Jeffrey Shainline: Neuromorphic Computing and Optoelectronic Intelligence

Lex Fridman (01:22.000)

is focused on superconducting electronics for computation.

Jeffrey Shainline (01:25.840)

I won't go into why that is,

Lex Fridman (01:27.760)

but that might make a little more sense in context

Jeffrey Shainline (01:31.440)

if we first describe what that is in contrast to,

Lex Fridman (01:35.440)

which is semiconducting electronics.

Lex Fridman (01:37.880)

So is it worth taking a couple minutes

Lex Fridman (01:39.720)

to describe semiconducting electronics?

Jeffrey Shainline (01:42.820)

It might even be worthwhile to step back

Lex Fridman (01:45.960)

and talk about electricity and circuits

Lex Fridman (01:49.960)

and how circuits work

Lex Fridman (01:52.240)

before we talk about superconductivity.

Jeffrey Shainline (01:54.760)

Right, okay.

Lex Fridman (01:56.560)

How does a computer work, Jeff?

Jeffrey Shainline (01:58.080)

Well, I won't go into everything

Lex Fridman (01:59.800)

that makes a computer work,

Lex Fridman (02:01.120)

but let's talk about the basic building blocks,

Lex Fridman (02:05.200)

a transistor, and even more basic than that,

Jeffrey Shainline (02:08.440)

a semiconductor material, silicon, say.

Lex Fridman (02:11.520)

So in silicon, silicon is a semiconductor,

Lex Fridman (02:15.440)

and what that means is at low temperature,

Lex Fridman (02:18.440)

there are no free charges,

Jeffrey Shainline (02:20.760)

no free electrons that can move around.

Lex Fridman (02:22.880)

So when you talk about electricity,

Jeffrey Shainline (02:24.920)

you're talking about predominantly electrons

Lex Fridman (02:28.040)

moving to establish electrical currents,

Lex Fridman (02:30.780)

and they move under the influence of voltages.

Lex Fridman (02:33.080)

So you apply voltages, electrons move around,

Jeffrey Shainline (02:36.640)

those can be measured as currents,

Lex Fridman (02:38.420)

and you can represent information in that way.

Lex Fridman (02:40.920)

So semiconductors are special

Lex Fridman (02:43.840)

in the sense that they are really malleable.

Lex Fridman (02:46.860)

So if you have a semiconductor material,

Lex Fridman (02:50.080)

you can change the number of free electrons

Jeffrey Shainline (02:52.600)

that can move around by putting different elements,

Lex Fridman (02:56.020)

different atoms in lattice sites.

Lex Fridman (02:58.280)

So what is a lattice site?

Lex Fridman (03:00.200)

Well, a semiconductor is a crystal,

Jeffrey Shainline (03:02.200)

which means all the atoms that comprise the material

Lex Fridman (03:06.480)

are at exact locations

Jeffrey Shainline (03:09.000)

that are perfectly periodic in space.

Lex Fridman (03:10.880)

So if you start at any one atom

Lex Fridman (03:12.220)

and you go along what are called the lattice vectors,

Lex Fridman (03:14.660)

you get to another atom and another atom and another atom,

Lex Fridman (03:17.320)

and for high quality devices,

Lex Fridman (03:19.440)

it's important that it's a perfect crystal

Jeffrey Shainline (03:21.920)

with very few defects,

Lex Fridman (03:23.820)

but you can intentionally replace a silicon atom

Jeffrey Shainline (03:27.440)

with say a phosphorus atom,

Lex Fridman (03:29.240)

and then you can change the number of free electrons

Jeffrey Shainline (03:32.160)

that are in a region of space

Lex Fridman (03:33.980)

that has that excess of what are called dopants.

Lex Fridman (03:37.140)

So picture a device that has a left terminal

Lex Fridman (03:40.260)

and a right terminal,

Lex Fridman (03:42.040)

and if you apply a voltage between those two,

Lex Fridman (03:44.160)

you can cause electrical current to flow between them.

Jeffrey Shainline (03:47.740)

Now we add a third terminal up on top there,

Lex Fridman (03:52.120)

and depending on the voltage

Jeffrey Shainline (03:53.800)

between the left and right terminal and that third voltage,

Lex Fridman (03:57.160)

you can change that current.

Lex Fridman (03:58.720)

So what's commonly done in digital electronic circuits

Lex Fridman (04:02.280)

is to leave a fixed voltage from left to right,

Lex Fridman (04:06.560)

and then change that voltage

Lex Fridman (04:08.600)

that's applied at what's called the gate,

Jeffrey Shainline (04:10.000)

the gate of the transistor.

Lex Fridman (04:11.220)

So what you do is you make it to where

Jeffrey Shainline (04:13.960)

there's an excess of electrons on the left,

Lex Fridman (04:15.680)

excess of electrons on the right,

Lex Fridman (04:17.360)

and very few electrons in the middle,

Lex Fridman (04:18.880)

and you do this by changing the concentration

Jeffrey Shainline (04:21.520)

of different dopants in the lattice spatially.

Lex Fridman (04:24.800)

And then when you apply a voltage to that gate,

Jeffrey Shainline (04:27.900)

you can either cause current to flow or turn it off,

Lex Fridman (04:30.680)

and so that's sort of your zero and one.

Jeffrey Shainline (04:33.440)

If you apply voltage, current can flow,

Lex Fridman (04:35.360)

that current is representing a digital one,

Lex Fridman (04:38.380)

and from that, from that basic element,

Lex Fridman (04:41.960)

you can build up all the complexity

Jeffrey Shainline (04:44.200)

of digital electronic circuits

Lex Fridman (04:45.520)

that have really had a profound influence on our society.

Jeffrey Shainline (04:50.080)

Now you're talking about electrons.

Lex Fridman (04:51.600)

Can you give a sense of what scale we're talking about

Jeffrey Shainline (04:54.680)

when we're talking about in silicon

Lex Fridman (04:57.600)

being able to mass manufacture these kinds of gates?

Jeffrey Shainline (05:01.560)

Yeah, so scale in a number of different senses.

Lex Fridman (05:04.480)

Well, at the scale of the silicon lattice,

Jeffrey Shainline (05:07.520)

the distance between two atoms there is half a nanometer.

Lex Fridman (05:10.880)

So people often like to compare these things

Jeffrey Shainline (05:14.480)

to the width of a human hair.

Lex Fridman (05:16.540)

I think it's some six orders of magnitude smaller

Jeffrey Shainline (05:20.040)

than the width of a human hair, something on that order.

Lex Fridman (05:24.440)

So remarkably small,

Jeffrey Shainline (05:25.840)

we're talking about individual atoms here,

Lex Fridman (05:27.280)

and electrons are of that length scale

Jeffrey Shainline (05:29.480)

when they're in that environment.

Lex Fridman (05:31.520)

But there's another sense

Jeffrey Shainline (05:32.440)

that scale matters in digital electronics.

Lex Fridman (05:34.380)

This is perhaps the more important sense,

Jeffrey Shainline (05:36.460)

although they're related.

Lex Fridman (05:37.880)

Scale refers to a number of things.

Jeffrey Shainline (05:41.340)

It refers to the size of that transistor.

Lex Fridman (05:44.560)

So for example, I said you have a left contact,

Jeffrey Shainline (05:47.280)

a right contact, and some space between them

Lex Fridman (05:49.800)

where the gate electrode sits.

Jeffrey Shainline (05:52.280)

That's called the channel width or the channel length.

Lex Fridman (05:56.000)

And what has enabled what we think of as Moore's law

Jeffrey Shainline (06:00.120)

or the continued increased performance

Lex Fridman (06:03.040)

in silicon microelectronic circuits

Jeffrey Shainline (06:05.560)

is the ability to make that size, that feature size,

Lex Fridman (06:08.960)

ever smaller, ever smaller at a really remarkable pace.

Jeffrey Shainline (06:14.800)

I mean, that feature size has decreased consistently

Lex Fridman (06:20.160)

every couple of years since the 1960s.

Lex Fridman (06:23.760)

And that was what Moore predicted in the 1960s.

Lex Fridman (06:27.160)

He thought it would continue for at least two more decades,

Lex Fridman (06:29.480)

and it's been much longer than that.

Lex Fridman (06:30.800)

And so that is why we've been able to fit ever more devices,

Jeffrey Shainline (06:35.800)

ever more transistors, ever more computational power

Lex Fridman (06:39.000)

on essentially the same size of chip.

Lex Fridman (06:41.440)

So a user sits back and does essentially nothing.

Lex Fridman (06:44.400)

You're running the same computer program,

Lex Fridman (06:45.960)

but those devices are getting smaller, so they get faster,

Lex Fridman (06:48.360)

they get more energy efficient,

Lex Fridman (06:50.080)

and all of our computing performance

Lex Fridman (06:51.520)

just continues to improve.

Lex Fridman (06:53.240)

And we don't have to think too hard

Lex Fridman (06:56.200)

about what we're doing as, say,

Jeffrey Shainline (06:59.200)

a software designer or something like that.

Lex Fridman (07:00.520)

I absolutely don't mean to say

Jeffrey Shainline (07:02.200)

that there's no innovation in software or the user side

Lex Fridman (07:05.960)

of things, of course there is.

Lex Fridman (07:07.040)

But from the hardware perspective,

Lex Fridman (07:09.400)

we just have been given this gift

Jeffrey Shainline (07:12.240)

of continued performance improvement

Lex Fridman (07:14.520)

through this scaling that is ever smaller feature sizes

Jeffrey Shainline (07:19.080)

with very similar, say, power consumption.

Lex Fridman (07:22.680)

That power consumption has not continued to scale

Jeffrey Shainline (07:25.880)

in the most recent decades, but nevertheless,

Lex Fridman (07:29.360)

we had a really good run there for a while.

Lex Fridman (07:31.200)

And now we're down to gates that are seven nanometers,

Lex Fridman (07:34.160)

which is state of the art right now.

Jeffrey Shainline (07:36.080)

Maybe GlobalFoundries is trying to push it

Lex Fridman (07:38.760)

even lower than that.

Jeffrey Shainline (07:39.600)

I can't keep up with where the predictions are

Lex Fridman (07:42.440)

that it's gonna end.

Lex Fridman (07:43.280)

But seven nanometer transistor has just a few tens of atoms

Lex Fridman (07:49.880)

along the length of the conduction pathway.

Lex Fridman (07:51.720)

So a naive semiconductor device physicist

Lex Fridman (07:56.640)

would think you can't go much further than that

Jeffrey Shainline (07:58.560)

without some kind of revolution in the way we think

Lex Fridman (08:02.240)

about the physics of our devices.

Jeffrey Shainline (08:03.800)

Is there something to be said

Lex Fridman (08:04.960)

about the mass manufacture of these devices?

Jeffrey Shainline (08:08.160)

Right, right, so that's another thing.

Lex Fridman (08:09.320)

So how have we been able

Lex Fridman (08:10.800)

to make those transistors smaller and smaller?

Lex Fridman (08:13.840)

Well, companies like Intel, GlobalFoundries,

Jeffrey Shainline (08:17.720)

they invest a lot of money in the lithography.

Lex Fridman (08:20.280)

So how are these chips actually made?

Jeffrey Shainline (08:22.800)

Well, one of the most important steps

Lex Fridman (08:24.320)

is this what's called ion implantation.

Lex Fridman (08:27.640)

So you start with sort of a pristine silicon crystal

Lex Fridman (08:31.920)

and then using photolithography,

Jeffrey Shainline (08:34.040)

which is a technique where you can pattern

Lex Fridman (08:36.440)

different shapes using light,

Jeffrey Shainline (08:38.760)

you can define which regions of space

Lex Fridman (08:41.360)

you're going to implant with different species of ions

Jeffrey Shainline (08:45.600)

that are going to change

Lex Fridman (08:46.560)

the local electrical properties right there.

Lex Fridman (08:49.520)

So by using ever shorter wavelengths of light

Lex Fridman (08:52.640)

and different kinds of optical techniques

Lex Fridman (08:54.360)

and different kinds of lithographic techniques,

Lex Fridman (08:56.560)

things that go far beyond my knowledge base,

Jeffrey Shainline (09:00.920)

you can just simply shrink that feature size down.

Lex Fridman (09:03.280)

And you say you're at seven nanometers.

Jeffrey Shainline (09:04.640)

Well, the wavelength of light that's being used

Lex Fridman (09:07.200)

is over a hundred nanometers.

Jeffrey Shainline (09:08.560)

That's already deep in the UV.

Lex Fridman (09:10.040)

So how are those minute features patterned?

Jeffrey Shainline (09:14.520)

Well, there's an extraordinary amount of innovation

Lex Fridman (09:16.720)

that has gone into that,

Lex Fridman (09:18.000)

but nevertheless, it stayed very consistent

Lex Fridman (09:20.080)

in this ever shrinking feature size.

Lex Fridman (09:21.720)

And now the question is, can you make it smaller?

Lex Fridman (09:24.560)

And even if you do, do you still continue

Lex Fridman (09:26.760)

to get performance improvements?

Lex Fridman (09:28.120)

But that's another kind of scaling

Jeffrey Shainline (09:30.040)

where these companies have been able to...

Lex Fridman (09:34.080)

So, okay, you picture a chip that has a processor on it.

Jeffrey Shainline (09:36.840)

Well, that chip is not made as a chip.

Lex Fridman (09:38.600)

It's made on a wafer.

Lex Fridman (09:40.680)

And using photolithography,

Lex Fridman (09:43.160)

you basically print the same pattern on different dyes

Jeffrey Shainline (09:47.000)

all across the wafer, multiple layers,

Lex Fridman (09:49.160)

tens, probably a hundred some layers

Jeffrey Shainline (09:53.120)

in a mature foundry process.

Lex Fridman (09:55.000)

And you do this on ever bigger wafers too.

Jeffrey Shainline (09:57.360)

That's another aspect of scaling

Lex Fridman (09:58.800)

that's occurred in the last several decades.

Lex Fridman (10:00.680)

So now you have this 300 millimeter wafer.

Lex Fridman (10:02.720)

It's like as big as a pizza

Lex Fridman (10:04.040)

and it has maybe a thousand processors on it.

Lex Fridman (10:06.360)

And then you dice that up using a saw.

Lex Fridman (10:08.640)

And now you can sell these things so cheap

Lex Fridman (10:11.520)

because the manufacturing process was so streamlined.

Jeffrey Shainline (10:14.920)

I think a technology as revolutionary

Lex Fridman (10:17.160)

as silicon microelectronics has to have

Jeffrey Shainline (10:19.880)

that kind of manufacturing scalability,

Lex Fridman (10:23.480)

which I will just emphasize,

Jeffrey Shainline (10:25.400)

I believe is enabled by physics.

Lex Fridman (10:28.920)

It's not, I mean, of course there's human ingenuity

Jeffrey Shainline (10:31.480)

that goes into it, but at least from my side where I sit,

Lex Fridman (10:35.840)

it sure looks like the physics of our universe

Jeffrey Shainline (10:38.760)

allows us to produce that.

Lex Fridman (10:40.880)

And we've discovered how more so than we've invented it,

Jeffrey Shainline (10:45.720)

although of course we have invented it,

Lex Fridman (10:47.160)

humans have invented it,

Lex Fridman (10:48.440)

but it's almost as if it was there

Lex Fridman (10:50.960)

waiting for us to discover it.

Jeffrey Shainline (10:52.960)

You mean the entirety of it

Lex Fridman (10:54.160)

or are you specifically talking about

Jeffrey Shainline (10:55.920)

the techniques of photolithography,

Lex Fridman (10:58.320)

like the optics involved?

Jeffrey Shainline (10:59.680)

I mean, the entirety of the scaling down

Lex Fridman (11:02.560)

to the seven nanometers,

Jeffrey Shainline (11:04.320)

you're able to have electrons not interfere with each other

Lex Fridman (11:08.200)

in such a way that you could still have gates.

Jeffrey Shainline (11:11.280)

Like that's enabled.

Lex Fridman (11:13.320)

To achieve that scale, spatial and temporal,

Jeffrey Shainline (11:16.560)

it seems to be very special

Lex Fridman (11:18.720)

and is enabled by the physics of our world.

Jeffrey Shainline (11:21.480)

All of the things you just said.

Lex Fridman (11:22.760)

So starting with the silicon material itself,

Jeffrey Shainline (11:25.960)

silicon is a unique semiconductor.

Lex Fridman (11:28.760)

It has essentially ideal properties

Jeffrey Shainline (11:31.560)

for making a specific kind of transistor

Lex Fridman (11:33.720)

that's extraordinarily useful.

Lex Fridman (11:35.200)

So I mentioned that silicon has,

Lex Fridman (11:39.560)

well, when you make a transistor,

Jeffrey Shainline (11:40.640)

you have this gate contact

Lex Fridman (11:42.000)

that sits on top of the conduction channel.

Lex Fridman (11:44.520)

And depending on the voltage you apply there,

Lex Fridman (11:47.240)

you pull more carriers into the conduction channel

Jeffrey Shainline (11:50.000)

or push them away so it becomes more or less conductive.

Lex Fridman (11:53.040)

In order to have that work

Jeffrey Shainline (11:54.680)

without just sucking those carriers right into that contact,

Lex Fridman (11:57.120)

you need a very thin insulator.

Lex Fridman (11:59.080)

And part of scaling has been to gradually decrease

Lex Fridman (12:03.200)

the thickness of that gate insulator

Lex Fridman (12:06.040)

so that you can use a roughly similar voltage

Lex Fridman (12:09.040)

and still have the same current voltage characteristics.

Lex Fridman (12:12.080)

So the material that's used to do that,

Lex Fridman (12:14.600)

or I should say was initially used to do that

Jeffrey Shainline (12:16.880)

was a silicon dioxide,

Lex Fridman (12:18.720)

which just naturally grows on the silicon surface.

Lex Fridman (12:21.800)

So you expose silicon to the atmosphere that we breathe

Lex Fridman (12:25.080)

and well, if you're manufacturing,

Jeffrey Shainline (12:27.560)

you're gonna purify these gases,

Lex Fridman (12:29.720)

but nevertheless,

Jeffrey Shainline (12:30.560)

that what's called a native oxide will grow there.

Lex Fridman (12:33.520)

There are essentially no other materials

Jeffrey Shainline (12:36.040)

on the entire periodic table

Lex Fridman (12:37.480)

that have as good of a gate insulator

Jeffrey Shainline (12:42.360)

as that silicon dioxide.

Lex Fridman (12:43.680)

And that has to do with nothing but the physics

Jeffrey Shainline (12:46.280)

of the interaction between silicon and oxygen.

Lex Fridman (12:49.240)

And if it wasn't that way,

Jeffrey Shainline (12:51.240)

transistors could not perform

Lex Fridman (12:54.720)

in nearly the degree of capability that they have.

Lex Fridman (12:58.800)

And that has to do with the way that the oxide grows,

Lex Fridman (13:02.880)

the reduced density of defects there,

Jeffrey Shainline (13:05.760)

it's insulation, meaning essentially it's energy gaps.

Lex Fridman (13:08.600)

You can apply a very large voltage there

Jeffrey Shainline (13:10.320)

without having current leak through it.

Lex Fridman (13:12.280)

So that's physics right there.

Jeffrey Shainline (13:15.760)

There are other things too.

Lex Fridman (13:16.880)

Silicon is a semiconductor in an elemental sense.

Jeffrey Shainline (13:19.840)

You only need silicon atoms.

Lex Fridman (13:21.560)

A lot of other semiconductors,

Jeffrey Shainline (13:22.800)

you need two different kinds of atoms,

Lex Fridman (13:24.880)

like a compound from group three

Lex Fridman (13:26.800)

and a compound from group five.

Lex Fridman (13:28.560)

That opens you up to lots of defects that can occur

Jeffrey Shainline (13:31.680)

where one atom's not sitting quite at the lattice site,

Lex Fridman (13:34.360)

it is and it's switched with another one

Jeffrey Shainline (13:35.960)

that degrades performance.

Lex Fridman (13:38.280)

But then also on the side that you mentioned

Jeffrey Shainline (13:40.280)

with the manufacturing,

Lex Fridman (13:43.480)

we have access to light sources

Jeffrey Shainline (13:45.840)

that can produce these very short wavelengths of light.

Lex Fridman (13:49.480)

How does photolithography occur?

Jeffrey Shainline (13:50.880)

Well, you actually put this polymer on top of your wafer

Lex Fridman (13:54.440)

and you expose it to light,

Lex Fridman (13:56.200)

and then you use a aqueous chemical processing

Lex Fridman (14:00.080)

to dissolve away the regions that were exposed to light

Lex Fridman (14:03.440)

and leave the regions that were not.

Lex Fridman (14:05.920)

And we are blessed with these polymers

Jeffrey Shainline (14:08.520)

that have the right property

Lex Fridman (14:09.640)

where they can cause scission events

Jeffrey Shainline (14:13.560)

where the polymer splits where a photon hits.

Lex Fridman (14:16.160)

I mean, maybe that's not too surprising,

Lex Fridman (14:19.720)

but I don't know, it all comes together

Lex Fridman (14:21.520)

to have this really complex,

Jeffrey Shainline (14:24.240)

manufacturable ecosystem

Lex Fridman (14:26.120)

where very sophisticated technologies can be devised

Lex Fridman (14:30.600)

and it works quite well.

Lex Fridman (14:33.000)

And amazingly, like you said,

Jeffrey Shainline (14:34.360)

with a wavelength at like 100 nanometers

Lex Fridman (14:36.240)

or something like that,

Jeffrey Shainline (14:37.080)

you're still able to achieve on this polymer

Lex Fridman (14:39.720)

precision of whatever we said, seven nanometers.

Jeffrey Shainline (14:43.440)

I think I've heard like four nanometers

Lex Fridman (14:45.720)

being talked about, something like that.

Jeffrey Shainline (14:48.400)

If we could just pause on this

Lex Fridman (14:49.800)

and we'll return to superconductivity,

Lex Fridman (14:52.080)

but in this whole journey from a history perspective,

Lex Fridman (14:56.360)

what do you think is the most beautiful

Jeffrey Shainline (14:59.440)

at the intersection of engineering and physics

Lex Fridman (15:01.680)

to you in this whole process

Jeffrey Shainline (15:03.240)

that we talked about with silicon and photolithography,

Lex Fridman (15:06.520)

things that people were able to achieve

Lex Fridman (15:08.320)

in order to push Moore's law forward?

Lex Fridman (15:12.280)

Is it the early days,

Lex Fridman (15:13.640)

the invention of the transistor itself?

Lex Fridman (15:16.240)

Is it some particular cool little thing

Lex Fridman (15:19.280)

that maybe not many people know about?

Lex Fridman (15:21.960)

Like, what do you think is most beautiful

Lex Fridman (15:24.480)

in this whole process, journey?

Lex Fridman (15:26.760)

The most beautiful is a little difficult to answer.

Jeffrey Shainline (15:29.560)

Let me try and sidestep it a little bit

Lex Fridman (15:32.040)

and just say what strikes me about looking

Jeffrey Shainline (15:35.840)

at the history of silicon microelectronics is that,

Lex Fridman (15:42.000)

so when quantum mechanics was developed,

Jeffrey Shainline (15:44.600)

people quickly began applying it to semiconductors

Lex Fridman (15:47.440)

and it was broadly understood

Jeffrey Shainline (15:49.360)

that these are fascinating systems

Lex Fridman (15:50.760)

and people cared about them for their basic physics,

Lex Fridman (15:52.720)

but also their utility as devices.

Lex Fridman (15:55.040)

And then the transistor was invented in the late forties

Jeffrey Shainline (15:59.280)

in a relatively crude experimental setup

Lex Fridman (16:02.080)

where you just crammed a metal electrode

Jeffrey Shainline (16:04.280)

into the semiconductor and that was ingenious.

Lex Fridman (16:08.040)

These people were able to make it work.

Lex Fridman (16:13.000)

But so what I wanna get to that really strikes me

Lex Fridman (16:16.840)

is that in those early days,

Jeffrey Shainline (16:19.320)

there were a number of different semiconductors

Lex Fridman (16:21.200)

that were being considered.

Jeffrey Shainline (16:22.120)

They had different properties, different strengths,

Lex Fridman (16:23.840)

different weaknesses.

Jeffrey Shainline (16:24.920)

Most people thought germanium was the way to go.

Lex Fridman (16:28.480)

It had some nice properties related to things

Jeffrey Shainline (16:33.680)

about how the electrons move inside the lattice.

Lex Fridman (16:37.320)

But other people thought that compound semiconductors

Jeffrey Shainline (16:39.800)

with group three and group five also had

Lex Fridman (16:42.000)

really, really extraordinary properties

Jeffrey Shainline (16:46.240)

that might be conducive to making the best devices.

Lex Fridman (16:50.080)

So there were different groups exploring each of these

Lex Fridman (16:52.560)

and that's great, that's how science works.

Lex Fridman (16:54.240)

You have to cast a broad net.

Lex Fridman (16:56.160)

But then what I find striking is why is it that silicon won?

Lex Fridman (17:02.120)

Because it's not that germanium is a useless material

Lex Fridman (17:05.280)

and it's not present in technology

Lex Fridman (17:06.760)

or compound semiconductors.

Jeffrey Shainline (17:08.080)

They're both doing exciting and important things,

Lex Fridman (17:12.640)

slightly more niche applications

Jeffrey Shainline (17:14.400)

whereas silicon is the semiconductor material

Lex Fridman (17:18.080)

for microelectronics which is the platform

Jeffrey Shainline (17:20.120)

for digital computing which has transformed our world.

Lex Fridman (17:22.760)

Why did silicon win?

Jeffrey Shainline (17:24.200)

It's because of a remarkable assemblage of qualities

Lex Fridman (17:28.720)

that no one of them was the clear winner

Lex Fridman (17:32.120)

but it made these sort of compromises

Lex Fridman (17:34.560)

between a number of different influences.

Jeffrey Shainline (17:36.680)

It had that really excellent gate oxide

Lex Fridman (17:40.520)

that allowed us to make MOSFETs,

Jeffrey Shainline (17:43.240)

these high performance transistors,

Lex Fridman (17:45.400)

so quickly and cheaply and easily

Jeffrey Shainline (17:47.200)

without having to do a lot of materials development.

Lex Fridman (17:49.360)

The band gap of silicon is actually,

Lex Fridman (17:53.400)

so in a semiconductor there's an important parameter

Lex Fridman (17:56.280)

which is called the band gap

Jeffrey Shainline (17:57.480)

which tells you there are sort of electrons

Lex Fridman (18:00.600)

that fill up to one level in the energy diagram

Lex Fridman (18:04.600)

and then there's a gap where electrons aren't allowed

Lex Fridman (18:06.800)

to have an energy in a certain range

Lex Fridman (18:08.280)

and then there's another energy level above that.

Lex Fridman (18:11.320)

And that difference between the lower sort of filled level

Lex Fridman (18:14.960)

and the unoccupied level,

Lex Fridman (18:16.880)

that tells you how much voltage you have to apply

Jeffrey Shainline (18:19.640)

in order to induce a current to flow.

Lex Fridman (18:22.160)

So with germanium, that's about 0.75 electron volts.

Jeffrey Shainline (18:27.320)

That means you have to apply 0.75 volts

Lex Fridman (18:29.640)

to get a current moving.

Lex Fridman (18:32.000)

And it turns out that if you compare that

Lex Fridman (18:34.280)

to the thermal excitations that are induced

Jeffrey Shainline (18:38.520)

just by the temperature of our environment,

Lex Fridman (18:40.680)

that gap's not quite big enough.

Jeffrey Shainline (18:42.120)

You start to use it to perform computations,

Lex Fridman (18:45.120)

it gets a little hot and you get all these accidental

Jeffrey Shainline (18:47.640)

carriers that are excited into the conduction band

Lex Fridman (18:50.720)

and it causes errors in your computation.

Jeffrey Shainline (18:53.360)

Silicon's band gap is just a little higher,

Lex Fridman (18:56.200)

1.1 electron volts,

Lex Fridman (18:58.960)

but you have an exponential dependence

Lex Fridman (19:01.280)

on the number of carriers that are present

Jeffrey Shainline (19:04.200)

that can induce those errors.

Lex Fridman (19:06.600)

It decays exponentially with that voltage.

Lex Fridman (19:08.480)

So just that slight extra energy in that band gap

Lex Fridman (19:12.760)

really puts it in an ideal position to be operated

Jeffrey Shainline (19:17.040)

in the conditions of our ambient environment.

Lex Fridman (19:20.200)

It's kind of fascinating that, like you mentioned,

Jeffrey Shainline (19:22.440)

errors decrease exponentially with the voltage.

Lex Fridman (19:27.500)

So it's funny because this error thing comes up

Jeffrey Shainline (19:32.040)

when you start talking about quantum computing.

Lex Fridman (19:34.600)

And it's kind of amazing that everything

Jeffrey Shainline (19:36.020)

we've been talking about, the errors,

Lex Fridman (19:37.920)

as we scale down, seems to be extremely low.

Jeffrey Shainline (19:41.480)

Yes.

Lex Fridman (19:42.320)

And like all of our computation is based

Jeffrey Shainline (19:45.960)

on the assumption that it's extremely low.

Lex Fridman (19:47.760)

Yes, well it's digital computation.

Jeffrey Shainline (19:49.560)

Digital, sorry, digital computation.

Lex Fridman (19:51.480)

So as opposed to our biological computation in our brain,

Jeffrey Shainline (19:55.000)

is like the assumption is stuff is gonna fail

Lex Fridman (19:58.240)

all over the place and we somehow

Jeffrey Shainline (19:59.800)

have to still be robust to that.

Lex Fridman (1:00:00.700)

at the kind of information processing it does.

Lex Fridman (1:00:03.180)

And I wanna think from first principles

Lex Fridman (1:00:05.540)

about what hardware is best going to enable us

Jeffrey Shainline (1:00:10.580)

to capture those information processing principles

Lex Fridman (1:00:14.020)

in an artificial system.

Lex Fridman (1:00:16.020)

And that's where I live.

Lex Fridman (1:00:17.420)

That's where I'm doing my exploration these days.

Lex Fridman (1:00:21.740)

So what are the first principles

Lex Fridman (1:00:25.800)

of brain like computation communication?

Jeffrey Shainline (1:00:29.960)

Right, yeah, this is so important

Lex Fridman (1:00:32.580)

and I'm glad we booked 14 hours for this because.

Jeffrey Shainline (1:00:35.540)

I only have 13, I'm sorry.

Lex Fridman (1:00:38.380)

Okay, so the brain is notoriously complicated.

Lex Fridman (1:00:41.500)

And I think that's an important part

Lex Fridman (1:00:44.060)

of why it can do what it does.

Lex Fridman (1:00:46.300)

But okay, let me try to break it down.

Lex Fridman (1:00:49.620)

Starting with the devices, neurons, as I said before,

Jeffrey Shainline (1:00:54.580)

they're sophisticated devices in and of themselves

Lex Fridman (1:00:57.100)

and synapses are too.

Jeffrey Shainline (1:00:58.220)

They can change their state based on the activity.

Lex Fridman (1:01:03.060)

So they adapt over time.

Jeffrey Shainline (1:01:04.900)

That's crucial to the way the brain works.

Lex Fridman (1:01:06.980)

They don't just adapt on one timescale,

Jeffrey Shainline (1:01:09.380)

they can adapt on myriad timescales

Lex Fridman (1:01:12.460)

from the spacing between pulses,

Jeffrey Shainline (1:01:16.060)

the spacing between spikes that come from neurons

Lex Fridman (1:01:18.700)

all the way to the age of the organism.

Jeffrey Shainline (1:01:23.100)

Also relevant, perhaps I think the most important thing

Lex Fridman (1:01:28.100)

that's guided my thinking is the network structure

Jeffrey Shainline (1:01:32.320)

of the brain, so.

Lex Fridman (1:01:33.880)

Which can also be adjusted on different scales.

Jeffrey Shainline (1:01:36.600)

Absolutely, yes, so you're making new,

Lex Fridman (1:01:39.440)

you're changing the strength of contacts,

Jeffrey Shainline (1:01:41.360)

you're changing the spatial distribution of them,

Lex Fridman (1:01:44.120)

although spatial distribution doesn't change that much

Jeffrey Shainline (1:01:46.900)

once you're a mature organism.

Lex Fridman (1:01:49.400)

But that network structure is really crucial.

Lex Fridman (1:01:52.880)

So let me dwell on that for a second.

Lex Fridman (1:01:55.400)

You can't talk about the brain without emphasizing

Jeffrey Shainline (1:01:58.960)

that most of the neurons in the neocortex

Lex Fridman (1:02:02.880)

or the prefrontal cortex, the part of the brain

Jeffrey Shainline (1:02:04.840)

that we think is most responsible for high level reasoning

Lex Fridman (1:02:08.080)

and things like that,

Jeffrey Shainline (1:02:09.080)

those neurons make thousands of connections.

Lex Fridman (1:02:11.400)

So you have this network that is highly interconnected.

Lex Fridman (1:02:15.560)

And I think it's safe to say that one of the primary reasons

Lex Fridman (1:02:19.880)

that they make so many different connections

Jeffrey Shainline (1:02:23.180)

is that allows information to be communicated very rapidly

Lex Fridman (1:02:26.880)

from any spot in the network

Jeffrey Shainline (1:02:28.420)

to any other spot in the network.

Lex Fridman (1:02:30.320)

So that's a sort of spatial aspect of it.

Jeffrey Shainline (1:02:33.920)

You can quantify this in terms of concepts

Lex Fridman (1:02:38.480)

that are related to fractals and scale invariants,

Jeffrey Shainline (1:02:41.020)

which I think is a very beautiful concept.

Lex Fridman (1:02:43.240)

So what I mean by that is kind of,

Jeffrey Shainline (1:02:46.200)

no matter what spatial scale you're looking at in the brain

Lex Fridman (1:02:50.720)

within certain bounds, you see the same

Jeffrey Shainline (1:02:53.520)

general statistical pattern.

Lex Fridman (1:02:54.960)

So if I draw a box around some region of my cortex,

Jeffrey Shainline (1:02:59.280)

most of the connections that those neurons

Lex Fridman (1:03:02.280)

within that box make are gonna be within the box

Jeffrey Shainline (1:03:04.480)

to each other in their local neighborhood.

Lex Fridman (1:03:06.200)

And that's sort of called clustering, loosely speaking.

Lex Fridman (1:03:09.280)

But a non negligible fraction

Lex Fridman (1:03:10.920)

is gonna go outside of that box.

Lex Fridman (1:03:12.600)

And then if I draw a bigger box,

Lex Fridman (1:03:14.080)

the pattern is gonna be exactly the same.

Lex Fridman (1:03:16.400)

So you have this scale invariants,

Lex Fridman (1:03:18.400)

and you also have a non vanishing probability

Jeffrey Shainline (1:03:22.720)

of a neuron making connection very far away.

Lex Fridman (1:03:25.400)

So suppose you wanna plot the probability

Jeffrey Shainline (1:03:28.720)

of a neuron making a connection as a function of distance.

Lex Fridman (1:03:32.420)

If that were an exponential function,

Jeffrey Shainline (1:03:34.240)

it would go e to the minus radius

Lex Fridman (1:03:36.720)

over some characteristic radius,

Lex Fridman (1:03:38.700)

and it would drop off up to some certain radius,

Lex Fridman (1:03:41.840)

the probability would be reasonably close to one,

Lex Fridman (1:03:44.920)

and then beyond that characteristic length R zero,

Lex Fridman (1:03:49.120)

it would drop off sharply.

Lex Fridman (1:03:51.280)

And so that would mean that the neurons in your brain

Lex Fridman (1:03:53.560)

are really localized, and that's not what we observe.

Jeffrey Shainline (1:03:58.640)

Instead, what you see is that the probability

Lex Fridman (1:04:00.680)

of making a longer distance connection, it does drop off,

Lex Fridman (1:04:03.760)

but it drops off as a power law.

Lex Fridman (1:04:05.760)

So the probability that you're gonna have a connection

Jeffrey Shainline (1:04:08.420)

at some radius R goes as R to the minus some power.

Lex Fridman (1:04:13.180)

And that's more, that's what we see with forces in nature,

Jeffrey Shainline (1:04:16.800)

like the electromagnetic force

Lex Fridman (1:04:18.440)

between two particles or gravity

Jeffrey Shainline (1:04:20.440)

goes as one over the radius squared.

Lex Fridman (1:04:23.000)

So you can see this in fractals.

Jeffrey Shainline (1:04:24.420)

I love that there's like a fractal dynamics of the brain

Lex Fridman (1:04:28.600)

that if you zoom out, you draw the box

Lex Fridman (1:04:31.460)

and you increase that box by certain step sizes,

Lex Fridman (1:04:35.040)

you're gonna see the same statistics.

Jeffrey Shainline (1:04:36.720)

I think that's probably very important

Lex Fridman (1:04:40.000)

to the way the brain processes information.

Jeffrey Shainline (1:04:41.880)

It's not just in the spatial domain,

Lex Fridman (1:04:43.600)

it's also in the temporal domain.

Lex Fridman (1:04:45.640)

And what I mean by that is...

Lex Fridman (1:04:48.640)

That's incredible that this emerged

Jeffrey Shainline (1:04:50.640)

through the evolutionary process

Lex Fridman (1:04:52.320)

that potentially somehow connected

Jeffrey Shainline (1:04:54.880)

to the way the physics of the universe works.

Lex Fridman (1:04:57.800)

Yeah, I couldn't agree more that it's a deep

Lex Fridman (1:05:00.360)

and fascinating subject that I hope to be able

Lex Fridman (1:05:02.720)

to spend the rest of my life studying.

Jeffrey Shainline (1:05:04.160)

You think you need to solve, understand this,

Lex Fridman (1:05:07.280)

this fractal nature in order to understand intelligence

Lex Fridman (1:05:10.120)

and communication. I do think so.

Lex Fridman (1:05:11.520)

I think they're deeply intertwined.

Jeffrey Shainline (1:05:13.320)

Yes, I think power laws are right at the heart of it.

Lex Fridman (1:05:16.880)

So just to push that one through,

Jeffrey Shainline (1:05:19.400)

the same thing happens in the temporal domain.

Lex Fridman (1:05:21.480)

So suppose your neurons in your brain

Jeffrey Shainline (1:05:26.000)

were always oscillating at the same frequency,

Lex Fridman (1:05:28.160)

then the probability of finding a neuron oscillating

Jeffrey Shainline (1:05:31.320)

as a function of frequency

Lex Fridman (1:05:32.520)

would be this narrowly peaked function

Jeffrey Shainline (1:05:34.520)

around that certain characteristic frequency.

Lex Fridman (1:05:36.520)

That's not at all what we see.

Jeffrey Shainline (1:05:37.880)

The probability of finding neurons oscillating

Lex Fridman (1:05:40.240)

or producing spikes at a certain frequency

Jeffrey Shainline (1:05:43.880)

is again a power law,

Lex Fridman (1:05:45.200)

which means there's no defined scale

Jeffrey Shainline (1:05:49.640)

of the temporal activity in the brain.

Lex Fridman (1:05:53.560)

At what speed do your thoughts occur?

Jeffrey Shainline (1:05:56.040)

Well, there's a fastest speed they can occur

Lex Fridman (1:05:58.280)

and that is limited by communication and other things,

Lex Fridman (1:06:01.520)

but there's not a characteristic scale.

Lex Fridman (1:06:03.960)

We have thoughts on all temporal scales

Jeffrey Shainline (1:06:06.880)

from a few tens of milliseconds,

Lex Fridman (1:06:10.800)

which is physiologically limited by our devices,

Jeffrey Shainline (1:06:13.360)

compare that to tens of picoseconds

Lex Fridman (1:06:15.720)

that I talked about in superconductors,

Jeffrey Shainline (1:06:17.120)

all the way up to the lifetime of the organism.

Lex Fridman (1:06:19.240)

You can still think about things

Jeffrey Shainline (1:06:20.720)

that happened to you when you were a kid.

Lex Fridman (1:06:22.560)

Or if you wanna be really trippy

Jeffrey Shainline (1:06:24.040)

then across multiple organisms

Lex Fridman (1:06:25.840)

in the entirety of human civilization,

Lex Fridman (1:06:27.440)

you have thoughts that span organisms, right?

Lex Fridman (1:06:29.400)

Yes, taking it to that level, yes.

Jeffrey Shainline (1:06:31.200)

If you're willing to see the entirety of the human species

Lex Fridman (1:06:34.600)

as a single organism with a collective intelligence

Lex Fridman (1:06:37.160)

and that too on a spatial and temporal scale,

Lex Fridman (1:06:39.880)

there's thoughts occurring.

Lex Fridman (1:06:41.080)

And then if you look at not just the human species,

Lex Fridman (1:06:44.000)

but the entirety of life on earth

Jeffrey Shainline (1:06:46.440)

as an organism with thoughts that are occurring,

Lex Fridman (1:06:49.600)

that are greater and greater sophisticated thoughts,

Jeffrey Shainline (1:06:51.600)

there's a different spatial and temporal scale there.

Lex Fridman (1:06:54.640)

This is getting very suspicious.

Jeffrey Shainline (1:06:57.200)

Well, hold on though, before we're done,

Lex Fridman (1:06:58.640)

I just wanna just tie the bow

Lex Fridman (1:07:00.960)

and say that the spatial and temporal aspects

Lex Fridman (1:07:04.280)

are intimately interrelated with each other.

Lex Fridman (1:07:06.440)

So activity between neurons that are very close to each other

Lex Fridman (1:07:10.320)

is more likely to happen on this faster timescale

Lex Fridman (1:07:13.560)

and information is gonna propagate

Lex Fridman (1:07:15.280)

and encompass more of the brain,

Jeffrey Shainline (1:07:17.200)

more of your cortices, different modules in the brain

Lex Fridman (1:07:20.280)

are gonna be engaged in information processing

Jeffrey Shainline (1:07:23.720)

on longer timescales.

Lex Fridman (1:07:25.280)

So there's this concept of information integration

Jeffrey Shainline (1:07:27.960)

where neurons are specialized.

Lex Fridman (1:07:31.960)

Any given neuron or any cluster of neuron

Jeffrey Shainline (1:07:33.960)

has its specific purpose,

Lex Fridman (1:07:35.720)

but they're also very much integrated.

Lex Fridman (1:07:39.880)

So you have neurons that specialize,

Lex Fridman (1:07:41.880)

but share their information.

Lex Fridman (1:07:43.640)

And so that happens through these fractal nested oscillations

Lex Fridman (1:07:47.560)

that occur across spatial and temporal scales.

Jeffrey Shainline (1:07:49.400)

I think capturing those dynamics in hardware,

Lex Fridman (1:07:53.640)

to me, that's the goal of neuromorphic computing.

Lex Fridman (1:07:57.040)

So does it need to look,

Lex Fridman (1:07:58.680)

so first of all, that's fascinating.

Jeffrey Shainline (1:08:00.800)

We stated some clear principles here.

Lex Fridman (1:08:03.960)

Now, does it have to look like the brain

Lex Fridman (1:08:08.120)

outside of those principles as well?

Lex Fridman (1:08:09.800)

Like what other characteristics

Lex Fridman (1:08:11.320)

have to look like the human brain?

Lex Fridman (1:08:13.080)

Or can it be something very different?

Jeffrey Shainline (1:08:15.840)

Well, it depends on what you're trying to use it for.

Lex Fridman (1:08:18.000)

And so I think a lot of the community

Jeffrey Shainline (1:08:21.720)

asks that question a lot.

Lex Fridman (1:08:23.080)

What are you gonna do with it?

Lex Fridman (1:08:24.360)

And I completely get it.

Lex Fridman (1:08:26.600)

I think that's a very important question.

Lex Fridman (1:08:28.040)

And it's also sometimes not the most helpful question.

Lex Fridman (1:08:31.840)

What if what you wanna do with it is study it?

Lex Fridman (1:08:33.800)

What if you just wanna see,

Lex Fridman (1:08:37.400)

what do you have to build into your hardware

Lex Fridman (1:08:39.280)

in order to observe these dynamical principles?

Lex Fridman (1:08:43.200)

And also, I ask myself that question every day

Lex Fridman (1:08:47.520)

and I'm not sure I'm able to answer that.

Lex Fridman (1:08:49.880)

So like, what are you gonna do

Lex Fridman (1:08:51.200)

with this particular neuromorphic machine?

Lex Fridman (1:08:53.480)

So suppose what we're trying to do with it

Jeffrey Shainline (1:08:55.320)

is build something that thinks.

Lex Fridman (1:08:56.960)

We're not trying to get it to make us any money

Jeffrey Shainline (1:08:59.160)

or drive a car.

Lex Fridman (1:09:00.240)

Maybe we'll be able to do that, but that's not our goal.

Jeffrey Shainline (1:09:02.640)

Our goal is to see if we can get the same types of behaviors

Lex Fridman (1:09:07.600)

that we observe in our own brain.

Lex Fridman (1:09:08.920)

And by behaviors in this sense,

Lex Fridman (1:09:10.480)

what I mean the behaviors of the components,

Jeffrey Shainline (1:09:14.320)

the neurons, the network, that kind of stuff.

Lex Fridman (1:09:16.000)

I think there's another element that I didn't really hit on

Jeffrey Shainline (1:09:19.120)

that you also have to build into this.

Lex Fridman (1:09:21.200)

And those are architectural principles.

Jeffrey Shainline (1:09:22.920)

They have to do with the hierarchical modular construction

Lex Fridman (1:09:26.680)

of the network.

Lex Fridman (1:09:27.520)

And without getting too lost in jargon,

Lex Fridman (1:09:30.200)

the main point that I think is relevant there,

Jeffrey Shainline (1:09:33.680)

let me try and illustrate it with a cartoon picture

Lex Fridman (1:09:35.720)

of the architecture of the brain.

Lex Fridman (1:09:38.200)

So in the brain, you have the cortex,

Lex Fridman (1:09:41.120)

which is sort of this outer sheet.

Jeffrey Shainline (1:09:44.440)

It's actually, it's a layered structure.

Lex Fridman (1:09:46.720)

You can, if you could take it out of your brain,

Jeffrey Shainline (1:09:48.480)

you could unroll it on the table

Lex Fridman (1:09:50.680)

and it would be about the size of a pizza sitting there.

Lex Fridman (1:09:53.560)

And that's a module.

Lex Fridman (1:09:56.400)

It does certain things.

Jeffrey Shainline (1:09:57.800)

It processes as Yogi Buzaki would say,

Lex Fridman (1:10:00.680)

it processes the what of what's going on around you.

Lex Fridman (1:10:03.560)

But you have another really crucial module

Lex Fridman (1:10:06.200)

that's called the hippocampus.

Lex Fridman (1:10:08.040)

And that network is structured entirely differently.

Lex Fridman (1:10:10.520)

First of all, this cortex that had described

Jeffrey Shainline (1:10:12.800)

10 billion neurons in there.

Lex Fridman (1:10:14.640)

So numbers matter here.

Lex Fridman (1:10:16.840)

And they're organized in that sort of power law distribution

Lex Fridman (1:10:20.360)

where the probability of making a connection drops off

Jeffrey Shainline (1:10:22.880)

as a power law in space.

Lex Fridman (1:10:24.520)

The hippocampus is another module that's important

Jeffrey Shainline (1:10:26.840)

for understanding how, where you are,

Lex Fridman (1:10:30.880)

when you are keeping track of your position

Jeffrey Shainline (1:10:36.240)

in space and time.

Lex Fridman (1:10:37.280)

And that network is very much random.

Lex Fridman (1:10:39.280)

So the probability of making a connection,

Lex Fridman (1:10:41.960)

it almost doesn't even drop off as a function of distance.

Jeffrey Shainline (1:10:44.760)

It's the same probability that you'll make it here

Lex Fridman (1:10:46.720)

to over there, but there are only about 100 million neurons

Jeffrey Shainline (1:10:50.520)

there, so you can have that huge densely connected module

Lex Fridman (1:10:54.680)

because it's not so big.

Lex Fridman (1:10:57.280)

And the neocortex or the cortex and the hippocampus,

Lex Fridman (1:11:02.040)

they talk to each other constantly.

Lex Fridman (1:11:04.920)

And that communication is largely facilitated

Lex Fridman (1:11:07.920)

by what's called the thalamus.

Jeffrey Shainline (1:11:09.720)

I'm not a neuroscientist here.

Lex Fridman (1:11:10.880)

I'm trying to do my best to recite things.

Jeffrey Shainline (1:11:12.960)

Cartoon picture of the brain, I gotcha.

Lex Fridman (1:11:14.680)

Yeah, something like that.

Lex Fridman (1:11:15.560)

So this thalamus is coordinating the activity

Lex Fridman (1:11:18.640)

between the neocortex and the hippocampus

Lex Fridman (1:11:20.760)

and making sure that they talk to each other

Lex Fridman (1:11:23.560)

at the right time and send messages

Jeffrey Shainline (1:11:25.280)

that will be useful to one another.

Lex Fridman (1:11:26.840)

So this all taken together is called

Jeffrey Shainline (1:11:29.120)

the thalamocortical complex.

Lex Fridman (1:11:31.600)

And it seems like building something like that

Jeffrey Shainline (1:11:34.880)

is going to be crucial to capturing the types of activity

Lex Fridman (1:11:39.280)

we're looking for because those responsibilities,

Jeffrey Shainline (1:11:43.400)

those separate modules, they do different things,

Lex Fridman (1:11:45.720)

that's gotta be central to achieving these states

Jeffrey Shainline (1:11:51.760)

of efficient information integration across space and time.

Lex Fridman (1:11:55.720)

By the way, I am able to achieve this state

Jeffrey Shainline (1:11:58.960)

by watching simulations, visualizations

Lex Fridman (1:12:01.800)

of the thalamocortical complex.

Jeffrey Shainline (1:12:03.800)

There's a few people I forget from where.

Lex Fridman (1:12:06.440)

They've created these incredible visual illustrations

Jeffrey Shainline (1:12:09.880)

of visual stimulation from the eye or something like that.

Lex Fridman (1:12:14.880)

And this image flowing through the brain.

Jeffrey Shainline (1:12:18.520)

Wow, I haven't seen that, I gotta check that out.

Lex Fridman (1:12:20.880)

So it's one of those things,

Jeffrey Shainline (1:12:22.120)

you find this stuff in the world,

Lex Fridman (1:12:24.280)

and you see on YouTube, it has 1,000 views,

Jeffrey Shainline (1:12:26.960)

these visualizations of the human brain

Lex Fridman (1:12:30.800)

processing information.

Lex Fridman (1:12:32.120)

And because there's chemistry there,

Lex Fridman (1:12:36.440)

because this is from actual human brains,

Jeffrey Shainline (1:12:38.880)

I don't know how they're doing the coloring,

Lex Fridman (1:12:40.720)

but they're able to actually trace

Jeffrey Shainline (1:12:42.840)

the different, the chemical and the electrical signals

Lex Fridman (1:12:46.680)

throughout the brain, and the visual thing,

Jeffrey Shainline (1:12:48.880)

it's like, whoa, because it looks kinda like the universe,

Lex Fridman (1:12:51.800)

I mean, the whole thing is just incredible.

Jeffrey Shainline (1:12:53.800)

I recommend it highly, I'll probably post a link to it.

Lex Fridman (1:12:56.640)

But you can just look for, one of the things they simulate

Jeffrey Shainline (1:13:00.960)

is the thalamocortical complex and just visualization.

Lex Fridman (1:13:05.840)

You can find that yourself on YouTube, but it's beautiful.

Jeffrey Shainline (1:13:09.520)

The other question I have for you is,

Lex Fridman (1:13:11.320)

how does memory play into all of this?

Jeffrey Shainline (1:13:14.440)

Because all the signals sending back and forth,

Lex Fridman (1:13:17.120)

that's computation and communication,

Lex Fridman (1:13:20.880)

but that's kinda like processing of inputs and outputs,

Lex Fridman (1:13:26.240)

to produce outputs in the system,

Jeffrey Shainline (1:13:27.560)

that's kinda like maybe reasoning,

Lex Fridman (1:13:29.000)

maybe there's some kind of recurrence.

Lex Fridman (1:13:30.920)

But is there a storage mechanism that you think about

Lex Fridman (1:13:33.920)

in the context of neuromorphic computing?

Jeffrey Shainline (1:13:35.840)

Yeah, absolutely, so that's gotta be central.

Lex Fridman (1:13:37.760)

You have to have a way that you can store memories.

Lex Fridman (1:13:41.520)

And there are a lot of different kinds

Lex Fridman (1:13:43.600)

of memory in the brain.

Jeffrey Shainline (1:13:45.480)

That's yet another example of how it's not a simple system.

Lex Fridman (1:13:49.160)

So there's one kind of memory,

Jeffrey Shainline (1:13:53.000)

one way of talking about memory,

Lex Fridman (1:13:56.040)

usually starts in the context of Hopfield networks.

Jeffrey Shainline (1:13:59.040)

You were lucky to talk to John Hopfield on this program.

Lex Fridman (1:14:02.440)

But the basic idea there is working memory

Jeffrey Shainline (1:14:05.840)

is stored in the dynamical patterns

Lex Fridman (1:14:07.840)

of activity between neurons.

Lex Fridman (1:14:10.400)

And you can think of a certain pattern of activity

Lex Fridman (1:14:14.760)

as an attractor, meaning if you put in some signal

Jeffrey Shainline (1:14:19.680)

that's similar enough to other

Lex Fridman (1:14:22.400)

previously experienced signals like that,

Jeffrey Shainline (1:14:26.480)

then you're going to converge to the same network dynamics

Lex Fridman (1:14:29.600)

and you will see these neurons

Jeffrey Shainline (1:14:31.760)

participate in the same network patterns of activity

Lex Fridman (1:14:36.200)

that they have in the past.

Lex Fridman (1:14:37.600)

So you can talk about the probability

Lex Fridman (1:14:39.720)

that different inputs will allow you to converge

Jeffrey Shainline (1:14:42.520)

to different basins of attraction

Lex Fridman (1:14:44.240)

and you might think of that as,

Jeffrey Shainline (1:14:46.600)

oh, I saw this face and then I excited

Lex Fridman (1:14:49.040)

this network pattern of activity

Jeffrey Shainline (1:14:50.920)

because last time I saw that face,

Lex Fridman (1:14:53.080)

I was at some movie and that's a famous person

Jeffrey Shainline (1:14:56.960)

that's on the screen or something like that.

Lex Fridman (1:14:58.120)

So that's one memory storage mechanism.

Lex Fridman (1:15:00.560)

But crucial to the ability to imprint those memories

Lex Fridman (1:15:04.400)

in your brain is the ability to change

Jeffrey Shainline (1:15:07.040)

the strength of connection between one neuron and another,

Lex Fridman (1:15:11.360)

that synaptic connection between them.

Lex Fridman (1:15:13.280)

So synaptic weight update is a massive field of neuroscience

Lex Fridman (1:15:18.000)

and neuromorphic computing as well.

Lex Fridman (1:15:19.560)

So there are two poles on that spectrum.

Lex Fridman (1:15:26.720)

Okay, so more in the language of machine learning,

Jeffrey Shainline (1:15:28.880)

we would talk about supervised and unsupervised learning.

Lex Fridman (1:15:32.000)

And when I'm trying to tie that down

Jeffrey Shainline (1:15:33.960)

to neuromorphic computing,

Lex Fridman (1:15:35.520)

I will use a definition of supervised learning,

Jeffrey Shainline (1:15:38.440)

which basically means the external user,

Lex Fridman (1:15:42.960)

the person who's controlling this hardware

Jeffrey Shainline (1:15:45.520)

has some knob that they can tune

Lex Fridman (1:15:48.360)

to change each of the synaptic weights,

Jeffrey Shainline (1:15:50.400)

depending on whether or not the network

Lex Fridman (1:15:52.160)

is doing what you want it to do.

Jeffrey Shainline (1:15:53.560)

Whereas what I mean in this conversation

Lex Fridman (1:15:56.120)

when I say unsupervised learning

Jeffrey Shainline (1:15:57.600)

is that those synaptic weights

Lex Fridman (1:15:59.400)

are dynamically changing in your network

Jeffrey Shainline (1:16:03.120)

based on nothing that the user is doing,

Lex Fridman (1:16:05.000)

nothing that there's no wire from the outside

Jeffrey Shainline (1:16:07.080)

going into any of those synapses.

Lex Fridman (1:16:09.040)

The network itself is reconfiguring those synaptic weights

Jeffrey Shainline (1:16:12.080)

based on physical properties

Lex Fridman (1:16:15.760)

that you've built into the devices.

Lex Fridman (1:16:17.600)

So if the synapse receives a pulse from here

Lex Fridman (1:16:21.400)

and that causes the neuron to spike,

Jeffrey Shainline (1:16:23.360)

some circuit built in there with no help from me

Lex Fridman (1:16:27.040)

or anybody else adjust the weight

Jeffrey Shainline (1:16:29.200)

in a way that makes it more likely

Lex Fridman (1:16:31.400)

to store the useful information

Lex Fridman (1:16:34.600)

and excite the useful network patterns

Lex Fridman (1:16:36.360)

and makes it less likely that random noise,

Jeffrey Shainline (1:16:39.360)

useless communication events

Lex Fridman (1:16:41.440)

will have an important effect on the network activity.

Lex Fridman (1:16:45.320)

So there's memory encoded in the weights,

Lex Fridman (1:16:48.280)

the synaptic weights.

Lex Fridman (1:16:49.760)

What about the formation of something

Lex Fridman (1:16:51.880)

that's not often done in machine learning,

Lex Fridman (1:16:53.680)

the formation of new synaptic connections?

Lex Fridman (1:16:56.280)

Right, well, that seems to,

Lex Fridman (1:16:57.440)

so again, not a neuroscientist here,

Lex Fridman (1:17:00.120)

but my reading of the literature

Jeffrey Shainline (1:17:01.960)

is that that's particularly crucial

Lex Fridman (1:17:04.000)

in early stages of brain development

Jeffrey Shainline (1:17:06.400)

where a newborn is born

Lex Fridman (1:17:09.160)

with tons of extra synaptic connections

Lex Fridman (1:17:11.680)

and it's actually pruned over time.

Lex Fridman (1:17:13.880)

So the number of synapses decreases

Jeffrey Shainline (1:17:16.800)

as opposed to growing new long distance connections.

Lex Fridman (1:17:19.680)

It is possible in the brain to grow new neurons

Lex Fridman (1:17:22.280)

and assign new synaptic connections

Lex Fridman (1:17:26.080)

but it doesn't seem to be the primary mechanism

Jeffrey Shainline (1:17:29.120)

by which the brain is learning.

Lex Fridman (1:17:31.840)

So for example, like right now,

Jeffrey Shainline (1:17:34.280)

sitting here talking to you,

Lex Fridman (1:17:35.720)

you say lots of interesting things

Lex Fridman (1:17:37.000)

and I learn from you

Lex Fridman (1:17:38.760)

and I can remember things that you just said

Lex Fridman (1:17:41.240)

and I didn't grow new axonal connections

Lex Fridman (1:17:44.720)

down to new synapses to enable those.

Jeffrey Shainline (1:17:47.360)

It's plasticity mechanisms

Lex Fridman (1:17:50.160)

in the synaptic connections between neurons

Jeffrey Shainline (1:17:52.920)

that enable me to learn on that timescale.

Lex Fridman (1:17:55.960)

So at the very least,

Jeffrey Shainline (1:17:57.560)

you can sufficiently approximate that

Lex Fridman (1:17:59.880)

with just weight updates.

Jeffrey Shainline (1:18:01.360)

You don't need to form new connections.

Lex Fridman (1:18:02.920)

I would say weight updates are a big part of it.

Jeffrey Shainline (1:18:05.040)

I also think there's more

Lex Fridman (1:18:06.200)

because broadly speaking,

Jeffrey Shainline (1:18:08.600)

when we're doing machine learning,

Lex Fridman (1:18:10.440)

our networks, say we're talking about feed forward,

Jeffrey Shainline (1:18:12.480)

deep neural networks,

Lex Fridman (1:18:14.120)

the temporal domain is not really part of it.

Jeffrey Shainline (1:18:16.960)

Okay, you're gonna put in an image

Lex Fridman (1:18:18.200)

and you're gonna get out a classification

Lex Fridman (1:18:20.400)

and you're gonna do that as fast as possible.

Lex Fridman (1:18:22.000)

So you care about time

Lex Fridman (1:18:23.160)

but time is not part of the essence of this thing really.

Lex Fridman (1:18:27.560)

Whereas in spiking neural networks,

Lex Fridman (1:18:30.040)

what we see in the brain,

Lex Fridman (1:18:31.760)

time is as crucial as space

Lex Fridman (1:18:33.360)

and they're intimately intertwined

Lex Fridman (1:18:34.600)

as I've tried to say.

Lex Fridman (1:18:36.000)

And so adaptation on different timescales

Lex Fridman (1:18:40.280)

is important not just in memory formation,

Jeffrey Shainline (1:18:44.120)

although it plays a key role there,

Lex Fridman (1:18:45.360)

but also in just keeping the activity

Jeffrey Shainline (1:18:48.240)

in a useful dynamic range.

Lex Fridman (1:18:50.320)

So you have other plasticity mechanisms,

Jeffrey Shainline (1:18:52.520)

not just weight update,

Lex Fridman (1:18:54.200)

or at least not on the timescale

Jeffrey Shainline (1:18:56.760)

of many action potentials,

Lex Fridman (1:18:58.760)

but even on the shorter timescale.

Lex Fridman (1:19:00.200)

So a synapse can become much less efficacious.

Lex Fridman (1:19:04.720)

It can transmit a weaker signal

Jeffrey Shainline (1:19:07.200)

after the second, third, fourth,

Lex Fridman (1:19:08.800)

that can second, third, fourth action potential

Jeffrey Shainline (1:19:11.960)

to occur in a sequence.

Lex Fridman (1:19:13.040)

So that's what's called short term synaptic plasticity,

Jeffrey Shainline (1:19:15.960)

which is a form of learning.

Lex Fridman (1:19:17.600)

You're learning that I'm getting too much stimulus

Jeffrey Shainline (1:19:19.640)

from looking at something bright right now.

Lex Fridman (1:19:21.640)

So I need to tone that down.

Jeffrey Shainline (1:19:24.960)

There's also another really important mechanism

Lex Fridman (1:19:28.080)

in learning that's called metoplasticity.

Lex Fridman (1:19:30.520)

What that seems to be is a way

Lex Fridman (1:19:33.560)

that you change not the weights themselves,

Lex Fridman (1:19:37.400)

but the rate at which the weights change.

Lex Fridman (1:19:40.280)

So when I am in say a lecture hall and my,

Jeffrey Shainline (1:19:45.440)

this is a potentially terrible cartoon example,

Lex Fridman (1:19:48.400)

but let's say I'm in a lecture hall

Lex Fridman (1:19:49.680)

and it's time to learn, right?

Lex Fridman (1:19:51.960)

So my brain will release more,

Jeffrey Shainline (1:19:54.280)

perhaps dopamine or some neuromodulator

Lex Fridman (1:19:57.240)

that's gonna change the rate

Jeffrey Shainline (1:20:00.320)

at which synaptic plasticity occurs.

Lex Fridman (1:20:02.240)

So that can make me more sensitive

Jeffrey Shainline (1:20:03.840)

to learning at certain times,

Lex Fridman (1:20:05.320)

more sensitive to overriding previous information

Lex Fridman (1:20:08.360)

and less sensitive at other times.

Lex Fridman (1:20:10.320)

And finally, as long as I'm rattling off the list,

Jeffrey Shainline (1:20:13.200)

I think another concept that falls in the category

Lex Fridman (1:20:16.480)

of learning or memory adaptation is homeostasis

Jeffrey Shainline (1:20:20.560)

or homeostatic adaptation,

Lex Fridman (1:20:22.440)

where neurons have the ability

Jeffrey Shainline (1:20:24.960)

to control their firing rate.

Lex Fridman (1:20:27.800)

So if one neuron is just like blasting way too much,

Jeffrey Shainline (1:20:31.200)

it will naturally tone itself down.

Lex Fridman (1:20:33.000)

Its threshold will adjust

Lex Fridman (1:20:35.520)

so that it stays in a useful dynamical range.

Lex Fridman (1:20:38.520)

And we see that that's captured in deep neural networks

Jeffrey Shainline (1:20:41.680)

where you don't just change the synaptic weights,

Lex Fridman (1:20:43.320)

but you can also move the thresholds of simple neurons

Jeffrey Shainline (1:20:46.680)

in those models.

Lex Fridman (1:20:47.520)

And so to achieve the spiking neural networks,

Jeffrey Shainline (1:20:53.800)

you want to use,

Lex Fridman (1:20:58.360)

you want to implement the first principles

Jeffrey Shainline (1:21:01.200)

that you mentioned of the temporal

Lex Fridman (1:21:03.400)

and the spatial fractal dynamics here.

Lex Fridman (1:21:07.040)

So you can communicate locally,

Lex Fridman (1:21:09.240)

you can communicate across much greater distances

Lex Fridman (1:21:13.320)

and do the same thing in space

Lex Fridman (1:21:16.000)

and do the same thing in time.

Jeffrey Shainline (1:21:18.040)

Now, you have like a chapter called

Lex Fridman (1:21:21.040)

Superconducting Hardware for Neuromorphic Computing.

Lex Fridman (1:21:24.360)

So what are some ideas that integrate

Lex Fridman (1:21:27.760)

some of the things we've been talking about

Jeffrey Shainline (1:21:29.080)

in terms of the first principles of neuromorphic computing

Lex Fridman (1:21:32.080)

and the ideas that you outline

Lex Fridman (1:21:34.280)

in optoelectronic intelligence?

Lex Fridman (1:21:38.040)

Yeah, so let me start, I guess,

Jeffrey Shainline (1:21:40.920)

on the communication side of things,

Lex Fridman (1:21:42.520)

because that's what led us down this track

Jeffrey Shainline (1:21:46.280)

in the first place.

Lex Fridman (1:21:47.120)

By us, I'm talking about my team of colleagues at NIST,

Jeffrey Shainline (1:21:51.800)

Saeed Han, Bryce Brimavera, Sonia Buckley,

Lex Fridman (1:21:54.800)

Jeff Chiles, Adam McCallum to name,

Jeffrey Shainline (1:21:57.200)

Alex Tate to name a few,

Lex Fridman (1:21:58.720)

our group leaders, Saewoo Nam and Rich Mirren.

Jeffrey Shainline (1:22:01.240)

We've all contributed to this.

Lex Fridman (1:22:02.480)

So this is not me saying necessarily

Jeffrey Shainline (1:22:05.880)

just the things that I've proposed,

Lex Fridman (1:22:07.560)

but sort of where our team's thinking

Jeffrey Shainline (1:22:09.600)

has evolved over the years.

Lex Fridman (1:22:11.560)

Can I quickly ask, what is NIST

Lex Fridman (1:22:14.720)

and where is this amazing group of people located?

Lex Fridman (1:22:18.080)

NIST is the National Institute of Standards and Technology.

Jeffrey Shainline (1:22:23.120)

The larger facility is out in Gaithersburg, Maryland.

Lex Fridman (1:22:26.720)

Our team is located in Boulder, Colorado.

Jeffrey Shainline (1:22:31.960)

NIST is a federal agency under the Department of Commerce.

Lex Fridman (1:22:36.240)

We do a lot with, by we, I mean other people at NIST,

Jeffrey Shainline (1:22:40.160)

do a lot with standards,

Lex Fridman (1:22:43.640)

making sure that we understand the system of units,

Jeffrey Shainline (1:22:46.080)

international system of units, precision measurements.

Lex Fridman (1:22:49.320)

There's a lot going on in electrical engineering,

Jeffrey Shainline (1:22:53.560)

material science.

Lex Fridman (1:22:54.760)

And it's historic.

Jeffrey Shainline (1:22:56.000)

I mean, it's one of those, it's like MIT

Lex Fridman (1:22:58.280)

or something like that.

Jeffrey Shainline (1:22:59.120)

It has a reputation over many decades

Lex Fridman (1:23:00.960)

of just being this really a place

Jeffrey Shainline (1:23:04.240)

where there's a lot of brilliant people have done

Lex Fridman (1:23:06.520)

a lot of amazing things.

Lex Fridman (1:23:07.600)

But in terms of the people in your team,

Lex Fridman (1:23:10.600)

in this team of people involved

Jeffrey Shainline (1:23:12.760)

in the concept we're talking about now,

Lex Fridman (1:23:14.600)

I'm just curious,

Lex Fridman (1:23:15.440)

what kind of disciplines are we talking about?

Lex Fridman (1:23:17.240)

What is it?

Jeffrey Shainline (1:23:18.080)

Mostly physicists and electrical engineers,

Lex Fridman (1:23:20.240)

some material scientists,

Lex Fridman (1:23:23.000)

but I would say,

Lex Fridman (1:23:24.840)

yeah, I think physicists and electrical engineers,

Jeffrey Shainline (1:23:27.240)

my background is in photonics,

Lex Fridman (1:23:29.480)

the use of light for technology.

Lex Fridman (1:23:31.040)

So coming from there, I tend to have found colleagues

Lex Fridman (1:23:36.840)

that are more from that background.

Jeffrey Shainline (1:23:38.240)

Although Adam McConn,

Lex Fridman (1:23:40.240)

more of a superconducting electronics background,

Jeffrey Shainline (1:23:42.720)

we need a diversity of folks.

Lex Fridman (1:23:44.280)

This project is sort of cross disciplinary.

Jeffrey Shainline (1:23:46.840)

I would love to be working more

Lex Fridman (1:23:48.280)

with neuroscientists and things,

Lex Fridman (1:23:50.800)

but we haven't reached that scale yet.

Lex Fridman (1:23:53.880)

But yeah.

Jeffrey Shainline (1:23:54.720)

You're focused on the hardware side,

Lex Fridman (1:23:56.480)

which requires all the disciplines that you mentioned.

Lex Fridman (1:23:59.160)

And then of course,

Lex Fridman (1:24:00.000)

neuroscientists may be a source of inspiration

Jeffrey Shainline (1:24:02.120)

for some of the longterm vision.

Lex Fridman (1:24:04.360)

I would actually call it more than inspiration.

Jeffrey Shainline (1:24:06.240)

I would call it sort of a roadmap.

Lex Fridman (1:24:11.120)

We're not trying to build exactly the brain,

Lex Fridman (1:24:15.000)

but I don't think it's enough to just say,

Lex Fridman (1:24:17.520)

oh, neurons kind of work like that.

Jeffrey Shainline (1:24:19.240)

Let's kind of do that thing.

Lex Fridman (1:24:20.760)

I mean, we're very much following the concepts

Jeffrey Shainline (1:24:25.360)

that the cognitive sciences have laid out for us,

Lex Fridman (1:24:27.440)

which I believe is a really robust roadmap.

Jeffrey Shainline (1:24:30.520)

I mean, just on a little bit of a tangent,

Lex Fridman (1:24:33.040)

it's often stated that we just don't understand the brain.

Lex Fridman (1:24:36.080)

And so it's really hard to replicate it

Lex Fridman (1:24:37.960)

because we just don't know what's going on there.

Lex Fridman (1:24:40.200)

And maybe five or seven years ago,

Lex Fridman (1:24:43.560)

I would have said that,

Lex Fridman (1:24:44.800)

but as I got more interested in the subject,

Lex Fridman (1:24:47.880)

I read more of the neuroscience literature

Lex Fridman (1:24:50.480)

and I was just taken by the exact opposite sense.

Lex Fridman (1:24:53.640)

I can't believe how much they know about this.

Jeffrey Shainline (1:24:55.880)

I can't believe how mathematically rigorous

Lex Fridman (1:24:59.320)

and sort of theoretically complete

Jeffrey Shainline (1:25:02.960)

a lot of the concepts are.

Lex Fridman (1:25:04.240)

That's not to say we understand consciousness

Jeffrey Shainline (1:25:06.680)

or we understand the self or anything like that,

Lex Fridman (1:25:08.560)

but what is the brain doing

Lex Fridman (1:25:11.040)

and why is it doing those things?

Lex Fridman (1:25:13.600)

Neuroscientists have a lot of answers to those questions.

Lex Fridman (1:25:16.000)

So if you're a hardware designer

Lex Fridman (1:25:17.840)

that just wants to get going,

Jeffrey Shainline (1:25:19.440)

whoa, it's pretty clear which direction to go in, I think.

Lex Fridman (1:25:23.000)

Okay, so I love the optimism behind that,

Lex Fridman (1:25:28.280)

but in the implementation of these systems

Lex Fridman (1:25:32.640)

that uses superconductivity, how do you make it happen?

Lex Fridman (1:25:39.320)

So to me, it starts with thinking

Lex Fridman (1:25:41.880)

about the communication network.

Jeffrey Shainline (1:25:43.400)

You know for sure that the ability of each neuron

Lex Fridman (1:25:47.560)

to communicate to many thousands of colleagues

Jeffrey Shainline (1:25:50.560)

across the network is indispensable.

Lex Fridman (1:25:52.360)

I take that as a core principle of my architecture,

Jeffrey Shainline (1:25:56.280)

my thinking on the subject.

Lex Fridman (1:25:58.440)

So coming from a background in photonics,

Jeffrey Shainline (1:26:02.280)

it was very natural to say,

Lex Fridman (1:26:03.560)

okay, we're gonna use light for communication.

Jeffrey Shainline (1:26:05.360)

Just in case listeners may not know,

Lex Fridman (1:26:08.720)

light is often used in communication.

Jeffrey Shainline (1:26:10.840)

I mean, if you think about radio, that's light,

Lex Fridman (1:26:12.720)

it's long wavelengths, but it's electromagnetic radiation.

Jeffrey Shainline (1:26:15.320)

It's the same physical phenomenon

Lex Fridman (1:26:17.640)

obeying exactly the same Maxwell's equations.

Lex Fridman (1:26:20.200)

And then all the way down to fiber, fiber optics.

Lex Fridman (1:26:24.920)

Now you're using visible

Jeffrey Shainline (1:26:26.240)

or near infrared wavelengths of light,

Lex Fridman (1:26:27.800)

but the way you send messages across the ocean

Jeffrey Shainline (1:26:30.360)

is now contemporary over optical fibers.

Lex Fridman (1:26:33.200)

So using light for communication is not a stretch.

Jeffrey Shainline (1:26:37.480)

It makes perfect sense.

Lex Fridman (1:26:38.960)

So you might ask, well, why don't you use light

Lex Fridman (1:26:41.520)

for communication in a conventional microchip?

Lex Fridman (1:26:45.280)

And the answer to that is, I believe, physical.

Jeffrey Shainline (1:26:49.280)

If we had a light source on a silicon chip

Lex Fridman (1:26:53.080)

that was as simple as a transistor,

Jeffrey Shainline (1:26:55.880)

there would not be a processor in the world

Lex Fridman (1:26:58.240)

that didn't use light for communication,

Jeffrey Shainline (1:26:59.760)

at least above some distance.

Lex Fridman (1:27:01.800)

How many light sources are needed?

Jeffrey Shainline (1:27:04.080)

Oh, you need a light source at every single point.

Lex Fridman (1:27:06.840)

A light source per neuron.

Jeffrey Shainline (1:27:08.440)

Per neuron, per little,

Lex Fridman (1:27:09.960)

but then if you could have a really small

Lex Fridman (1:27:13.120)

and nice light source,

Lex Fridman (1:27:15.080)

your definition of neuron could be flexible.

Jeffrey Shainline (1:27:17.960)

Could be, yes, yes.

Lex Fridman (1:27:19.200)

Sometimes it's helpful to me to say,

Jeffrey Shainline (1:27:21.720)

in this hardware, a neuron is that entity

Lex Fridman (1:27:24.560)

which has a light source.

Jeffrey Shainline (1:27:25.720)

That, and I can explain.

Lex Fridman (1:27:27.960)

And then there was light.

Jeffrey Shainline (1:27:29.520)

I mean, I can explain more about that, but.

Lex Fridman (1:27:32.280)

Somehow this like rhymes with consciousness

Jeffrey Shainline (1:27:34.680)

because people will often say the light of consciousness.

Lex Fridman (1:27:38.240)

So that consciousness is that which is conscious.

Jeffrey Shainline (1:27:41.680)

I got it.

Lex Fridman (1:27:43.600)

That's not my quote.

Jeffrey Shainline (1:27:44.840)

That's me, that's my quote.

Lex Fridman (1:27:47.000)

You see, that quote comes from my background.

Jeffrey Shainline (1:27:49.520)

Yours is in optics, mine in light, mine's in darkness.

Lex Fridman (1:27:55.440)

So go ahead.

Lex Fridman (1:27:56.920)

So the point I was making there is that

Lex Fridman (1:27:59.640)

if it was easy to manufacture light sources

Jeffrey Shainline (1:28:02.960)

along with transistors on a silicon chip,

Lex Fridman (1:28:05.760)

they would be everywhere.

Lex Fridman (1:28:07.240)

And it's not easy.

Lex Fridman (1:28:08.880)

People have been trying for decades

Lex Fridman (1:28:10.160)

and it's actually extremely difficult.

Lex Fridman (1:28:11.960)

I think an important part of our research

Jeffrey Shainline (1:28:14.040)

is dwelling right at that spot there.

Lex Fridman (1:28:16.960)

So.

Lex Fridman (1:28:17.800)

Is it physics or engineering?

Lex Fridman (1:28:18.720)

It's physics.

Jeffrey Shainline (1:28:19.560)

So, okay, so it's physics, I think.

Lex Fridman (1:28:22.280)

So what I mean by that is, as we discussed,

Jeffrey Shainline (1:28:26.920)

silicon is the material of choice for transistors

Lex Fridman (1:28:29.400)

and it's very difficult to imagine

Jeffrey Shainline (1:28:33.280)

that that's gonna change anytime soon.

Lex Fridman (1:28:35.200)

Silicon is notoriously bad at emitting light.

Lex Fridman (1:28:39.160)

And that has to do with the immutable properties

Lex Fridman (1:28:43.320)

of silicon itself.

Jeffrey Shainline (1:28:44.160)

The way that the energy bands are structured in silicon,

Lex Fridman (1:28:47.880)

you're never going to make silicon efficient

Jeffrey Shainline (1:28:50.520)

as a light source at room temperature

Lex Fridman (1:28:53.520)

without doing very exotic things

Jeffrey Shainline (1:28:55.640)

that degrade its ability to interface nicely

Lex Fridman (1:28:58.320)

with those transistors in the first place.

Lex Fridman (1:28:59.840)

So that's like one of these things where it's,

Lex Fridman (1:29:02.600)

why is nature dealing us that blow?

Jeffrey Shainline (1:29:05.200)

You give us these beautiful transistors

Lex Fridman (1:29:07.000)

and you give us all the motivation

Jeffrey Shainline (1:29:08.640)

to use light for communication,

Lex Fridman (1:29:10.120)

but then you don't give us a light source.

Jeffrey Shainline (1:29:11.520)

So, well, okay, you do give us a light source.

Lex Fridman (1:29:14.040)

Compound semiconductors,

Jeffrey Shainline (1:29:15.320)

like we talked about back at the beginning,

Lex Fridman (1:29:16.880)

an element from group three and an element from group five

Jeffrey Shainline (1:29:19.640)

form an alloy where every other lattice site

Lex Fridman (1:29:21.800)

switches which element it is.

Jeffrey Shainline (1:29:23.680)

Those have much better properties for generating light.

Lex Fridman (1:29:27.240)

You put electrons in, light comes out.

Jeffrey Shainline (1:29:30.040)

Almost 100% of the electron hold,

Lex Fridman (1:29:33.960)

it can be made efficient.

Jeffrey Shainline (1:29:36.080)

I'll take your word for it, okay.

Lex Fridman (1:29:37.440)

However, I say it's physics, not engineering,

Jeffrey Shainline (1:29:39.600)

because it's very difficult

Lex Fridman (1:29:41.920)

to get those compound semiconductor light sources

Jeffrey Shainline (1:29:45.240)

situated with your silicon.

Lex Fridman (1:29:47.520)

In order to do that ion implantation

Jeffrey Shainline (1:29:49.400)

that I talked about at the beginning,

Lex Fridman (1:29:50.760)

high temperatures are required.

Lex Fridman (1:29:52.720)

So you gotta make all of your transistors first

Lex Fridman (1:29:55.680)

and then put the compound semiconductors on top of there.

Jeffrey Shainline (1:29:58.560)

You can't grow them afterwards

Lex Fridman (1:30:00.800)

because that requires high temperature.

Jeffrey Shainline (1:30:02.360)

It screws up all your transistors.

Lex Fridman (1:30:04.000)

You try and stick them on there.

Jeffrey Shainline (1:30:05.800)

They don't have the same lattice constant.

Lex Fridman (1:30:07.960)

The spacing between atoms is different enough

Jeffrey Shainline (1:30:10.320)

that it just doesn't work.

Lex Fridman (1:30:11.640)

So nature does not seem to be telling us that,

Jeffrey Shainline (1:30:15.520)

hey, go ahead and combine light sources

Lex Fridman (1:30:17.960)

with your digital switches

Jeffrey Shainline (1:30:19.960)

for conventional digital computing.

Lex Fridman (1:30:22.680)

And conventional digital computing

Jeffrey Shainline (1:30:24.720)

will often require smaller scale, I guess,

Lex Fridman (1:30:27.640)

in terms of like smartphone.

Lex Fridman (1:30:30.600)

So in which kind of systems does nature hint

Lex Fridman (1:30:35.440)

that we can use light and photons for communication?

Jeffrey Shainline (1:30:40.440)

Well, so let me just try and be clear.

Lex Fridman (1:30:42.920)

You can use light for communication in digital systems,

Jeffrey Shainline (1:30:46.240)

just the light sources are not intimately integrated

Lex Fridman (1:30:49.640)

with the silicon.

Jeffrey Shainline (1:30:50.480)

You manufacture all the silicon,

Lex Fridman (1:30:52.080)

you have your microchip, plunk it down.

Lex Fridman (1:30:54.680)

And then you manufacture your light sources,

Lex Fridman (1:30:56.640)

separate chip, completely different process

Jeffrey Shainline (1:30:58.720)

made in a different foundry.

Lex Fridman (1:31:00.440)

And then you put those together at the package level.

Lex Fridman (1:31:03.200)

So now you have some,

Lex Fridman (1:31:06.920)

I would say a great deal of architectural limitations

Jeffrey Shainline (1:31:09.480)

that are introduced by that sort of

Lex Fridman (1:31:13.920)

package level integration

Jeffrey Shainline (1:31:15.400)

as opposed to monolithic on the same chip integration,

Lex Fridman (1:31:18.000)

but it's still a very useful thing to do.

Lex Fridman (1:31:19.600)

And that's where I had done some work previously

Lex Fridman (1:31:23.120)

before I came to NIST.

Jeffrey Shainline (1:31:24.080)

There's a project led by Vladimir Stoyanovich

Lex Fridman (1:31:27.920)

that now spun out into a company called IR Labs

Jeffrey Shainline (1:31:30.800)

led by Mark Wade and Chen Sun

Lex Fridman (1:31:33.200)

where they're doing exactly that.

Lex Fridman (1:31:34.440)

So you have your light source chip,

Lex Fridman (1:31:36.440)

your silicon chip, whatever it may be doing,

Jeffrey Shainline (1:31:39.760)

maybe it's digital electronics,

Lex Fridman (1:31:40.880)

maybe it's some other control purpose, something.

Lex Fridman (1:31:43.600)

And the silicon chip drives the light source chip

Lex Fridman (1:31:47.760)

and modulates the intensity of the lights.

Jeffrey Shainline (1:31:49.800)

You can get data out of the package on an optical fiber.

Lex Fridman (1:31:52.640)

And that still gives you tremendous advantages in bandwidth

Jeffrey Shainline (1:31:56.560)

as opposed to sending those signals out

Lex Fridman (1:31:58.800)

over electrical lines.

Lex Fridman (1:32:00.760)

But it is somewhat peculiar to my eye

Lex Fridman (1:32:05.160)

that they have to be integrated at this package level.

Lex Fridman (1:32:07.880)

And those people, I mean, they're so smart.

Lex Fridman (1:32:09.760)

Those are my colleagues that I respect a great deal.

Lex Fridman (1:32:12.880)

So it's very clear that it's not just

Lex Fridman (1:32:16.760)

they're making a bad choice.

Jeffrey Shainline (1:32:18.760)

This is what physics is telling us.

Lex Fridman (1:32:20.480)

It just wouldn't make any sense

Jeffrey Shainline (1:32:22.280)

to try to stick them together.

Lex Fridman (1:32:24.240)

Yeah, so even if it's difficult,

Jeffrey Shainline (1:32:28.160)

it's easier than the alternative, unfortunately.

Lex Fridman (1:32:30.880)

I think so, yes.

Lex Fridman (1:32:31.720)

And again, I need to go back

Lex Fridman (1:32:33.280)

and make sure that I'm not taking the wrong way.

Jeffrey Shainline (1:32:35.080)

I'm not saying that the pursuit

Lex Fridman (1:32:36.800)

of integrating compound semiconductors with silicon

Jeffrey Shainline (1:32:39.400)

is fruitless and shouldn't be pursued.

Lex Fridman (1:32:41.120)

It should, and people are doing great work.

Jeffrey Shainline (1:32:43.440)

Kai Mei Lau and John Bowers, others,

Lex Fridman (1:32:45.720)

they're doing it and they're making progress.

Lex Fridman (1:32:48.240)

But to my eye, it doesn't look like that's ever going to be

Lex Fridman (1:32:53.680)

just the standard monolithic light source

Jeffrey Shainline (1:32:57.720)

on silicon process.

Lex Fridman (1:32:58.600)

I just don't see it.

Jeffrey Shainline (1:33:00.040)

Yeah, so nature kind of points the way usually.

Lex Fridman (1:33:02.840)

And if you resist nature,

Jeffrey Shainline (1:33:04.480)

you're gonna have to do a lot more work.

Lex Fridman (1:33:05.680)

And it's gonna be expensive and not scalable.

Jeffrey Shainline (1:33:07.720)

Got it.

Lex Fridman (1:33:08.560)

But okay, so let's go far into the future.

Jeffrey Shainline (1:33:11.320)

Let's imagine this gigantic neuromorphic computing system

Lex Fridman (1:33:14.760)

that simulates all of our realities.

Jeffrey Shainline (1:33:17.360)

It currently is Mantra Matrix 4.

Lex Fridman (1:33:19.080)

So this thing, this powerful computer,

Lex Fridman (1:33:23.200)

how does it operate?

Lex Fridman (1:33:24.880)

So what are the neurons?

Lex Fridman (1:33:27.520)

What is the communication?

Lex Fridman (1:33:29.040)

What's your sense?

Jeffrey Shainline (1:33:30.000)

All right, so let me now,

Lex Fridman (1:33:32.480)

after spending 45 minutes trashing

Jeffrey Shainline (1:33:34.720)

light source integration with silicon,

Lex Fridman (1:33:36.160)

let me now say why I'm basing my entire life,

Jeffrey Shainline (1:33:40.160)

professional life, on integrating light sources

Lex Fridman (1:33:43.680)

with electronics.

Jeffrey Shainline (1:33:44.960)

I think the game is completely different

Lex Fridman (1:33:47.040)

when you're talking about superconducting electronics.

Jeffrey Shainline (1:33:49.560)

For several reasons, let me try to go through them.

Lex Fridman (1:33:54.240)

One is that, as I mentioned,

Jeffrey Shainline (1:33:56.480)

it's difficult to integrate

Lex Fridman (1:33:57.960)

those compound semiconductor light sources with silicon.

Jeffrey Shainline (1:34:01.280)

With silicon is a requirement that is introduced

Lex Fridman (1:34:04.800)

by the fact that you're using semiconducting electronics.

Jeffrey Shainline (1:34:07.280)

In superconducting electronics,

Lex Fridman (1:34:08.840)

you're still gonna start with a silicon wafer,

Lex Fridman (1:34:10.840)

but it's just the bread for your sandwich in a lot of ways.

Lex Fridman (1:34:13.880)

You're not using that silicon

Jeffrey Shainline (1:34:15.800)

in precisely the same way for the electronics.

Lex Fridman (1:34:17.720)

You're now depositing superconducting materials

Jeffrey Shainline (1:34:20.440)

on top of that.

Lex Fridman (1:34:21.840)

The prospects for integrating light sources

Jeffrey Shainline (1:34:24.520)

with that kind of an electronic process

Lex Fridman (1:34:27.400)

are certainly less explored,

Lex Fridman (1:34:30.480)

but I think much more promising

Lex Fridman (1:34:31.960)

because you don't need those light sources

Jeffrey Shainline (1:34:34.280)

to be intimately integrated with the transistors.

Lex Fridman (1:34:36.600)

That's where the problems come up.

Jeffrey Shainline (1:34:37.920)

They don't need to be lattice matched to the silicon,

Lex Fridman (1:34:39.920)

all that kind of stuff.

Jeffrey Shainline (1:34:41.160)

Instead, it seems possible

Lex Fridman (1:34:43.640)

that you can take those compound semiconductor light sources,

Jeffrey Shainline (1:34:47.320)

stick them on the silicon wafer,

Lex Fridman (1:34:49.160)

and then grow your superconducting electronics

Jeffrey Shainline (1:34:51.400)

on the top of that.

Lex Fridman (1:34:52.320)

It's at least not obviously going to fail.

Lex Fridman (1:34:55.800)

So the computation would be done

Lex Fridman (1:34:57.280)

on the superconductive material as well?

Jeffrey Shainline (1:35:00.120)

Yes, the computation is done

Lex Fridman (1:35:01.920)

in the superconducting electronics,

Lex Fridman (1:35:03.920)

and the light sources receive signals

Lex Fridman (1:35:06.400)

that say, hey, a neuron reached threshold,

Jeffrey Shainline (1:35:08.200)

produce a pulse of light,

Lex Fridman (1:35:09.800)

send it out to all your downstream synaptic connections.

Jeffrey Shainline (1:35:12.480)

Those are, again, superconducting electronics.

Lex Fridman (1:35:16.240)

Perform your computation,

Lex Fridman (1:35:18.000)

and you're off to the races.

Lex Fridman (1:35:19.640)

Your network works.

Lex Fridman (1:35:20.760)

So then if we can rewind real quick,

Lex Fridman (1:35:22.600)

so what are the limitations of the challenges

Jeffrey Shainline (1:35:25.960)

of superconducting electronics

Lex Fridman (1:35:28.920)

when we think about constructing these kinds of systems?

Lex Fridman (1:35:31.480)

So actually, let me say one other thing

Lex Fridman (1:35:35.640)

about the light sources,

Lex Fridman (1:35:37.560)

and then I'll move on, I promise,

Lex Fridman (1:35:39.840)

because this is probably tedious for some.

Jeffrey Shainline (1:35:42.320)

This is super exciting.

Lex Fridman (1:35:44.000)

Okay, one other thing about the light sources.

Jeffrey Shainline (1:35:45.720)

I said that silicon is terrible at emitting photons.

Lex Fridman (1:35:48.920)

It's just not what it's meant to do.

Jeffrey Shainline (1:35:50.640)

However, the game is different

Lex Fridman (1:35:52.800)

when you're at low temperature.

Jeffrey Shainline (1:35:54.080)

If you're working with superconductors,

Lex Fridman (1:35:55.720)

you have to be at low temperature

Jeffrey Shainline (1:35:56.960)

because they don't work otherwise.

Lex Fridman (1:35:58.680)

When you're at four Kelvin,

Jeffrey Shainline (1:36:00.240)

silicon is not obviously a terrible light source.

Lex Fridman (1:36:03.440)

It's still not as efficient as compound semiconductors,

Lex Fridman (1:36:05.840)

but it might be good enough for this application.

Lex Fridman (1:36:08.680)

The final thing that I'll mention about that is, again,

Jeffrey Shainline (1:36:11.320)

leveraging superconductors, as I said,

Lex Fridman (1:36:13.960)

in a different context,

Jeffrey Shainline (1:36:15.360)

superconducting detectors can receive one single photon.

Lex Fridman (1:36:19.520)

In that conversation, I failed to mention

Jeffrey Shainline (1:36:21.240)

that semiconductors can also receive photons.

Lex Fridman (1:36:23.480)

That's the primary mechanism by which it's done.

Jeffrey Shainline (1:36:26.200)

A camera in your phone that's receptive to visible light

Lex Fridman (1:36:29.400)

is receiving photons.

Jeffrey Shainline (1:36:31.080)

It's based on silicon,

Lex Fridman (1:36:32.200)

or you can make it in different semiconductors

Jeffrey Shainline (1:36:34.360)

for different wavelengths,

Lex Fridman (1:36:36.600)

but it requires on the order of a thousand,

Jeffrey Shainline (1:36:39.600)

a few thousand photons to receive a pulse.

Lex Fridman (1:36:43.560)

Now, when you're using a superconducting detector,

Jeffrey Shainline (1:36:46.160)

you need one photon, exactly one.

Lex Fridman (1:36:48.360)

I mean, one or more.

Lex Fridman (1:36:50.840)

So the fact that your synapses can now be based

Lex Fridman (1:36:54.840)

on superconducting detectors

Jeffrey Shainline (1:36:56.680)

instead of semiconducting detectors

Lex Fridman (1:36:58.800)

brings the light levels that are required

Jeffrey Shainline (1:37:00.720)

down by some three orders of magnitude.

Lex Fridman (1:37:03.160)

So now you don't need good light sources.

Jeffrey Shainline (1:37:06.520)

You can have the world's worst light sources.

Lex Fridman (1:37:08.800)

As long as they spit out maybe a few thousand photons

Jeffrey Shainline (1:37:11.840)

every time a neuron fires,

Lex Fridman (1:37:13.800)

you have the hardware principles in place

Jeffrey Shainline (1:37:17.640)

that you might be able to perform

Lex Fridman (1:37:19.640)

this optoelectronic integration.

Jeffrey Shainline (1:37:21.720)

To me optoelectronic integration is, it's just so enticing.

Lex Fridman (1:37:25.040)

We want to be able to leverage electronics for computation,

Jeffrey Shainline (1:37:28.680)

light for communication,

Lex Fridman (1:37:30.400)

working with silicon microelectronics at room temperature

Jeffrey Shainline (1:37:32.800)

that has been exceedingly difficult.

Lex Fridman (1:37:35.000)

And I hope that when we move to the superconducting domain,

Jeffrey Shainline (1:37:40.000)

target a different application space

Lex Fridman (1:37:41.840)

that is neuromorphic instead of digital

Lex Fridman (1:37:44.920)

and use superconducting detectors,

Lex Fridman (1:37:47.680)

maybe optoelectronic integration comes to us.

Jeffrey Shainline (1:37:50.120)

Okay, so there's a bunch of questions.

Lex Fridman (1:37:51.720)

So one is temperature.

Lex Fridman (1:37:53.680)

So in these kinds of hybrid heterogeneous systems,

Lex Fridman (1:37:58.320)

what's the temperature?

Lex Fridman (1:37:59.560)

What are some of the constraints to the operation here?

Lex Fridman (1:38:01.600)

Does it all have to be a four Kelvin as well?

Jeffrey Shainline (1:38:03.560)

Four Kelvin.

Lex Fridman (1:38:04.400)

Everything has to be at four Kelvin.

Jeffrey Shainline (1:38:06.840)

Okay, so what are the other engineering challenges

Lex Fridman (1:38:09.720)

of making this kind of optoelectronic systems?

Jeffrey Shainline (1:38:14.320)

Let me just dwell on that four Kelvin for a second

Lex Fridman (1:38:16.720)

because some people hear four Kelvin

Lex Fridman (1:38:18.280)

and they just get up and leave.

Lex Fridman (1:38:19.280)

They just say, I'm not doing it, you know?

Lex Fridman (1:38:21.480)

And to me, that's very earth centric, species centric.

Lex Fridman (1:38:25.360)

We live in 300 Kelvin.

Lex Fridman (1:38:27.240)

So we want our technologies to operate there too.

Lex Fridman (1:38:29.120)

I totally get it.

Lex Fridman (1:38:30.080)

Yeah, what's zero Celsius?

Lex Fridman (1:38:31.880)

Zero Celsius is 273 Kelvin.

Lex Fridman (1:38:34.520)

So we're talking very, very cold here.

Lex Fridman (1:38:37.640)

This is...

Jeffrey Shainline (1:38:38.480)

Not even Boston cold.

Lex Fridman (1:38:39.400)

No.

Jeffrey Shainline (1:38:40.240)

This is real cold.

Lex Fridman (1:38:42.400)

Yeah.

Jeffrey Shainline (1:38:43.240)

Siberia cold, no.

Lex Fridman (1:38:44.280)

Okay, so just for reference,

Jeffrey Shainline (1:38:45.680)

the temperature of the cosmic microwave background

Lex Fridman (1:38:47.920)

is about 2.7 Kelvin.

Lex Fridman (1:38:49.400)

So we're still warmer than deep space.

Lex Fridman (1:38:51.680)

Yeah, good.

Lex Fridman (1:38:52.960)

So that when the universe dies out,

Lex Fridman (1:38:56.520)

it'll be colder than four K.

Jeffrey Shainline (1:38:57.920)

It's already colder than four K.

Lex Fridman (1:38:59.400)

In the expanses, you know,

Jeffrey Shainline (1:39:01.760)

you don't have to get that far away from the earth

Lex Fridman (1:39:05.000)

in order to drop down to not far from four Kelvin.

Lex Fridman (1:39:08.040)

So what you're saying is the aliens that live at the edge

Lex Fridman (1:39:11.200)

of the observable universe

Jeffrey Shainline (1:39:13.280)

are using superconductive material for their computation.

Lex Fridman (1:39:16.440)

They don't have to live at the edge of the universe.

Jeffrey Shainline (1:39:17.880)

The aliens that are more advanced than us

Lex Fridman (1:39:21.040)

in their solar system are doing this

Jeffrey Shainline (1:39:24.480)

in their asteroid belt.

Lex Fridman (1:39:26.560)

We can get to that.

Jeffrey Shainline (1:39:27.800)

Oh, because they can get that

Lex Fridman (1:39:30.320)

to that temperature easier there?

Jeffrey Shainline (1:39:31.640)

Sure, yeah.

Lex Fridman (1:39:32.480)

All you have to do is reflect the sunlight away

Lex Fridman (1:39:34.120)

and you have a huge headstart.

Lex Fridman (1:39:36.080)

Oh, so the sun is the problem here.

Jeffrey Shainline (1:39:37.680)

Like it's warm here on earth.

Lex Fridman (1:39:39.000)

Got it. Yeah.

Jeffrey Shainline (1:39:39.840)

Okay, so can you...

Lex Fridman (1:39:41.560)

So how do we get to four K?

Jeffrey Shainline (1:39:42.920)

What's...

Lex Fridman (1:39:43.760)

Well, okay, so what I want to say about temperature...

Jeffrey Shainline (1:39:47.320)

Yeah.

Lex Fridman (1:39:48.160)

What I want to say about temperature is that

Jeffrey Shainline (1:39:50.640)

if you can swallow that,

Lex Fridman (1:39:52.600)

if you can say, all right, I give up applications

Jeffrey Shainline (1:39:56.200)

that have to do with my cell phone

Lex Fridman (1:39:58.200)

and the convenience of a laptop on a train

Lex Fridman (1:40:02.040)

and you instead...

Lex Fridman (1:40:03.760)

For me, I'm very much in the scientific head space.

Jeffrey Shainline (1:40:06.440)

I'm not looking at products.

Lex Fridman (1:40:07.960)

I'm not looking at what this will be useful

Jeffrey Shainline (1:40:09.880)

to sell to consumers.

Lex Fridman (1:40:11.000)

Instead, I'm thinking about scientific questions.

Jeffrey Shainline (1:40:13.400)

Well, it's just not that bad to have to work at four Kelvin.

Lex Fridman (1:40:16.560)

We do it all the time in our labs at NIST.

Lex Fridman (1:40:19.000)

And so does...

Lex Fridman (1:40:19.840)

I mean, for reference,

Jeffrey Shainline (1:40:21.440)

the entire quantum computing sector

Lex Fridman (1:40:25.560)

usually has to work at something like 100 millikelvin,

Jeffrey Shainline (1:40:28.680)

50 millikelvin.

Lex Fridman (1:40:29.680)

So now you're talking of another factor of 100

Jeffrey Shainline (1:40:32.120)

even colder than that, a fraction of a degree.

Lex Fridman (1:40:35.160)

And everybody seems to think quantum computing

Jeffrey Shainline (1:40:37.360)

is going to take over the world.

Lex Fridman (1:40:39.200)

It's so much more expensive

Jeffrey Shainline (1:40:40.720)

to have to get that extra factor of 10 or whatever colder.

Lex Fridman (1:40:46.600)

And yet it's not stopping people from investing in that area.

Lex Fridman (1:40:50.280)

And by investing, I mean putting their research into it

Lex Fridman (1:40:53.840)

as well as venture capital or whatever.

Jeffrey Shainline (1:40:55.760)

So...

Lex Fridman (1:40:56.600)

Oh, so based on the energy of what you're commenting on,

Jeffrey Shainline (1:40:59.600)

I'm getting a sense that's one of the criticism

Lex Fridman (1:41:01.960)

of this approach is 4K, 4 Kelvin is a big negative.

Jeffrey Shainline (1:41:06.720)

It is the showstopper for a lot of people.

Lex Fridman (1:41:10.680)

They just, I mean, and understandably,

Jeffrey Shainline (1:41:12.880)

I'm not saying that that's not a consideration.

Lex Fridman (1:41:16.840)

Of course it is.

Jeffrey Shainline (1:41:17.840)

For some...

Lex Fridman (1:41:18.800)

Okay, so different motivations for different people.

Jeffrey Shainline (1:41:21.440)

In the academic world,

Lex Fridman (1:41:23.000)

suppose you spent your whole life

Jeffrey Shainline (1:41:24.400)

learning about silicon microelectronic circuits.

Lex Fridman (1:41:26.760)

You send a design to a foundry,

Jeffrey Shainline (1:41:28.880)

they send you back a chip

Lex Fridman (1:41:30.520)

and you go test it at your tabletop.

Lex Fridman (1:41:33.000)

And now I'm saying,

Lex Fridman (1:41:34.360)

here now learn how to use all these cryogenics

Lex Fridman (1:41:36.520)

so you can do that at 4 Kelvin.

Lex Fridman (1:41:38.440)

No, come on, man.

Jeffrey Shainline (1:41:39.880)

I don't wanna do that.

Lex Fridman (1:41:41.200)

That sounds bad.

Jeffrey Shainline (1:41:42.040)

It's the old momentum, the Titanic of the turning.

Lex Fridman (1:41:44.520)

Yeah, kind of.

Lex Fridman (1:41:45.600)

But you're saying that's not too much of a...

Lex Fridman (1:41:48.360)

When we're looking at large systems

Lex Fridman (1:41:50.360)

and the gain you can potentially get from them,

Lex Fridman (1:41:52.320)

that's not that much of a cost.

Lex Fridman (1:41:53.440)

And when you wanna answer the scientific question

Lex Fridman (1:41:55.160)

about what are the physical limits of cognition?

Jeffrey Shainline (1:41:58.120)

Well, the physical limits,

Lex Fridman (1:41:59.800)

they don't care if you're at 4 Kelvin.

Jeffrey Shainline (1:42:01.440)

If you can perform cognition at a scale

Lex Fridman (1:42:04.560)

orders of magnitude beyond any room temperature technology,

Lex Fridman (1:42:07.680)

but you gotta get cold to do it,

Lex Fridman (1:42:09.760)

you're gonna do it.

Lex Fridman (1:42:10.600)

And to me, that's the interesting application space.

Lex Fridman (1:42:14.600)

It's not even an application space,

Jeffrey Shainline (1:42:16.120)

that's the interesting scientific paradigm.

Lex Fridman (1:42:18.920)

So I personally am not going to let low temperature

Jeffrey Shainline (1:42:22.560)

stop me from realizing a technological domain or realm

Lex Fridman (1:42:29.000)

that is achieving in most ways everything else

Jeffrey Shainline (1:42:33.800)

that I'm looking for in my hardware.

Lex Fridman (1:42:36.160)

So that, okay, that's a big one.

Jeffrey Shainline (1:42:37.640)

Is there other kind of engineering challenges

Lex Fridman (1:42:40.000)

that you envision?

Jeffrey Shainline (1:42:40.840)

Yeah, yeah, yeah.

Lex Fridman (1:42:41.680)

So let me take a moment here

Jeffrey Shainline (1:42:43.120)

because I haven't really described what I mean

Lex Fridman (1:42:45.760)

by a neuron or a network in this particular hardware.

Jeffrey Shainline (1:42:49.000)

Yeah, do you wanna talk about loop neurons

Lex Fridman (1:42:51.640)

and there's so many fascinating...

Lex Fridman (1:42:53.680)

But you just have so many amazing papers

Lex Fridman (1:42:55.960)

that people should definitely check out

Lex Fridman (1:42:57.720)

and the titles alone are just killer.

Lex Fridman (1:42:59.880)

So anyway, go ahead.

Jeffrey Shainline (1:43:01.080)

Right, so let me say big picture,

Lex Fridman (1:43:03.680)

based on optics, photonics for communication,

Jeffrey Shainline (1:43:07.720)

superconducting electronics for computation,

Lex Fridman (1:43:10.120)

how does this all work?

Lex Fridman (1:43:11.520)

So a neuron in this hardware platform

Lex Fridman (1:43:17.480)

can be thought of as circuits

Jeffrey Shainline (1:43:19.520)

that are based on Josephson junctions,

Lex Fridman (1:43:21.080)

like we talked about before,

Jeffrey Shainline (1:43:22.680)

where every time a photon comes in...

Lex Fridman (1:43:25.320)

So let's start by talking about a synapse.

Jeffrey Shainline (1:43:27.120)

A synapse receives a photon, one or more,

Lex Fridman (1:43:29.680)

from a different neuron

Lex Fridman (1:43:31.400)

and it converts that optical signal

Lex Fridman (1:43:33.600)

to an electrical signal.

Jeffrey Shainline (1:43:35.320)

The amount of current that that adds to a loop

Lex Fridman (1:43:38.920)

is controlled by the synaptic weight.

Lex Fridman (1:43:40.840)

So as I said before,

Lex Fridman (1:43:42.360)

you're popping fluxons into a loop, right?

Lex Fridman (1:43:44.520)

So a photon comes in,

Lex Fridman (1:43:46.440)

it hits a superconducting single photon detector,

Jeffrey Shainline (1:43:49.200)

one photon, the absolute physical minimum

Lex Fridman (1:43:52.120)

that you can communicate

Jeffrey Shainline (1:43:53.040)

from one place to another with light.

Lex Fridman (1:43:54.520)

And that detector then converts that

Jeffrey Shainline (1:43:56.120)

into an electrical signal

Lex Fridman (1:43:57.360)

and the amount of signal

Jeffrey Shainline (1:43:58.720)

is correlated with some kind of weight.

Lex Fridman (1:44:01.200)

Yeah, so the synaptic weight will tell you

Lex Fridman (1:44:02.940)

how many fluxons you pop into the loop.

Lex Fridman (1:44:05.440)

It's an analog number.

Jeffrey Shainline (1:44:06.560)

We're doing analog computation now.

Lex Fridman (1:44:08.160)

Well, can you just linger on that?

Lex Fridman (1:44:09.760)

What the heck is a fluxon?

Lex Fridman (1:44:10.980)

Are we supposed to know this?

Jeffrey Shainline (1:44:11.960)

Or is this a funny,

Lex Fridman (1:44:14.220)

is this like the big bang?

Lex Fridman (1:44:15.600)

Is this a funny word for something deeply technical?

Lex Fridman (1:44:18.840)

No, let's try to avoid using the word fluxon

Jeffrey Shainline (1:44:21.040)

because it's not actually necessary.

Lex Fridman (1:44:22.960)

When a photon...

Jeffrey Shainline (1:44:24.360)

It's fun to say though.

Lex Fridman (1:44:25.440)

So it's very necessary, I would say.

Jeffrey Shainline (1:44:29.400)

When a photon hits

Lex Fridman (1:44:30.360)

that superconducting single photon detector,

Jeffrey Shainline (1:44:32.920)

current is added to a superconducting loop.

Lex Fridman (1:44:36.560)

And the amount of current that you add

Jeffrey Shainline (1:44:39.200)

is an analog value,

Lex Fridman (1:44:40.340)

can have eight bit equivalent resolution,

Jeffrey Shainline (1:44:42.980)

something like that.

Lex Fridman (1:44:44.240)

10 bits, maybe.

Jeffrey Shainline (1:44:45.440)

That's amazing, by the way.

Lex Fridman (1:44:46.920)

This is starting to make a lot more sense.

Jeffrey Shainline (1:44:48.480)

When you're using superconductors for this,

Lex Fridman (1:44:50.620)

the energy of that circulating current

Jeffrey Shainline (1:44:54.720)

is less than the energy of that photon.

Lex Fridman (1:44:58.600)

So your energy budget is not destroyed

Jeffrey Shainline (1:45:01.760)

by doing this analog computation.

Lex Fridman (1:45:04.000)

So now in the language of a neuroscientist,

Jeffrey Shainline (1:45:07.080)

you would say that's your postsynaptic signal.

Lex Fridman (1:45:09.180)

You have this current being stored in a loop.

Jeffrey Shainline (1:45:11.680)

You can decide what you wanna do with it.

Lex Fridman (1:45:13.480)

Most likely you're gonna have it decay exponentially.

Lex Fridman (1:45:16.160)

So every single synapse

Lex Fridman (1:45:18.280)

is gonna have some given time constant.

Lex Fridman (1:45:20.960)

And that's determined by putting some resistor

Lex Fridman (1:45:25.360)

in that superconducting loop.

Lex Fridman (1:45:27.040)

So a synapse event occurs when a photon strikes a detector,

Lex Fridman (1:45:31.120)

adds current to that loop, it decays over time.

Jeffrey Shainline (1:45:33.880)

That's the postsynaptic signal.

Lex Fridman (1:45:35.560)

Then you can process that in a dendritic tree.

Jeffrey Shainline (1:45:38.440)

Bryce Primavera and I have a paper

Lex Fridman (1:45:41.080)

that we've submitted about that.

Jeffrey Shainline (1:45:43.420)

For the more neuroscience oriented people,

Lex Fridman (1:45:45.380)

there's a lot of dendritic processing,

Jeffrey Shainline (1:45:47.080)

a lot of plasticity mechanisms you can implement

Lex Fridman (1:45:49.440)

with essentially exactly the same circuits.

Jeffrey Shainline (1:45:51.480)

You have this one simple building block circuit

Lex Fridman (1:45:54.460)

that you can use for a synapse, for a dendrite,

Jeffrey Shainline (1:45:57.240)

for the neuron cell body, for all the plasticity functions.

Lex Fridman (1:46:00.320)

It's all based on the same building block,

Jeffrey Shainline (1:46:02.360)

just tweaking a couple parameters.

Lex Fridman (1:46:03.760)

So this basic building block

Jeffrey Shainline (1:46:05.040)

has both an optical and an electrical component,

Lex Fridman (1:46:07.240)

and then you just build arbitrary large systems with that?

Jeffrey Shainline (1:46:11.680)

Close, you're not at fault

Lex Fridman (1:46:13.520)

for thinking that that's what I meant.

Lex Fridman (1:46:15.000)

What I should say is that if you want it to be a synapse,

Lex Fridman (1:46:18.780)

you tack a superconducting detector onto the front of it.

Lex Fridman (1:46:22.240)

And if you want it to be anything else,

Lex Fridman (1:46:23.440)

there's no optical component.

Jeffrey Shainline (1:46:25.280)

Got it, so at the front,

Lex Fridman (1:46:28.400)

optics in the front, electrical stuff in the back.

Jeffrey Shainline (1:46:32.160)

Electrical, yeah, in the processing

Lex Fridman (1:46:34.120)

and in the output signal that it sends

Jeffrey Shainline (1:46:36.480)

to the next stage of processing further.

Lex Fridman (1:46:39.380)

So the dendritic trees is electrical.

Jeffrey Shainline (1:46:41.480)

It's all electrical.

Lex Fridman (1:46:42.500)

It's all electrical in the superconducting domain.

Jeffrey Shainline (1:46:44.880)

For anybody who's up on their superconducting circuits,

Lex Fridman (1:46:48.560)

it's just based on a DC squid, the most ubiquitous,

Jeffrey Shainline (1:46:52.440)

which is a circuit composed of two Joseph's injunctions.

Lex Fridman (1:46:55.160)

So it's a very bread and butter kind of thing.

Lex Fridman (1:46:58.720)

And then the only place where you go beyond that

Lex Fridman (1:47:00.900)

is the neuron cell body itself.

Jeffrey Shainline (1:47:03.040)

It's receiving all these electrical inputs

Lex Fridman (1:47:05.300)

from the synapses or dendrites

Jeffrey Shainline (1:47:06.840)

or however you've structured that particular unique neuron.

Lex Fridman (1:47:09.840)

And when it reaches its threshold,

Jeffrey Shainline (1:47:12.320)

which occurs by driving a Joseph's injunction

Lex Fridman (1:47:14.360)

above its critical current,

Jeffrey Shainline (1:47:15.760)

it produces a pulse of current,

Lex Fridman (1:47:17.200)

which starts an amplification sequence,

Jeffrey Shainline (1:47:19.460)

voltage amplification,

Lex Fridman (1:47:21.380)

that produces light out of a transmitter.

Lex Fridman (1:47:24.480)

So one of our colleagues, Adam McCann,

Lex Fridman (1:47:26.880)

and Sonia Buckley as well,

Jeffrey Shainline (1:47:27.960)

did a lot of work on the light sources

Lex Fridman (1:47:30.980)

and the amplifiers that drive the current

Lex Fridman (1:47:34.660)

and produce sufficient voltage to drive current

Lex Fridman (1:47:37.560)

through that now semiconducting part.

Lex Fridman (1:47:39.160)

So that light source is the semiconducting part of a neuron.

Lex Fridman (1:47:43.160)

And that, so the neuron has reached threshold.

Jeffrey Shainline (1:47:45.620)

It produces a pulse of light.

Lex Fridman (1:47:47.640)

That light then fans out across a network of wave guides

Jeffrey Shainline (1:47:51.560)

to reach all the downstream synaptic terminals

Lex Fridman (1:47:54.520)

that perform this process themselves.

Lex Fridman (1:47:57.320)

So it's probably worth explaining

Lex Fridman (1:47:59.880)

what a network of wave guides is,

Jeffrey Shainline (1:48:02.320)

because a lot of listeners aren't gonna know that.

Lex Fridman (1:48:04.780)

Look up the papers by Jeff Chiles on this one.

Lex Fridman (1:48:07.120)

But basically, light can be guided in a simple,

Lex Fridman (1:48:11.380)

basically wire of usually an insulating material.

Lex Fridman (1:48:14.880)

So silicon, silicon nitride,

Lex Fridman (1:48:18.280)

different kinds of glass,

Jeffrey Shainline (1:48:20.040)

just like in a fiber optic, it's glass, silicon dioxide.

Lex Fridman (1:48:23.440)

That makes it a little bit big.

Jeffrey Shainline (1:48:24.840)

We wanna bring these down.

Lex Fridman (1:48:26.160)

So we use different materials like silicon nitride,

Lex Fridman (1:48:28.040)

but basically just imagine a rectangle of some material

Lex Fridman (1:48:32.980)

that just goes and branches,

Jeffrey Shainline (1:48:37.080)

forms different branch points

Lex Fridman (1:48:39.980)

that target different subregions of the network.

Jeffrey Shainline (1:48:43.060)

You can transition between layers of these.

Lex Fridman (1:48:45.000)

So now we're talking about building in the third dimension,

Jeffrey Shainline (1:48:47.160)

which is absolutely crucial.

Lex Fridman (1:48:48.840)

So that's what wave guides are.

Jeffrey Shainline (1:48:50.720)

Yeah, that's great.

Lex Fridman (1:48:52.040)

Why the third dimension is crucial?

Jeffrey Shainline (1:48:54.680)

Okay, so yes, you were talking about

Lex Fridman (1:48:56.640)

what are some of the technical limitations.

Jeffrey Shainline (1:48:59.660)

One of the things that I believe we have to grapple with

Lex Fridman (1:49:04.280)

is that our brains are miraculously compact.

Jeffrey Shainline (1:49:08.800)

For the number of neurons that are in our brain,

Lex Fridman (1:49:11.560)

it sure does fit in a small volume,

Jeffrey Shainline (1:49:13.520)

as it would have to if we're gonna be biological organisms

Lex Fridman (1:49:16.200)

that are resource limited and things like that.

Jeffrey Shainline (1:49:19.260)

Any kind of hardware neuron

Lex Fridman (1:49:20.960)

is almost certainly gonna be much bigger than that

Jeffrey Shainline (1:49:23.460)

if it is of comparable complexity,

Lex Fridman (1:49:26.480)

whether it's based on silicon transistors.

Jeffrey Shainline (1:49:28.540)

Okay, a transistor, seven nanometers,

Lex Fridman (1:49:30.720)

that doesn't mean a semiconductor based neuron

Jeffrey Shainline (1:49:33.760)

is seven nanometers.

Lex Fridman (1:49:34.680)

They're big.

Jeffrey Shainline (1:49:35.920)

They require many transistors,

Lex Fridman (1:49:38.780)

different other things like capacitors and things

Jeffrey Shainline (1:49:40.640)

that store charge.

Lex Fridman (1:49:41.540)

They end up being on the order of 100 microns

Jeffrey Shainline (1:49:44.660)

by 100 microns,

Lex Fridman (1:49:45.880)

and it's difficult to get them down any smaller than that.

Jeffrey Shainline (1:49:48.240)

The same is true for superconducting neurons,

Lex Fridman (1:49:50.600)

and the same is true

Jeffrey Shainline (1:49:52.080)

if we're trying to use light for communication.

Lex Fridman (1:49:54.000)

Even if you're using electrons for communication,

Jeffrey Shainline (1:49:56.860)

you have these wires where, okay,

Lex Fridman (1:50:00.680)

the size of an electron might be angstroms,

Lex Fridman (1:50:03.360)

but the size of a wire is not angstroms,

Lex Fridman (1:50:05.560)

and if you try and make it narrower,

Jeffrey Shainline (1:50:07.160)

the resistance just goes up,

Lex Fridman (1:50:08.520)

so you don't actually win.

Jeffrey Shainline (1:50:10.760)

To communicate over long distances,

Lex Fridman (1:50:12.360)

you need your wires to be microns wide,

Lex Fridman (1:50:15.640)

and it's the same thing for wave guides.

Lex Fridman (1:50:17.160)

Wave guides are essentially limited

Jeffrey Shainline (1:50:18.920)

by the wavelength of light,

Lex Fridman (1:50:20.280)

and that's gonna be about a micron,

Lex Fridman (1:50:21.920)

so whereas compare that to an axon,

Lex Fridman (1:50:24.680)

the analogous component in the brain,

Jeffrey Shainline (1:50:26.760)

which is 10 nanometers in diameter, something like that,

Lex Fridman (1:50:32.160)

they're bigger when they need to communicate

Jeffrey Shainline (1:50:33.680)

over long distances,

Lex Fridman (1:50:34.640)

but grappling with the size of these structures

Jeffrey Shainline (1:50:37.960)

is inevitable and crucial,

Lex Fridman (1:50:39.820)

and so in order to make systems of comparable scale

Jeffrey Shainline (1:50:45.000)

to the human brain, by scale here,

Lex Fridman (1:50:46.880)

I mean number of interconnected neurons,

Jeffrey Shainline (1:50:49.840)

you absolutely have to be using

Lex Fridman (1:50:51.760)

the third spatial dimension,

Lex Fridman (1:50:53.640)

and that means on the wafer,

Lex Fridman (1:50:55.960)

you need multiple layers

Jeffrey Shainline (1:50:57.360)

of both active and passive components.

Lex Fridman (1:50:59.600)

Active, I mean superconducting electronic circuits

Jeffrey Shainline (1:51:03.360)

that are performing computations,

Lex Fridman (1:51:05.360)

and passive, I mean these wave guides

Jeffrey Shainline (1:51:07.400)

that are routing the optical signals to different places,

Lex Fridman (1:51:10.140)

you have to be able to stack those.

Jeffrey Shainline (1:51:11.720)

If you can get to something like 10 planes

Lex Fridman (1:51:14.720)

of each of those, or maybe not even 10,

Jeffrey Shainline (1:51:16.160)

maybe five, six, something like that,

Lex Fridman (1:51:18.840)

then you're in business.

Jeffrey Shainline (1:51:19.760)

Now you can get millions of neurons on a wafer,

Lex Fridman (1:51:22.860)

but that's not anywhere close to the brain scale.

Jeffrey Shainline (1:51:26.380)

In order to get to the scale of the human brain,

Lex Fridman (1:51:27.960)

you're gonna have to also use the third dimension

Jeffrey Shainline (1:51:30.000)

in the sense that entire wafers

Lex Fridman (1:51:32.560)

need to be stacked on top of each other

Jeffrey Shainline (1:51:34.160)

with fiber optic communication between them,

Lex Fridman (1:51:36.080)

and we need to be able to fill a space

Jeffrey Shainline (1:51:38.640)

the size of this table with stacked wafers,

Lex Fridman (1:51:42.040)

and that's when you can get to some 10 billion neurons

Jeffrey Shainline (1:51:44.280)

like your human brain,

Lex Fridman (1:51:45.120)

and I don't think that's specific

Jeffrey Shainline (1:51:46.680)

to the optoelectronic approach that we're taking.

Lex Fridman (1:51:48.800)

I think that applies to any hardware

Jeffrey Shainline (1:51:51.400)

where you're trying to reach commensurate scale

Lex Fridman (1:51:53.560)

and complexity as the human brain.

Lex Fridman (1:51:55.000)

So you need that fractal stacking,

Lex Fridman (1:51:57.520)

so stacking on the wafer,

Lex Fridman (1:51:59.360)

and stacking of the wafers,

Lex Fridman (1:52:01.120)

and then whatever the system that combines,

Jeffrey Shainline (1:52:03.840)

this stacking of the tables with the wafers.

Lex Fridman (1:52:06.360)

And it has to be fractal all the way,

Jeffrey Shainline (1:52:07.720)

you're exactly right,

Lex Fridman (1:52:08.600)

because that's the only way

Jeffrey Shainline (1:52:10.060)

that you can efficiently get information

Lex Fridman (1:52:12.500)

from a small point to across that whole network.

Jeffrey Shainline (1:52:15.060)

It has to have the power law connected.

Lex Fridman (1:52:17.520)

And photons are like optics throughout.

Jeffrey Shainline (1:52:20.640)

Yeah, absolutely.

Lex Fridman (1:52:21.480)

Once you're at this scale, to me it's just obvious.

Jeffrey Shainline (1:52:23.680)

Of course you're using light for communication.

Lex Fridman (1:52:25.580)

You have fiber optics given to us from nature, so simple.

Jeffrey Shainline (1:52:30.580)

The thought of even trying to do

Lex Fridman (1:52:32.860)

any kind of electrical communication

Jeffrey Shainline (1:52:34.740)

just doesn't make sense to me.

Lex Fridman (1:52:37.140)

I'm not saying it's wrong, I don't know,

Lex Fridman (1:52:39.260)

but that's where I'm coming from.

Lex Fridman (1:52:40.980)

So let's return to loop neurons.

Lex Fridman (1:52:43.860)

Why are they called loop neurons?

Lex Fridman (1:52:46.460)

Yeah, the term loop neurons comes from the fact,

Jeffrey Shainline (1:52:48.900)

like we've been talking about,

Lex Fridman (1:52:49.980)

that they rely heavily on these superconducting loops.

Lex Fridman (1:52:53.260)

So even in a lot of forms of digital computing

Lex Fridman (1:52:57.420)

with superconductors,

Jeffrey Shainline (1:52:58.740)

storing a signal in a superconducting loop

Lex Fridman (1:53:02.540)

is a primary technique.

Jeffrey Shainline (1:53:05.060)

In this particular case,

Lex Fridman (1:53:06.660)

it's just loops everywhere you look.

Lex Fridman (1:53:08.620)

So the strength of a synaptic weight

Lex Fridman (1:53:11.620)

is gonna be set by the amount of current circulating

Jeffrey Shainline (1:53:15.060)

in a loop that is coupled to the synapse.

Lex Fridman (1:53:17.820)

So memory is implemented as current circulating

Jeffrey Shainline (1:53:22.820)

in a superconducting loop.

Lex Fridman (1:53:24.140)

The coupling between, say, a synapse and a dendrite

Jeffrey Shainline (1:53:27.140)

or a synapse in the neuron cell body

Lex Fridman (1:53:29.260)

occurs through loop coupling through transformers.

Lex Fridman (1:53:33.140)

So current circulating in a synapse

Lex Fridman (1:53:34.820)

is gonna induce current in a different loop,

Jeffrey Shainline (1:53:37.580)

a receiving loop in the neuron cell body.

Lex Fridman (1:53:40.820)

So since all of the computation is happening

Jeffrey Shainline (1:53:44.500)

in these flux storage loops

Lex Fridman (1:53:46.660)

and they play such a central role

Jeffrey Shainline (1:53:48.380)

in how the information is processed,

Lex Fridman (1:53:50.260)

how memories are formed, all that stuff,

Jeffrey Shainline (1:53:52.820)

I didn't think too much about it,

Lex Fridman (1:53:53.980)

I just called them loop neurons

Jeffrey Shainline (1:53:55.540)

because it rolls off the tongue a little bit better

Lex Fridman (1:53:58.180)

than superconducting optoelectronic neurons.

Lex Fridman (1:54:02.220)

Okay, so how do you design circuits for these loop neurons?

Lex Fridman (1:54:08.540)

That's a great question.

Jeffrey Shainline (1:54:09.740)

There's a lot of different scales of design.

Lex Fridman (1:54:12.340)

So at the level of just one synapse,

Jeffrey Shainline (1:54:16.340)

you can use conventional methods.

Lex Fridman (1:54:18.980)

They're not that complicated

Jeffrey Shainline (1:54:21.100)

as far as superconducting electronics goes.

Lex Fridman (1:54:23.220)

It's just four Joseph's injunctions or something like that

Jeffrey Shainline (1:54:27.100)

depending on how much complexity you wanna add.

Lex Fridman (1:54:29.180)

So you can just directly simulate each component in SPICE.

Lex Fridman (1:54:34.940)

What's SPICE?

Lex Fridman (1:54:35.860)

It's Standard Electrical Simulation Software, basically.

Lex Fridman (1:54:39.220)

So you're just explicitly solving the differential equations

Lex Fridman (1:54:42.580)

that describe the circuit elements.

Lex Fridman (1:54:44.340)

And then you can stack these things together

Lex Fridman (1:54:46.220)

in that simulation software to then build circuits.

Jeffrey Shainline (1:54:48.860)

You can, but that becomes computationally expensive.

Lex Fridman (1:54:51.580)

So one of the things when COVID hit,

Jeffrey Shainline (1:54:54.100)

we knew we had to turn some attention

Lex Fridman (1:54:55.780)

to more things you can do at home in your basement

Jeffrey Shainline (1:54:59.380)

or whatever, and one of them was computational modeling.

Lex Fridman (1:55:02.700)

So we started working on adapting,

Jeffrey Shainline (1:55:07.780)

abstracting out the circuit performance

Lex Fridman (1:55:10.180)

so that you don't have to explicitly solve

Jeffrey Shainline (1:55:12.540)

the circuit equations, which for Joseph's injunctions

Lex Fridman (1:55:15.860)

usually needs to be done on like a picosecond timescale

Lex Fridman (1:55:18.460)

and you have a lot of nodes in your circuit.

Lex Fridman (1:55:21.220)

So it results in a lot of differential equations

Jeffrey Shainline (1:55:24.860)

that need to be solved simultaneously.

Lex Fridman (1:55:26.340)

We were looking for a way to simulate these circuits

Jeffrey Shainline (1:55:29.420)

that is scalable up to networks of millions or so neurons

Lex Fridman (1:55:33.740)

is sort of where we're targeting right now.

Lex Fridman (1:55:36.620)

So we were able to analyze the behavior of these circuits.

Lex Fridman (1:55:40.820)

And as I said, it's based on these simple building blocks.

Lex Fridman (1:55:43.620)

So you really only need to understand

Lex Fridman (1:55:45.140)

this one building block.

Lex Fridman (1:55:46.340)

And if you get a good model of that, boom, it tiles.

Lex Fridman (1:55:48.980)

And you can change the parameters in there

Jeffrey Shainline (1:55:51.140)

to get different behaviors and stuff,

Lex Fridman (1:55:52.860)

but it's all based on now it's one differential equation

Jeffrey Shainline (1:55:56.180)

that you need to solve.

Lex Fridman (1:55:57.140)

So one differential equation for every synapse,

Jeffrey Shainline (1:56:00.780)

dendrite or neuron in your system.

Lex Fridman (1:56:03.740)

And for the neuroscientists out there,

Jeffrey Shainline (1:56:05.340)

it's just a simple leaky integrate and fire model,

Lex Fridman (1:56:08.420)

leaky integrator, basically.

Jeffrey Shainline (1:56:10.300)

A synapse is a leaky integrator,

Lex Fridman (1:56:11.860)

a dendrite is a leaky integrator.

Lex Fridman (1:56:13.460)

So I'm really fascinated by how this one simple component

Lex Fridman (1:56:18.460)

can be used to achieve lots of different types

Jeffrey Shainline (1:56:22.180)

of dynamical activity.

Lex Fridman (1:56:24.420)

And to me, that's where scalability comes from.

Lex Fridman (1:56:27.500)

And also complexity as well.

Lex Fridman (1:56:29.180)

Complexity is often characterized

Jeffrey Shainline (1:56:30.860)

by relatively simple building blocks

Lex Fridman (1:56:35.420)

connected in potentially simple

Jeffrey Shainline (1:56:37.820)

or sometimes complicated ways,

Lex Fridman (1:56:39.340)

and then emergent new behavior that was hard to predict

Jeffrey Shainline (1:56:41.940)

from those simple elements.

Lex Fridman (1:56:44.660)

And that's exactly what we're working with here.

Lex Fridman (1:56:46.620)

So it's a very exciting platform,

Lex Fridman (1:56:49.020)

both from a modeling perspective

Lex Fridman (1:56:50.380)

and from a hardware manifestation perspective

Lex Fridman (1:56:52.740)

where we can hopefully start to have this test bed

Jeffrey Shainline (1:56:57.380)

where we can explore things,

Lex Fridman (1:56:58.860)

not just related to neuroscience,

Lex Fridman (1:57:00.820)

but also related to other things

Lex Fridman (1:57:03.300)

that connected to other physics like critical phenomenon,

Jeffrey Shainline (1:57:07.140)

Ising models, things like that.

Lex Fridman (1:57:08.660)

So you were asking how we simulate these circuits.

Jeffrey Shainline (1:57:13.060)

It's at different levels

Lex Fridman (1:57:14.540)

and we've got the simple spice circuit stuff.

Jeffrey Shainline (1:57:18.300)

That's no problem.

Lex Fridman (1:57:19.540)

And now we're building these network models

Jeffrey Shainline (1:57:21.740)

based on this more efficient leaky integrator.

Lex Fridman (1:57:23.620)

So we can actually reduce every element

Jeffrey Shainline (1:57:26.220)

to one differential equation.

Lex Fridman (1:57:27.460)

And then we can also step through it

Jeffrey Shainline (1:57:28.860)

on a much coarser time grid.

Lex Fridman (1:57:30.700)

So it ends up being something like a factor

Jeffrey Shainline (1:57:32.380)

of a thousand to 10,000 speed improvement,

Lex Fridman (1:57:35.380)

which allows us to simulate,

Lex Fridman (1:57:37.740)

but hopefully up to millions of neurons.

Lex Fridman (1:57:40.540)

Whereas before we would have been limited to tens,

Jeffrey Shainline (1:57:44.660)

a hundred, something like that.

Lex Fridman (1:57:45.780)

And just like simulating quantum mechanical systems

Jeffrey Shainline (1:57:48.780)

with a quantum computer.

Lex Fridman (1:57:49.860)

So the goal here is to understand such systems.

Jeffrey Shainline (1:57:53.380)

For me, the goal is to study this

Lex Fridman (1:57:55.900)

as a scientific physical system.

Jeffrey Shainline (1:57:58.940)

I'm not drawn towards turning this

Lex Fridman (1:58:01.940)

into an enterprise at this point.

Jeffrey Shainline (1:58:03.460)

I feel short term applications

Lex Fridman (1:58:05.820)

that obviously make a lot of money

Jeffrey Shainline (1:58:07.460)

is not necessarily a curiosity driver for you at the moment.

Lex Fridman (1:58:11.380)

Absolutely not.

Jeffrey Shainline (1:58:12.220)

If you're interested in short term making money,

Lex Fridman (1:58:14.020)

go with deep learning, use silicon microelectronics.

Jeffrey Shainline (1:58:16.500)

If you wanna understand things like the physics

Lex Fridman (1:58:20.900)

of a fascinating system,

Jeffrey Shainline (1:58:23.020)

or if you wanna understand something more

Lex Fridman (1:58:25.620)

along the lines of the physical limits

Jeffrey Shainline (1:58:27.820)

of what can be achieved,

Lex Fridman (1:58:29.460)

then I think single photon communication,

Jeffrey Shainline (1:58:32.700)

superconducting electronics is extremely exciting.

Lex Fridman (1:58:35.940)

What if I wanna use superconducting hardware

Lex Fridman (1:58:39.620)

at four Kelvin to mine Bitcoin?

Lex Fridman (1:58:42.020)

That's my main interest.

Jeffrey Shainline (1:58:44.220)

The reason I wanted to talk to you today,

Lex Fridman (1:58:45.860)

I wanna say, no, I don't know.

Lex Fridman (1:58:47.660)

What's Bitcoin?

Lex Fridman (1:58:51.540)

Look it up on the internet.

Jeffrey Shainline (1:58:53.140)

Somebody told me about it.

Lex Fridman (1:58:54.780)

I'm not sure exactly what it is.

Lex Fridman (1:58:57.500)

But let me ask nevertheless

Lex Fridman (1:58:59.060)

about applications to machine learning.

Jeffrey Shainline (1:59:01.380)

Okay, so if you look at the scale of five, 10, 20 years,

Lex Fridman (1:59:07.300)

is it possible to, before we understand the nature

Jeffrey Shainline (1:59:11.900)

of human intelligence and general intelligence,

Lex Fridman (1:59:14.420)

do you think we'll start falling out of this exploration

Jeffrey Shainline (1:59:19.100)

of neuromorphic systems ability to solve some

Lex Fridman (1:59:23.100)

of the problems that the machine learning systems

Lex Fridman (1:59:25.180)

of today can't solve?

Lex Fridman (1:59:26.420)

Well, I'm really hesitant to over promise.

Lex Fridman (1:59:31.620)

So I really don't know.

Lex Fridman (1:59:34.100)

Also, I don't really understand machine learning

Jeffrey Shainline (1:59:36.740)

in a lot of senses.

Lex Fridman (1:59:37.580)

I mean, machine learning from my perspective appears

Jeffrey Shainline (1:59:42.900)

to require that you know precisely what your input is

Lex Fridman (1:59:49.180)

and also what your goal is.

Jeffrey Shainline (1:59:51.700)

You usually have some objective function

Lex Fridman (1:59:53.300)

or something like that.

Lex Fridman (1:59:54.140)

And that's very limiting.

Lex Fridman (1:59:57.300)

I mean, of course, a lot of times that's the case.

Jeffrey Shainline (20:01.720)

That's exactly right.

Lex Fridman (20:03.060)

So this also, this is gonna be the most controversial part

Jeffrey Shainline (20:05.820)

of our conversation where you're gonna make some enemies.

Lex Fridman (20:07.800)

So let me ask,

Jeffrey Shainline (20:09.040)

because we've been talking about physics and engineering.

Lex Fridman (20:14.040)

Which group of people is smarter

Lex Fridman (20:15.760)

and more important for this one?

Lex Fridman (20:17.800)

Let me ask the question in a better way.

Jeffrey Shainline (20:20.560)

Some of the big innovations,

Lex Fridman (20:22.560)

some of the beautiful things that we've been talking about,

Lex Fridman (20:25.720)

how much of it is physics?

Lex Fridman (20:26.960)

How much of it is engineering?

Jeffrey Shainline (20:28.440)

My dad is a physicist and he talks down

Lex Fridman (20:31.640)

to all the amazing engineering that we're doing

Jeffrey Shainline (20:34.400)

in the artificial intelligence and the computer science

Lex Fridman (20:37.920)

and the robotics and all that space.

Lex Fridman (20:39.560)

So we argue about this all the time.

Lex Fridman (20:41.640)

So what do you think?

Lex Fridman (20:42.480)

Who gets more credit?

Lex Fridman (20:43.920)

I'm genuinely not trying to just be politically correct here.

Jeffrey Shainline (20:46.760)

I don't see how you would have any of the,

Lex Fridman (20:50.480)

what we consider sort of the great accomplishments

Jeffrey Shainline (20:52.600)

of society without both.

Lex Fridman (20:54.120)

You absolutely need both of those things.

Jeffrey Shainline (20:55.880)

Physics tends to play a key role earlier in the development

Lex Fridman (20:59.640)

and then engineering optimization, these things take over.

Lex Fridman (21:04.640)

And I mean, the invention of the transistor

Lex Fridman (21:09.220)

or actually even before that,

Jeffrey Shainline (21:10.840)

the understanding of semiconductor physics

Lex Fridman (21:12.840)

that allowed the invention of the transistor,

Jeffrey Shainline (21:14.800)

that's all physics.

Lex Fridman (21:15.640)

So if you didn't have that physics,

Jeffrey Shainline (21:17.000)

you don't even get to get on the field.

Lex Fridman (21:20.300)

But once you have understood and demonstrated

Jeffrey Shainline (21:24.200)

that this is in principle possible,

Lex Fridman (21:26.520)

more so as engineering.

Lex Fridman (21:28.400)

Why we have computers more powerful

Lex Fridman (21:32.200)

than old supercomputers in each of our phones,

Jeffrey Shainline (21:36.400)

that's all engineering.

Lex Fridman (21:37.520)

And I think I would be quite foolish to say

Jeffrey Shainline (21:41.720)

that that's not valuable, that's not a great contribution.

Lex Fridman (21:46.920)

It's a beautiful dance.

Jeffrey Shainline (21:47.800)

Would you put like Silicon,

Lex Fridman (21:49.740)

the understanding of the material properties

Lex Fridman (21:52.760)

in the space of engineering?

Lex Fridman (21:54.320)

Like how does that whole process work?

Jeffrey Shainline (21:55.680)

To understand that it has all these nice properties

Lex Fridman (21:58.000)

or even the development of photolithography,

Jeffrey Shainline (22:02.240)

is that basically,

Lex Fridman (22:03.840)

would you put that in a category of engineering?

Jeffrey Shainline (22:06.140)

No, I would say that it is basic physics,

Lex Fridman (22:09.960)

it is applied physics, it's material science,

Jeffrey Shainline (22:12.760)

it's X ray crystallography, it's polymer chemistry,

Lex Fridman (22:17.640)

it's everything.

Lex Fridman (22:18.800)

Chemistry even is thrown in there?

Lex Fridman (22:20.280)

Absolutely, yes, absolutely.

Jeffrey Shainline (22:22.580)

Just no biology.

Lex Fridman (22:25.320)

We can get to biology.

Jeffrey Shainline (22:26.560)

Or the biologies and the humans

Lex Fridman (22:28.240)

that are engineering the system,

Lex Fridman (22:29.480)

so it's all integrated deeply.

Lex Fridman (22:31.200)

Okay, so let's return,

Jeffrey Shainline (22:32.560)

you mentioned this word superconductivity.

Lex Fridman (22:35.640)

So what does that have to do with what we're talking about?

Jeffrey Shainline (22:38.600)

Right, okay, so in a semiconductor,

Lex Fridman (22:40.560)

as I tried to describe a second ago,

Jeffrey Shainline (22:44.280)

you can sort of induce currents by applying voltages

Lex Fridman (22:50.040)

and those have sort of typical properties

Jeffrey Shainline (22:52.280)

that you would expect from some kind of a conductor.

Lex Fridman (22:55.400)

Those electrons, they don't just flow

Jeffrey Shainline (22:59.000)

perfectly without dissipation.

Lex Fridman (23:00.360)

If an electron collides with an imperfection in the lattice

Jeffrey Shainline (23:03.240)

or another electron, it's gonna slow down,

Lex Fridman (23:05.600)

it's gonna lose its momentum.

Lex Fridman (23:06.880)

So you have to keep applying that voltage

Lex Fridman (23:09.280)

in order to keep the current flowing.

Jeffrey Shainline (23:11.440)

In a superconductor, something different happens.

Lex Fridman (23:13.440)

If you get a current to start flowing,

Jeffrey Shainline (23:16.520)

it will continue to flow indefinitely.

Lex Fridman (23:18.280)

There's no dissipation.

Lex Fridman (23:19.880)

So that's crazy.

Lex Fridman (23:21.660)

How does that happen?

Jeffrey Shainline (23:22.500)

Well, it happens at low temperature and this is crucial.

Lex Fridman (23:26.800)

It has to be a quite low temperature

Lex Fridman (23:30.200)

and what I'm talking about there,

Lex Fridman (23:32.840)

for essentially all of our conversation,

Jeffrey Shainline (23:35.800)

I'm gonna be talking about conventional superconductors,

Lex Fridman (23:39.800)

sometimes called low TC superconductors,

Jeffrey Shainline (23:42.240)

low critical temperature superconductors.

Lex Fridman (23:45.120)

And so those materials have to be at a temperature around,

Jeffrey Shainline (23:50.640)

say around four Kelvin.

Lex Fridman (23:51.960)

I mean, their critical temperature might be 10 Kelvin,

Jeffrey Shainline (23:54.560)

something like that,

Lex Fridman (23:55.400)

but you wanna operate them at around four Kelvin,

Jeffrey Shainline (23:57.120)

four degrees above absolute zero.

Lex Fridman (23:59.280)

And what happens at that temperature,

Jeffrey Shainline (24:01.640)

at very low temperatures in certain materials

Lex Fridman (24:03.880)

is that the noise of atoms moving around,

Jeffrey Shainline (24:10.080)

the lattice vibrating, electrons colliding with each other,

Lex Fridman (24:13.580)

that becomes sufficiently low

Jeffrey Shainline (24:15.120)

that the electrons can settle into this very special state.

Lex Fridman (24:18.600)

It's sometimes referred to as a macroscopic quantum state

Jeffrey Shainline (24:22.280)

because if I had a piece of superconducting material here,

Lex Fridman (24:26.560)

let's say niobium is a very typical superconductor.

Jeffrey Shainline (24:30.920)

If I had a block of niobium here

Lex Fridman (24:33.640)

and we cooled it below its critical temperature,

Jeffrey Shainline (24:36.700)

all of the electrons in that superconducting state

Lex Fridman (24:40.840)

would be in one coherent quantum state.

Jeffrey Shainline (24:42.840)

The wave function of that state is described

Lex Fridman (24:47.520)

in terms of all of the particles simultaneously,

Lex Fridman (24:49.760)

but it extends across macroscopic dimensions,

Lex Fridman (24:52.220)

the size of whatever block of that material

Jeffrey Shainline (24:56.120)

I have sitting here.

Lex Fridman (24:57.040)

And the way this occurs is that,

Jeffrey Shainline (25:01.220)

let's try to be a little bit light on the technical details,

Lex Fridman (25:03.480)

but essentially the electrons coordinate with each other.

Jeffrey Shainline (25:06.520)

They are able to, in this macroscopic quantum state,

Lex Fridman (25:10.300)

they're able to sort of,

Jeffrey Shainline (25:12.260)

one can quickly take the place of the other.

Lex Fridman (25:14.400)

You can't tell electrons apart.

Jeffrey Shainline (25:15.800)

They're what's known as identical particles.

Lex Fridman (25:17.740)

So if this electron runs into a defect

Jeffrey Shainline (25:22.040)

that would otherwise cause it to scatter,

Lex Fridman (25:25.000)

it can just sort of almost miraculously avoid that defect

Jeffrey Shainline (25:30.440)

because it's not really in that location.

Lex Fridman (25:32.240)

It's part of a macroscopic quantum state

Lex Fridman (25:34.000)

and the entire quantum state

Lex Fridman (25:35.660)

was not scattered by that defect.

Lex Fridman (25:37.160)

So you can get a current that flows without dissipation

Lex Fridman (25:40.780)

and that's called a supercurrent.

Jeffrey Shainline (25:42.960)

That's sort of just very much scratching the surface

Lex Fridman (25:47.840)

of superconductivity.

Jeffrey Shainline (25:49.840)

There's very deep and rich physics there,

Lex Fridman (25:52.240)

just probably not the main subject

Jeffrey Shainline (25:54.520)

we need to go into right now.

Lex Fridman (25:55.640)

But it turns out that when you have this material,

Jeffrey Shainline (26:00.520)

you can do usual things like make wires out of it

Lex Fridman (26:03.560)

so you can get current to flow in a straight line on a chip,

Lex Fridman (26:06.440)

but you can also make other devices

Lex Fridman (26:08.920)

that perform different kinds of operations.

Jeffrey Shainline (26:11.880)

Some of them are kind of logic operations

Lex Fridman (26:14.760)

like you'd get in a transistor.

Jeffrey Shainline (26:16.680)

The most common or the most,

Lex Fridman (26:21.200)

I would say, diverse in its utility component

Jeffrey Shainline (26:25.480)

is a Josephson junction.

Lex Fridman (26:26.920)

It's not analogous to a transistor

Jeffrey Shainline (26:28.980)

in the sense that if you apply a voltage here,

Lex Fridman (26:31.480)

it changes how much current flows from left to right,

Lex Fridman (26:33.880)

but it is analogous in sort of a sense

Lex Fridman (26:36.360)

of it's the go to component

Jeffrey Shainline (26:39.200)

that a circuit engineer is going to use

Lex Fridman (26:42.040)

to start to build up more complexity.

Lex Fridman (26:44.520)

So these junctions serve as gates.

Lex Fridman (26:48.840)

They can serve as gates.

Lex Fridman (26:50.680)

So I'm not sure how concerned to be with semantics,

Lex Fridman (26:55.680)

but let me just briefly say what a Josephson junction is

Lex Fridman (26:58.880)

and we can talk about different ways that they can be used.

Lex Fridman (27:02.240)

Basically, if you have a superconducting wire

Lex Fridman (27:05.280)

and then a small gap of a different material

Lex Fridman (27:09.680)

that's not superconducting, an insulator or normal metal,

Lex Fridman (27:13.520)

and then another superconducting wire on the other side,

Lex Fridman (27:15.800)

that's a Josephson junction.

Lex Fridman (27:17.040)

So it's sometimes referred to

Lex Fridman (27:18.600)

as a superconducting weak link.

Lex Fridman (27:20.320)

So you have this superconducting state on one side

Lex Fridman (27:24.200)

and on the other side, and the superconducting wave function

Jeffrey Shainline (27:27.520)

actually tunnels across that gap.

Lex Fridman (27:30.720)

And when you create such a physical entity,

Jeffrey Shainline (27:35.460)

it has very unusual current voltage characteristics.

Lex Fridman (27:41.360)

In that gap, like weird stuff happens.

Jeffrey Shainline (27:44.320)

Through the entire circuit.

Lex Fridman (27:45.160)

So you can imagine, suppose you had a loop set up

Jeffrey Shainline (27:47.600)

that had one of those weak links in the loop.

Lex Fridman (27:51.240)

Current would flow in that loop independent,

Jeffrey Shainline (27:53.800)

even if you hadn't applied a voltage to it,

Lex Fridman (27:55.640)

and that's called the Josephson effect.

Lex Fridman (27:57.060)

So the fact that there's this phase difference

Lex Fridman (28:00.520)

in the quantum wave function from one side

Jeffrey Shainline (28:02.940)

of the tunneling barrier to the other

Lex Fridman (28:04.400)

induces current to flow.

Lex Fridman (28:05.720)

So how does you change state?

Lex Fridman (28:07.720)

Right, exactly.

Lex Fridman (28:08.560)

So how do you change state?

Lex Fridman (28:09.560)

Now picture if I have a current bias coming down

Jeffrey Shainline (28:13.880)

this line of my circuit and there's a Josephson junction

Lex Fridman (28:16.120)

right in the middle of it.

Lex Fridman (28:18.120)

And now I make another wire

Lex Fridman (28:19.960)

that goes around the Josephson junction.

Lex Fridman (28:21.800)

So I have a loop here, a superconducting loop.

Lex Fridman (28:24.600)

I can add current to that loop by exceeding

Jeffrey Shainline (28:28.960)

the critical current of that Josephson junction.

Lex Fridman (28:30.860)

So like any superconducting material,

Jeffrey Shainline (28:34.960)

it can carry this supercurrent that I've described,

Lex Fridman (28:37.700)

this current that can propagate without dissipation

Jeffrey Shainline (28:40.520)

up to a certain level.

Lex Fridman (28:41.840)

And if you try and pass more current than that

Jeffrey Shainline (28:44.240)

through the material, it's going to become

Lex Fridman (28:47.520)

a resistive material, normal material.

Lex Fridman (28:51.140)

So in the Josephson junction, the same thing happens.

Lex Fridman (28:54.140)

I can bias it above its critical current.

Lex Fridman (28:57.120)

And then what it's going to do,

Lex Fridman (28:58.220)

it's going to add a quantized amount of current

Jeffrey Shainline (29:03.560)

into that loop.

Lex Fridman (29:04.400)

And what I mean by quantized is it's going to come

Jeffrey Shainline (29:07.320)

in discrete packets with a well defined value of current.

Lex Fridman (29:11.200)

So in the vernacular of some people working

Jeffrey Shainline (29:15.320)

in this community, you would say you pop a flux on

Lex Fridman (29:19.520)

into the loop.

Lex Fridman (29:20.360)

So a flux on.

Lex Fridman (29:21.760)

You pop a flux on into the loop.

Jeffrey Shainline (29:23.680)

Yeah, so a flux on.

Lex Fridman (29:24.520)

Sounds like skateboarder talk, I love it.

Jeffrey Shainline (29:26.560)

Okay, sorry, go ahead.

Lex Fridman (29:28.960)

A flux on is one of these quantized sort of amounts

Jeffrey Shainline (29:33.640)

of current that you can add to a loop.

Lex Fridman (29:35.220)

And this is a cartoon picture,

Lex Fridman (29:36.600)

but I think it's sufficient for our purposes.

Lex Fridman (29:38.400)

So which, maybe it's useful to say,

Lex Fridman (29:41.700)

what is the speed at which these discrete packets

Lex Fridman (29:45.480)

of current travel?

Jeffrey Shainline (29:47.000)

Because we'll be talking about light a little bit.

Lex Fridman (29:49.160)

It seems like the speed is important.

Jeffrey Shainline (29:51.080)

The speed is important, that's an excellent question.

Lex Fridman (29:53.560)

Sometimes I wonder where you, how you became so astute.

Lex Fridman (29:57.800)

But so this.

Lex Fridman (2:00:00.540)

There's a picture and there's a horse in it, so you're done.

Lex Fridman (2:00:03.940)

But that's not a very interesting problem.

Lex Fridman (2:00:06.500)

I think when I think about intelligence,

Jeffrey Shainline (2:00:09.500)

it's almost defined by the ability to handle problems

Lex Fridman (2:00:13.200)

where you don't know what your inputs are going to be

Lex Fridman (2:00:15.940)

and you don't even necessarily know

Lex Fridman (2:00:17.420)

what you're trying to accomplish.

Jeffrey Shainline (2:00:18.580)

I mean, I'm not sure what I'm trying to accomplish

Lex Fridman (2:00:21.620)

in this world.

Jeffrey Shainline (2:00:22.540)

Yeah, at all scales.

Lex Fridman (2:00:24.540)

Yeah, at all scales, right.

Jeffrey Shainline (2:00:25.900)

I mean, so I'm more drawn to the underlying phenomena,

Lex Fridman (2:00:33.700)

the critical dynamics of this system,

Jeffrey Shainline (2:00:37.940)

trying to understand how elements that you build

Lex Fridman (2:00:41.900)

into your hardware result in emergent fascinating activity

Jeffrey Shainline (2:00:48.580)

that was very difficult to predict, things like that.

Lex Fridman (2:00:51.740)

So, but I gotta be really careful

Jeffrey Shainline (2:00:53.660)

because I think a lot of other people who,

Lex Fridman (2:00:55.580)

if they found themselves working on this project

Jeffrey Shainline (2:00:57.700)

in my shoes, they would say, all right,

Lex Fridman (2:00:59.300)

what are all the different ways we can use this

Lex Fridman (2:01:01.660)

for machine learning?

Lex Fridman (2:01:02.500)

Actually, let me just definitely mention colleague

Jeffrey Shainline (2:01:05.340)

at NIST, Mike Schneider.

Lex Fridman (2:01:06.660)

He's also very much interested,

Jeffrey Shainline (2:01:09.340)

particularly in the superconducting side of things,

Lex Fridman (2:01:11.720)

using the incredible speed, power efficiency,

Jeffrey Shainline (2:01:14.920)

also Ken Seagal at Colgate,

Lex Fridman (2:01:16.620)

other people working on specifically

Jeffrey Shainline (2:01:18.780)

the superconducting side of this for machine learning

Lex Fridman (2:01:22.020)

and deep feed forward neural networks.

Jeffrey Shainline (2:01:25.060)

There, the advantages are obvious.

Lex Fridman (2:01:27.400)

It's extremely fast.

Jeffrey Shainline (2:01:28.780)

Yeah, so that's less on the nature of intelligences

Lex Fridman (2:01:31.880)

and more on various characteristics of this hardware

Jeffrey Shainline (2:01:35.740)

that you can use for the basic computation

Lex Fridman (2:01:38.340)

as we know it today and communication.

Jeffrey Shainline (2:01:40.680)

One of the things that Mike Schneider's working on right now

Lex Fridman (2:01:42.900)

is an image classifier at a relatively small scale.

Jeffrey Shainline (2:01:46.160)

I think he's targeting that nine pixel problem

Lex Fridman (2:01:48.380)

where you can have three different characters

Lex Fridman (2:01:50.100)

and you put in a nine pixel image

Lex Fridman (2:01:54.240)

and you classify it as one of these three categories.

Lex Fridman (2:01:58.420)

And that's gonna be really interesting

Lex Fridman (2:02:00.160)

to see what happens there,

Jeffrey Shainline (2:02:01.260)

because if you can show that even at that scale,

Lex Fridman (2:02:05.720)

you just put these images in and you get it out

Lex Fridman (2:02:08.180)

and he thinks he can do it,

Lex Fridman (2:02:09.540)

I forgot if it's a nanosecond

Jeffrey Shainline (2:02:11.180)

or some extremely fast classification time,

Lex Fridman (2:02:14.040)

it's probably less,

Jeffrey Shainline (2:02:14.880)

it's probably a hundred picoseconds or something.

Lex Fridman (2:02:17.420)

There you have challenges though,

Jeffrey Shainline (2:02:18.820)

because the Joseph's injunctions themselves,

Lex Fridman (2:02:21.600)

the electronic circuit is extremely power efficient.

Jeffrey Shainline (2:02:24.580)

Some orders of magnitude for something more

Lex Fridman (2:02:26.880)

than a transistor doing the same thing,

Lex Fridman (2:02:29.280)

but when you have to cool it down to four Kelvin,

Lex Fridman (2:02:31.480)

you pay a huge overhead just for keeping it cold,

Jeffrey Shainline (2:02:33.900)

even if it's not doing anything.

Lex Fridman (2:02:35.300)

So it has to work at large scale

Jeffrey Shainline (2:02:40.680)

in order to overcome that power penalty,

Lex Fridman (2:02:43.720)

but that's possible.

Jeffrey Shainline (2:02:45.140)

It's just, it's gonna have to get that performance.

Lex Fridman (2:02:48.000)

And this is sort of what you were asking about before

Lex Fridman (2:02:49.720)

is like how much better than silicon would it need to be?

Lex Fridman (2:02:52.760)

And the answer is, I don't know.

Jeffrey Shainline (2:02:54.120)

I think if it's just overall better than silicon

Lex Fridman (2:02:57.220)

at a problem that a lot of people care about,

Jeffrey Shainline (2:03:00.200)

maybe it's image classification,

Lex Fridman (2:03:02.640)

maybe it's facial recognition,

Jeffrey Shainline (2:03:03.960)

maybe it's monitoring credit transactions, I don't know,

Lex Fridman (2:03:07.520)

then I think it will have a place.

Jeffrey Shainline (2:03:09.000)

It's not gonna be in your cell phone,

Lex Fridman (2:03:10.440)

but it could be in your data center.

Lex Fridman (2:03:12.200)

So what about in terms of the data center,

Lex Fridman (2:03:16.160)

I don't know if you're paying attention

Jeffrey Shainline (2:03:17.680)

to the various systems,

Lex Fridman (2:03:19.080)

like Tesla recently announced DOJO,

Jeffrey Shainline (2:03:23.880)

which is a large scale machine learning training system,

Lex Fridman (2:03:27.160)

that again, the bottleneck there

Jeffrey Shainline (2:03:28.940)

is probably going to be communication

Lex Fridman (2:03:30.900)

between those systems.

Jeffrey Shainline (2:03:32.800)

Is there something from your work

Lex Fridman (2:03:34.880)

on everything we've been talking about

Jeffrey Shainline (2:03:38.720)

in terms of superconductive hardware

Lex Fridman (2:03:41.640)

that could be useful there?

Jeffrey Shainline (2:03:43.560)

Oh, I mean, okay, tomorrow, no.

Lex Fridman (2:03:46.720)

In the long term, it could be the whole thing.

Jeffrey Shainline (2:03:49.000)

It could be nothing.

Lex Fridman (2:03:49.840)

I don't know, but definitely, definitely.

Jeffrey Shainline (2:03:54.040)

When you look at the,

Lex Fridman (2:03:55.160)

so I don't know that much about DOJO.

Lex Fridman (2:03:56.720)

My understanding is that that's new, right?

Lex Fridman (2:03:58.840)

That's just coming online.

Jeffrey Shainline (2:04:01.320)

Well, I don't even know where it hasn't come online.

Lex Fridman (2:04:06.960)

And when you announce big, sexy,

Lex Fridman (2:04:09.560)

so let me explain to you the way things work

Lex Fridman (2:04:11.360)

in the world of business and marketing.

Jeffrey Shainline (2:04:15.680)

It's not always clear where you are

Lex Fridman (2:04:18.400)

on the coming online part of that.

Lex Fridman (2:04:20.560)

So I don't know where they are exactly,

Lex Fridman (2:04:22.680)

but the vision is from a ground up

Jeffrey Shainline (2:04:25.240)

to build a very, very large scale,

Lex Fridman (2:04:28.520)

modular machine learning, ASIC,

Jeffrey Shainline (2:04:31.080)

basically hardware that's optimized

Lex Fridman (2:04:32.640)

for training neural networks.

Lex Fridman (2:04:34.120)

And of course, there's a lot of companies

Lex Fridman (2:04:36.080)

that are small and big working on this kind of problem.

Jeffrey Shainline (2:04:39.360)

The question is how to do it in a modular way

Lex Fridman (2:04:42.680)

that has very fast communication.

Jeffrey Shainline (2:04:45.640)

The interesting aspect of Tesla is you have a company

Lex Fridman (2:04:49.520)

that at least at this time is so singularly focused

Jeffrey Shainline (2:04:54.640)

on solving a particular machine learning problem

Lex Fridman (2:04:57.720)

and is making obviously a lot of money doing so

Jeffrey Shainline (2:05:00.880)

because the machine learning problem

Lex Fridman (2:05:02.200)

happens to be involved with autonomous driving.

Lex Fridman (2:05:05.200)

So you have a system that's driven by an application.

Lex Fridman (2:05:09.760)

And that's really interesting because you have maybe Google

Jeffrey Shainline (2:05:15.040)

working on TPUs and so on.

Lex Fridman (2:05:17.720)

You have all these other companies with ASICs.

Jeffrey Shainline (2:05:21.440)

They're usually more kind of always thinking general.

Lex Fridman (2:05:25.960)

So I like it when it's driven by a particular application

Jeffrey Shainline (2:05:29.160)

because then you can really get to the,

Lex Fridman (2:05:32.200)

it's somehow if you just talk broadly about intelligence,

Jeffrey Shainline (2:05:37.240)

you may not always get to the right solutions.

Lex Fridman (2:05:40.240)

It's nice to couple that sometimes

Jeffrey Shainline (2:05:41.560)

with specific clear illustration

Lex Fridman (2:05:45.600)

of something that requires general intelligence,

Jeffrey Shainline (2:05:47.640)

which for me driving is one such case.

Lex Fridman (2:05:49.880)

I think you're exactly right.

Jeffrey Shainline (2:05:51.080)

Sometimes just having that focus on that application

Lex Fridman (2:05:54.360)

brings a lot of people focuses their energy and attention.

Jeffrey Shainline (2:05:57.640)

I think that, so one of the things that's appealing

Lex Fridman (2:06:00.200)

about what you're saying is not just

Jeffrey Shainline (2:06:02.440)

that the application is specific,

Lex Fridman (2:06:03.840)

but also that the scale is big

Lex Fridman (2:06:06.040)

and that the benefit is also huge.

Lex Fridman (2:06:10.640)

Financial and to humanity.

Jeffrey Shainline (2:06:12.280)

Right, right, right.

Lex Fridman (2:06:13.120)

Yeah, so I guess let me just try to understand

Jeffrey Shainline (2:06:15.480)

is the point of this dojo system

Lex Fridman (2:06:17.960)

to figure out the parameters

Jeffrey Shainline (2:06:21.840)

that then plug into neural networks

Lex Fridman (2:06:23.840)

and then you don't need to retrain,

Jeffrey Shainline (2:06:26.520)

you just make copies of a certain chip

Lex Fridman (2:06:29.080)

that has all the other parameters established or?

Jeffrey Shainline (2:06:32.320)

No, it's straight up retraining a large neural network

Lex Fridman (2:06:36.760)

over and over and over.

Lex Fridman (2:06:38.560)

So you have to do it once for every new car?

Lex Fridman (2:06:41.640)

No, no, you have to, so they do this interesting process,

Jeffrey Shainline (2:06:44.800)

which I think is a process for machine learning,

Lex Fridman (2:06:47.000)

supervised machine learning systems

Jeffrey Shainline (2:06:49.200)

you're going to have to do, which is you have a system,

Lex Fridman (2:06:53.720)

you train your network once, it takes a long time.

Jeffrey Shainline (2:06:56.400)

I don't know how long, but maybe a week.

Lex Fridman (2:06:58.720)

Okay. To train.

Lex Fridman (2:07:00.840)

And then you deploy it on, let's say about a million cars.

Lex Fridman (2:07:05.080)

I don't know what the number is.

Lex Fridman (2:07:05.920)

But that part, you just write software

Lex Fridman (2:07:07.840)

that updates some weights in a table and yeah, okay.

Lex Fridman (2:07:10.680)

But there's a loop back.

Lex Fridman (2:07:12.640)

Yeah, yeah, okay.

Jeffrey Shainline (2:07:13.480)

Each of those cars run into trouble, rarely,

Lex Fridman (2:07:18.840)

but they catch the edge cases

Jeffrey Shainline (2:07:23.720)

of the performance of that particular system

Lex Fridman (2:07:26.080)

and then send that data back

Lex Fridman (2:07:28.320)

and either automatically or by humans,

Lex Fridman (2:07:31.440)

that weird edge case data is annotated

Lex Fridman (2:07:34.760)

and then the network has to become smart enough

Lex Fridman (2:07:37.560)

to now be able to perform in those edge cases,

Lex Fridman (2:07:40.120)

so it has to get retrained.

Lex Fridman (2:07:41.800)

There's clever ways of retraining different parts

Jeffrey Shainline (2:07:43.960)

of that network, but for the most part,

Lex Fridman (2:07:46.240)

I think they prefer to retrain the entire thing.

Lex Fridman (2:07:49.280)

So you have this giant monster

Lex Fridman (2:07:51.320)

that kind of has to be retrained regularly.

Jeffrey Shainline (2:07:54.880)

I think the vision with Dojo is to have

Lex Fridman (2:07:58.960)

a very large machine learning focused,

Jeffrey Shainline (2:08:02.320)

driving focused supercomputer

Lex Fridman (2:08:05.200)

that then is sufficiently modular

Jeffrey Shainline (2:08:07.600)

that can be scaled to other machine learning applications.

Lex Fridman (2:08:11.040)

So they're not limiting themselves completely

Jeffrey Shainline (2:08:12.760)

to this particular application,

Lex Fridman (2:08:14.000)

but this application is the way they kind of test

Jeffrey Shainline (2:08:17.440)

this iterative process of machine learning

Lex Fridman (2:08:19.760)

is you make a system that's very dumb,

Jeffrey Shainline (2:08:23.440)

deploy it, get the edge cases where it fails,

Lex Fridman (2:08:27.160)

make it a little smarter, it becomes a little less dumb

Lex Fridman (2:08:30.000)

and that iterative process achieves something

Lex Fridman (2:08:33.000)

that you can call intelligent or is smart enough

Jeffrey Shainline (2:08:36.160)

to be able to solve this particular application.

Lex Fridman (2:08:37.960)

So it has to do with training neural networks fast

Lex Fridman (2:08:43.680)

and training neural networks that are large.

Lex Fridman (2:08:45.920)

But also based on an extraordinary amount of diverse input.

Jeffrey Shainline (2:08:49.920)

Data, yeah.

Lex Fridman (2:08:50.760)

And that's one of the things,

Lex Fridman (2:08:51.920)

so this does seem like one of those spaces

Lex Fridman (2:08:54.520)

where the scale of superconducting optoelectronics,

Jeffrey Shainline (2:08:58.920)

the way that, so when you talk about the weaknesses,

Lex Fridman (2:09:02.520)

like I said, okay, well, you have to cool it down.

Jeffrey Shainline (2:09:04.120)

At this scale, that's fine.

Lex Fridman (2:09:05.760)

Because that's not too much of an added cost.

Jeffrey Shainline (2:09:09.440)

Most of your power is being dissipated

Lex Fridman (2:09:10.960)

by the circuits themselves, not the cooling.

Lex Fridman (2:09:12.920)

And also you have one centralized kind of cognitive hub,

Lex Fridman (2:09:19.400)

if you will.

Lex Fridman (2:09:20.800)

And so if we're talking about putting

Lex Fridman (2:09:24.840)

a superconducting system in a car, that's questionable.

Lex Fridman (2:09:28.640)

Do you really wanna cryostat

Lex Fridman (2:09:30.080)

in the trunk of everyone in your car?

Jeffrey Shainline (2:09:31.400)

It'll fit, it's not that big of a deal,

Lex Fridman (2:09:32.920)

but hopefully there's a better way, right?

Lex Fridman (2:09:35.720)

But since this is sort of a central supreme intelligence

Lex Fridman (2:09:39.080)

or something like that,

Lex Fridman (2:09:40.360)

and it needs to really have this massive data acquisition,

Lex Fridman (2:09:45.120)

massive data integration,

Jeffrey Shainline (2:09:47.080)

I would think that that's where large scale

Lex Fridman (2:09:49.120)

spiking neural networks with vast communication

Lex Fridman (2:09:51.280)

and all these things would have something

Lex Fridman (2:09:53.160)

pretty tremendous to offer.

Jeffrey Shainline (2:09:54.280)

It's not gonna happen tomorrow.

Lex Fridman (2:09:55.760)

There's a lot of development that needs to be done.

Lex Fridman (2:09:58.280)

But we have to be patient with self driving cars

Lex Fridman (2:10:01.440)

for a lot of reasons.

Jeffrey Shainline (2:10:02.280)

We were all optimistic that they would be here by now.

Lex Fridman (2:10:04.600)

And okay, they are to some extent,

Lex Fridman (2:10:06.480)

but if we're thinking five or 10 years down the line,

Lex Fridman (2:10:09.560)

it's not unreasonable.

Jeffrey Shainline (2:10:12.080)

One other thing, let me just mention,

Lex Fridman (2:10:15.200)

getting into self driving cars and technologies

Jeffrey Shainline (2:10:17.400)

that are using AI out in the world,

Lex Fridman (2:10:19.680)

this is something NIST cares a lot about.

Jeffrey Shainline (2:10:21.520)

Elham Tabassi is leading up a much larger effort in AI

Lex Fridman (2:10:25.720)

at NIST than my little project.

Lex Fridman (2:10:29.320)

And really central to that mission

Lex Fridman (2:10:32.680)

is this concept of trustworthiness.

Lex Fridman (2:10:35.000)

So when you're going to deploy this neural network

Lex Fridman (2:10:39.880)

in every single automobile with so much on the line,

Jeffrey Shainline (2:10:43.320)

you have to be able to trust that.

Lex Fridman (2:10:45.240)

So now how do we know that we can trust that?

Lex Fridman (2:10:48.040)

How do we know that we can trust the self driving car

Lex Fridman (2:10:50.080)

or the supercomputer that trained it?

Jeffrey Shainline (2:10:53.720)

There's a lot of work there

Lex Fridman (2:10:54.840)

and there's a lot of that going on at NIST.

Lex Fridman (2:10:56.960)

And it's still early days.

Lex Fridman (2:10:58.200)

I mean, you're familiar with the problem and all that.

Lex Fridman (2:11:01.840)

But there's a fascinating dance in engineering

Lex Fridman (2:11:04.480)

with safety critical systems.

Jeffrey Shainline (2:11:06.680)

There's a desire in computer science,

Lex Fridman (2:11:08.840)

just recently talked to Don Knuth,

Jeffrey Shainline (2:11:13.120)

for algorithms and for systems,

Lex Fridman (2:11:14.800)

for them to be provably correct or provably safe.

Lex Fridman (2:11:17.440)

And this is one other difference

Lex Fridman (2:11:20.320)

between humans and biological systems

Jeffrey Shainline (2:11:22.320)

is we're not provably anything.

Lex Fridman (2:11:24.840)

And so there's some aspect of imperfection

Jeffrey Shainline (2:11:29.760)

that we need to have built in,

Lex Fridman (2:11:32.160)

like robustness to imperfection be part of our systems,

Jeffrey Shainline (2:11:37.920)

which is a difficult thing for engineers to contend with.

Lex Fridman (2:11:40.680)

They're very uncomfortable with the idea

Jeffrey Shainline (2:11:42.800)

that you have to be okay with failure

Lex Fridman (2:11:46.880)

and almost engineer failure into the system.

Jeffrey Shainline (2:11:49.720)

Mathematicians hate it too.

Lex Fridman (2:11:50.960)

But I think it was Turing who said something

Jeffrey Shainline (2:11:53.640)

along the lines of,

Lex Fridman (2:11:55.040)

I can give you an intelligent system

Jeffrey Shainline (2:11:57.080)

or I can give you a flawless system,

Lex Fridman (2:11:59.520)

but I can't give you both.

Lex Fridman (2:12:00.880)

And it's in sort of creativity and abstract thinking

Lex Fridman (2:12:04.120)

seem to rely somewhat on stochasticity

Lex Fridman (2:12:07.840)

and not having components

Lex Fridman (2:12:11.320)

that perform exactly the same way every time.

Jeffrey Shainline (2:12:13.760)

This is where like the disagreement I have with,

Lex Fridman (2:12:16.000)

not disagreement, but a different view on the world.

Jeffrey Shainline (2:12:18.560)

I'm with Turing,

Lex Fridman (2:12:19.680)

but when I talk to robotic, robot colleagues,

Jeffrey Shainline (2:12:24.600)

that sounds like I'm talking to robots,

Lex Fridman (2:12:26.440)

colleagues that are roboticists,

Jeffrey Shainline (2:12:29.760)

the goal is perfection.

Lex Fridman (2:12:31.920)

And to me is like, no,

Jeffrey Shainline (2:12:33.960)

I think the goal should be imperfection

Lex Fridman (2:12:38.880)

that's communicated.

Lex Fridman (2:12:40.960)

And through the interaction between humans and robots,

Lex Fridman (2:12:44.160)

that imperfection becomes a feature, not a bug.

Jeffrey Shainline (2:12:49.600)

Like together, seen as a system,

Lex Fridman (2:12:52.160)

the human and the robot together

Jeffrey Shainline (2:12:53.680)

are better than either of them individually,

Lex Fridman (2:12:56.400)

but the robot itself is not perfect in any way.

Jeffrey Shainline (2:13:00.520)

Of course, there's a bunch of disagreements,

Lex Fridman (2:13:02.760)

including with Mr. Elon about,

Jeffrey Shainline (2:13:06.360)

to me, autonomous driving is fundamentally

Lex Fridman (2:13:08.640)

a human robot interaction problem,

Jeffrey Shainline (2:13:10.760)

not a robotics problem.

Lex Fridman (2:13:12.360)

To Elon, it's a robotics problem.

Jeffrey Shainline (2:13:14.320)

That's actually an open and fascinating question,

Lex Fridman (2:13:18.560)

whether humans can be removed from the loop completely.

Jeffrey Shainline (2:13:24.400)

We've talked about a lot of fascinating chemistry

Lex Fridman (2:13:27.680)

and physics and engineering,

Lex Fridman (2:13:31.240)

and we're always running up against this issue

Lex Fridman (2:13:33.680)

that nature seems to dictate what's easy and what's hard.

Lex Fridman (2:13:37.560)

So you have this cool little paper

Lex Fridman (2:13:40.080)

that I'd love to just ask you about.

Jeffrey Shainline (2:13:43.680)

It's titled,

Lex Fridman (2:13:44.520)

Does Cosmological Evolution Select for Technology?

Lex Fridman (2:13:48.200)

So in physics, there's parameters

Lex Fridman (2:13:53.240)

that seem to define the way our universe works,

Jeffrey Shainline (2:13:56.200)

that physics works, that if it worked any differently,

Lex Fridman (2:13:59.320)

we would get a very different world.

Lex Fridman (2:14:01.720)

So it seems like the parameters are very fine tuned

Lex Fridman (2:14:04.240)

to the kind of physics that we see.

Jeffrey Shainline (2:14:06.480)

All the beautiful E equals MC squared,

Lex Fridman (2:14:08.560)

they would get these nice, beautiful laws.

Jeffrey Shainline (2:14:10.440)

It seems like very fine tuned for that.

Lex Fridman (2:14:13.160)

So what you argue in this article

Jeffrey Shainline (2:14:15.960)

is it may be that the universe has also fine tuned

Lex Fridman (2:14:20.600)

its parameters that enable the kind of technological

Jeffrey Shainline (2:14:25.400)

innovation that we see, the technology that we see.

Lex Fridman (2:14:29.600)

Can you explain this idea?

Jeffrey Shainline (2:14:31.520)

Yeah, I think you've introduced it nicely.

Lex Fridman (2:14:33.440)

Let me just try to say a few things in my language layout.

Lex Fridman (2:14:39.560)

What is this fine tuning problem?

Lex Fridman (2:14:41.680)

So physicists have spent centuries trying to understand

Jeffrey Shainline (2:14:46.560)

the system of equations that govern the way nature behaves,

Lex Fridman (2:14:51.640)

the way particles move and interact with each other.

Lex Fridman (2:14:55.120)

And as that understanding has become more clear over time,

Lex Fridman (2:15:00.120)

it became sort of evident that it's all well adjusted

Jeffrey Shainline (2:15:07.640)

to allow a universe like we see, very complex,

Lex Fridman (2:15:13.040)

this large, long lived universe.

Lex Fridman (2:15:16.480)

And so one answer to that is, well, of course it is

Lex Fridman (2:15:19.920)

because we wouldn't be here otherwise.

Lex Fridman (2:15:21.520)

But I don't know, that's not very satisfying.

Lex Fridman (2:15:24.400)

That's sort of, that's what's known

Jeffrey Shainline (2:15:25.560)

as the weak anthropic principle.

Lex Fridman (2:15:27.240)

It's a statement of selection bias.

Jeffrey Shainline (2:15:29.200)

We can only observe a universe that is fit for us to live in.

Lex Fridman (2:15:33.640)

So what does it mean for a universe

Lex Fridman (2:15:34.960)

to be fit for us to live in?

Lex Fridman (2:15:36.120)

Well, the pursuit of physics,

Jeffrey Shainline (2:15:38.400)

it is based partially on coming up with equations

Lex Fridman (2:15:42.600)

that describe how things behave

Lex Fridman (2:15:44.640)

and interact with each other.

Lex Fridman (2:15:46.280)

But in all those equations you have,

Lex Fridman (2:15:48.480)

so there's the form of the equation,

Lex Fridman (2:15:49.960)

sort of how different fields or particles

Jeffrey Shainline (2:15:54.200)

move in space and time.

Lex Fridman (2:15:56.480)

But then there are also the parameters

Jeffrey Shainline (2:15:58.480)

that just tell you sort of the strength

Lex Fridman (2:16:01.120)

of different couplings.

Lex Fridman (2:16:02.840)

How strongly does a charged particle

Lex Fridman (2:16:05.160)

couple to the electromagnetic field or masses?

Lex Fridman (2:16:07.600)

How strongly does a particle couple

Lex Fridman (2:16:10.760)

to the Higgs field or something like that?

Lex Fridman (2:16:12.960)

And those parameters that define,

Lex Fridman (2:16:16.960)

not the general structure of the equations,

Lex Fridman (2:16:19.760)

but the relative importance of different terms,

Lex Fridman (2:16:23.600)

they seem to be every bit as important

Jeffrey Shainline (2:16:25.240)

as the structure of the equations themselves.

Lex Fridman (2:16:27.760)

And so I forget who it was.

Jeffrey Shainline (2:16:29.400)

Somebody, when they were working through this

Lex Fridman (2:16:31.200)

and trying to see, okay, if I adjust the parameter,

Jeffrey Shainline (2:16:34.000)

this parameter over here,

Lex Fridman (2:16:34.840)

call it the, say the fine structure constant,

Jeffrey Shainline (2:16:36.800)

which tells us the strength

Lex Fridman (2:16:37.920)

of the electromagnetic interaction.

Jeffrey Shainline (2:16:40.400)

Oh boy, I can't change it very much.

Lex Fridman (2:16:42.320)

Otherwise nothing works.

Jeffrey Shainline (2:16:43.720)

The universe sort of doesn't,

Lex Fridman (2:16:45.360)

it just pops into existence and goes away

Jeffrey Shainline (2:16:47.240)

in a nanosecond or something like that.

Lex Fridman (2:16:48.920)

And somebody had the phrase,

Jeffrey Shainline (2:16:51.040)

this looks like a put up job,

Lex Fridman (2:16:52.920)

meaning every one of these parameters was dialed in.

Jeffrey Shainline (2:16:57.080)

It's arguable how precisely they have to be dialed in,

Lex Fridman (2:17:00.680)

but dialed in to some extent,

Jeffrey Shainline (2:17:03.040)

not just in order to enable our existence,

Lex Fridman (2:17:05.360)

that's a very anthropocentric view,

Lex Fridman (2:17:07.120)

but to enable a universe like this one.

Lex Fridman (2:17:10.000)

So, okay, maybe I think the majority position

Jeffrey Shainline (2:17:14.040)

of working physicists in the field is,

Lex Fridman (2:17:17.000)

it has to be that way in order for us to exist.

Jeffrey Shainline (2:17:18.960)

We're here, we shouldn't be surprised

Lex Fridman (2:17:20.400)

that that's the way the universe is.

Lex Fridman (2:17:22.800)

And I don't know, for a while,

Lex Fridman (2:17:24.520)

that never sat well with me,

Lex Fridman (2:17:26.120)

but I just kind of moved on

Lex Fridman (2:17:28.040)

because there are things to do

Lex Fridman (2:17:29.440)

and a lot of exciting work.

Lex Fridman (2:17:31.160)

It doesn't depend on resolving this puzzle,

Lex Fridman (2:17:33.760)

but as I started working more with technology,

Lex Fridman (2:17:39.320)

getting into the more recent years of my career,

Jeffrey Shainline (2:17:41.840)

particularly when I started,

Lex Fridman (2:17:43.720)

after having worked with silicon for a long time,

Jeffrey Shainline (2:17:46.520)

which was kind of eerie on its own,

Lex Fridman (2:17:49.000)

but then when I switched over to superconductors,

Jeffrey Shainline (2:17:51.120)

I was just like, this is crazy.

Lex Fridman (2:17:53.560)

It's just absolutely astonishing

Jeffrey Shainline (2:17:57.360)

that our universe gives us superconductivity.

Lex Fridman (2:18:00.560)

It's one of the most beautiful physical phenomena

Lex Fridman (2:18:02.440)

and it's also extraordinarily useful for technology.

Lex Fridman (2:18:06.560)

So you can argue that the universe

Jeffrey Shainline (2:18:07.920)

has to have the parameters it does for us to exist

Lex Fridman (2:18:11.280)

because we couldn't be here otherwise,

Lex Fridman (2:18:13.000)

but why does it give us technology?

Lex Fridman (2:18:14.760)

Why does it give us silicon that has this ideal oxide

Jeffrey Shainline (2:18:18.840)

that allows us to make a transistor

Lex Fridman (2:18:20.920)

without trying that hard?

Jeffrey Shainline (2:18:23.640)

That can't be explained by the same anthropic reasoning.

Lex Fridman (2:18:27.800)

Yeah, so it's asking the why question.

Jeffrey Shainline (2:18:30.360)

I mean, a slight natural extension of that question is,

Lex Fridman (2:18:34.680)

I wonder if the parameters were different

Jeffrey Shainline (2:18:39.440)

if we would simply have just another set of paint brushes

Lex Fridman (2:18:44.240)

to create totally other things

Jeffrey Shainline (2:18:46.880)

that wouldn't look like anything

Lex Fridman (2:18:49.240)

like the technology of today,

Lex Fridman (2:18:50.880)

but would nevertheless have incredible complexity,

Lex Fridman (2:18:54.520)

which is if you sort of zoom out and start defining things,

Jeffrey Shainline (2:18:57.160)

not by like how many batteries it needs

Lex Fridman (2:19:01.400)

and whether it can make toast,

Lex Fridman (2:19:03.560)

but more like how much complexity is within the system

Lex Fridman (2:19:06.440)

or something like that.

Jeffrey Shainline (2:19:07.280)

Well, yeah, you can start to quantify things.

Lex Fridman (2:19:10.000)

You're exactly right.

Lex Fridman (2:19:10.840)

So nowhere am I arguing that

Lex Fridman (2:19:13.720)

in all of the vast parameter space

Jeffrey Shainline (2:19:15.840)

of everything that could conceivably exist

Lex Fridman (2:19:18.000)

in the multiverse of nature,

Jeffrey Shainline (2:19:20.640)

there's this one point in parameter space

Lex Fridman (2:19:23.280)

where complexity arises.

Jeffrey Shainline (2:19:25.160)

I doubt it.

Lex Fridman (2:19:26.640)

That would be a shameful waste of resources, it seems.

Lex Fridman (2:19:31.120)

But it might be that we reside

Lex Fridman (2:19:33.880)

at one place in parameter space

Jeffrey Shainline (2:19:35.640)

that has been adapted through an evolutionary process

Lex Fridman (2:19:40.040)

to allow us to make certain technologies

Jeffrey Shainline (2:19:43.440)

that allow our particular kind of universe to arise

Lex Fridman (2:19:47.080)

and sort of achieve the things it does.

Jeffrey Shainline (2:19:49.760)

See, I wonder if nature in this kind of discussion,

Lex Fridman (2:19:52.720)

if nature is a catalyst for innovation

Jeffrey Shainline (2:19:55.720)

or if it's a ceiling for innovation.

Lex Fridman (2:19:57.680)

So like, is it going to always limit us?

Jeffrey Shainline (2:20:00.800)

Like you're talking about silicon.

Lex Fridman (2:20:04.000)

Is it just make it super easy to do awesome stuff

Jeffrey Shainline (2:20:06.640)

in a certain dimension,

Lex Fridman (2:20:08.000)

but we could still do awesome stuff in other ways,

Lex Fridman (2:20:10.240)

it'll just be harder?

Lex Fridman (2:20:11.560)

Or does it really set like the maximum we can do?

Jeffrey Shainline (2:20:15.440)

That's a good thing to,

Lex Fridman (2:20:17.840)

that's a good subject to discuss.

Jeffrey Shainline (2:20:19.400)

I guess I feel like we need to lay

Lex Fridman (2:20:20.960)

a little bit more groundwork.

Lex Fridman (2:20:23.160)

So I want to make sure that

Lex Fridman (2:20:27.560)

I introduce this in the context

Jeffrey Shainline (2:20:29.240)

of Lee Smolin's previous idea.

Lex Fridman (2:20:31.800)

So who's Lee Smolin and what kind of ideas does he have?

Jeffrey Shainline (2:20:35.640)

Okay, Lee Smolin is a theoretical physicist

Lex Fridman (2:20:39.000)

who back in the late 1980s published a paper

Jeffrey Shainline (2:20:42.440)

in the early 1990s introduced this idea

Lex Fridman (2:20:45.040)

of cosmological natural selection,

Jeffrey Shainline (2:20:47.000)

which argues that the universe did evolve.

Lex Fridman (2:20:51.440)

So his paper was called, did the universe evolve?

Lex Fridman (2:20:54.480)

And I gave myself the liberty of titling my paper

Lex Fridman (2:20:59.480)

does cosmological selection

Jeffrey Shainline (2:21:01.440)

or does cosmological evolution select for technology

Lex Fridman (2:21:03.960)

in reference to that.

Lex Fridman (2:21:05.000)

So he introduced that idea decades ago.

Lex Fridman (2:21:08.200)

Now he primarily works on quantum gravity,

Jeffrey Shainline (2:21:12.200)

loop quantum gravity, other approaches to

Lex Fridman (2:21:14.640)

unifying quantum mechanics with general relativity,

Jeffrey Shainline (2:21:19.280)

as you can read about in his most recent book, I believe,

Lex Fridman (2:21:22.360)

and he's been on your show as well.

Jeffrey Shainline (2:21:24.280)

So, but I want to introduce this idea

Lex Fridman (2:21:27.760)

of cosmological natural selection,

Jeffrey Shainline (2:21:29.360)

because I think that is one of the core ideas

Lex Fridman (2:21:32.640)

that could change our understanding

Jeffrey Shainline (2:21:35.320)

of how the universe got here, our role in it,

Lex Fridman (2:21:37.800)

what technology is doing here.

Lex Fridman (2:21:39.840)

But there's a couple more pieces

Lex Fridman (2:21:41.240)

that need to be set up first.

Lex Fridman (2:21:42.360)

So the beginning of our universe is largely accepted

Lex Fridman (2:21:46.320)

to be the big bang.

Lex Fridman (2:21:47.360)

And what that means is if you look back in time

Lex Fridman (2:21:49.920)

by looking far away in space,

Jeffrey Shainline (2:21:52.640)

you see that everything used to be at one point

Lex Fridman (2:21:56.960)

and it expanded away from there.

Jeffrey Shainline (2:21:58.920)

There was an era in the evolutionary process of our universe

Lex Fridman (2:22:04.040)

that was called inflation.

Lex Fridman (2:22:05.520)

And this idea was developed primarily by Alan Guth

Lex Fridman (2:22:08.880)

and others, Andre Linde and others in the 80s.

Lex Fridman (2:22:13.120)

And this idea of inflation is basically that

Lex Fridman (2:22:16.040)

when a singularity begins this process of growth,

Jeffrey Shainline (2:22:25.240)

there can be a temporary stage

Lex Fridman (2:22:27.560)

where it just accelerates incredibly rapidly.

Lex Fridman (2:22:30.880)

And based on quantum field theory,

Lex Fridman (2:22:33.760)

this tells us that this should produce matter

Jeffrey Shainline (2:22:35.720)

in precisely the proportions that we find

Lex Fridman (2:22:37.840)

of hydrogen and helium in the big bang,

Jeffrey Shainline (2:22:39.960)

lithium too, lithium also, and other things too.

Lex Fridman (2:22:44.800)

So the predictions that come out of big bang

Jeffrey Shainline (2:22:47.120)

inflationary cosmology have stood up extremely well

Lex Fridman (2:22:50.720)

to empirical verification,

Jeffrey Shainline (2:22:52.520)

the cosmic microwave background, things like this.

Lex Fridman (2:22:55.720)

So most scientists working in the field

Jeffrey Shainline (2:22:59.520)

think that the origin of our universe is the big bang.

Lex Fridman (2:23:03.720)

And I base all my thinking on that as well.

Jeffrey Shainline (2:23:08.040)

I'm just laying this out there so that people understand

Lex Fridman (2:23:11.440)

that where I'm coming from is an extension,

Jeffrey Shainline (2:23:14.160)

not a replacement of existing well founded ideas.

Lex Fridman (2:23:19.080)

In a paper, I believe it was 1986 with Alan Guth

Lex Fridman (2:23:23.800)

and another author Farhi,

Lex Fridman (2:23:26.560)

they wrote that a big bang,

Jeffrey Shainline (2:23:30.240)

I don't remember the exact quote,

Lex Fridman (2:23:31.520)

a big bang is inextricably linked with a black hole.

Jeffrey Shainline (2:23:35.280)

The singularity that we call our origin

Lex Fridman (2:23:39.240)

is mathematically indistinguishable from a black hole.

Jeffrey Shainline (2:23:42.000)

They're the same thing.

Lex Fridman (2:23:44.360)

And Lee Smolin based his thinking on that idea,

Jeffrey Shainline (2:23:48.880)

I believe, I don't mean to speak for him,

Lex Fridman (2:23:50.720)

but this is my reading of it.

Lex Fridman (2:23:52.080)

So what Lee Smolin will say is that

Lex Fridman (2:23:56.040)

a black hole in one universe is a big bang

Jeffrey Shainline (2:23:58.600)

in another universe.

Lex Fridman (2:24:00.720)

And this allows us to have progeny, offspring.

Lex Fridman (2:24:04.680)

So a universe can be said to have come

Lex Fridman (2:24:08.400)

before another universe.

Lex Fridman (2:24:10.480)

And very crucially, Smolin argues,

Lex Fridman (2:24:14.200)

I think this is potentially one of the great ideas

Jeffrey Shainline (2:24:16.920)

of all time, that's my opinion,

Lex Fridman (2:24:18.640)

that when a black hole forms, it's not a classical entity,

Jeffrey Shainline (2:24:22.400)

it's a quantum gravitational entity.

Lex Fridman (2:24:24.520)

So it is subject to the fluctuations

Jeffrey Shainline (2:24:27.240)

that are inherent in quantum mechanics, the properties,

Lex Fridman (2:24:34.080)

what we're calling the parameters

Jeffrey Shainline (2:24:35.440)

that describe the physics of that system

Lex Fridman (2:24:38.600)

are subject to slight mutations

Lex Fridman (2:24:40.680)

so that the offspring universe

Lex Fridman (2:24:42.600)

does not have the exact same parameters

Jeffrey Shainline (2:24:45.200)

defining its physics as its parent universe.

Lex Fridman (2:24:48.440)

They're close, but they're a little bit different.

Lex Fridman (2:24:50.440)

And so now you have a mechanism for evolution,

Lex Fridman (2:24:55.160)

for natural selection.

Lex Fridman (2:24:57.280)

So there's mutation, so there's,

Lex Fridman (2:24:59.680)

and then if you think about the DNA of the universe

Jeffrey Shainline (2:25:03.280)

are the basic parameters that govern its laws.

Lex Fridman (2:25:05.880)

Exactly, so what Smolin said is our universe results

Jeffrey Shainline (2:25:11.560)

from an evolutionary process that can be traced back

Lex Fridman (2:25:14.520)

some, he estimated, 200 million generations.

Jeffrey Shainline (2:25:17.640)

Initially, there was something like a vacuum fluctuation

Lex Fridman (2:25:20.640)

that produced through random chance a universe

Jeffrey Shainline (2:25:25.800)

that was able to reproduce just one.

Lex Fridman (2:25:27.280)

So now it had one offspring.

Lex Fridman (2:25:28.640)

And then over time, it was able to make more and more

Lex Fridman (2:25:30.960)

until it evolved into a highly structured universe

Jeffrey Shainline (2:25:35.280)

with a very long lifetime, with a great deal of complexity

Lex Fridman (2:25:40.160)

and importantly, especially importantly for Lee Smolin,

Jeffrey Shainline (2:25:44.080)

stars, stars make black holes.

Lex Fridman (2:25:47.160)

Therefore, we should expect our universe

Jeffrey Shainline (2:25:49.720)

to be optimized, have its physical parameters optimized

Lex Fridman (2:25:53.160)

to make very large numbers of stars

Jeffrey Shainline (2:25:55.600)

because that's how you make black holes

Lex Fridman (2:25:57.840)

and black holes make offspring.

Lex Fridman (2:25:59.440)

So we expect the physics of our universe to have evolved

Lex Fridman (2:26:03.720)

to maximize fecundity, the number of offspring.

Lex Fridman (2:26:06.720)

And the way Lee Smolin argues you do that

Lex Fridman (2:26:09.200)

is through stars that the biggest ones die

Jeffrey Shainline (2:26:12.160)

in these core collapse supernova

Lex Fridman (2:26:13.440)

that make a black hole and a child.

Jeffrey Shainline (2:26:15.720)

Okay, first of all, I agree with you

Lex Fridman (2:26:19.200)

that this is back to our fractal view of everything

Jeffrey Shainline (2:26:24.760)

from intelligence to our universe.

Lex Fridman (2:26:27.240)

That is very compelling and a very powerful idea

Jeffrey Shainline (2:26:31.120)

that unites the origin of life

Lex Fridman (2:26:36.120)

and perhaps the origin of ideas and intelligence.

Lex Fridman (2:26:39.760)

So from a Dawkins perspective here on earth,

Lex Fridman (2:26:42.200)

the evolution of those and then the evolution

Jeffrey Shainline (2:26:45.360)

of the laws of physics that led to us.

Lex Fridman (2:26:51.000)

I mean, it's beautiful.

Lex Fridman (2:26:52.280)

And then you stacking on top of that,

Lex Fridman (2:26:54.840)

that maybe we are one of the offspring.

Jeffrey Shainline (2:26:57.480)

Right, okay, so before getting into where I'd like

Lex Fridman (2:27:02.320)

to take that idea, let me just a little bit more groundwork.

Jeffrey Shainline (2:27:05.160)

There is this concept of the multiverse

Lex Fridman (2:27:07.080)

and it can be confusing.

Jeffrey Shainline (2:27:08.600)

Different people use the word multiverse in different ways.

Lex Fridman (2:27:11.760)

In the multiverse that I think is relevant to picture

Jeffrey Shainline (2:27:16.920)

when trying to grasp Lee Smolin's idea,

Lex Fridman (2:27:20.840)

essentially every vacuum fluctuation

Jeffrey Shainline (2:27:24.280)

can be referred to as a universe.

Lex Fridman (2:27:25.920)

It occurs, it borrows energy from the vacuum

Jeffrey Shainline (2:27:28.840)

for some finite amount of time

Lex Fridman (2:27:30.200)

and it evanesces back into the quantum vacuum.

Lex Fridman (2:27:33.960)

And ideas of Guth before that and Andrei Linde

Lex Fridman (2:27:38.960)

with eternal inflation aren't that different

Jeffrey Shainline (2:27:42.480)

that you would expect nature

Lex Fridman (2:27:44.480)

due to the quantum properties of the vacuum,

Jeffrey Shainline (2:27:46.920)

which we know exist, they're measurable

Lex Fridman (2:27:49.480)

through things like the Casimir effect and others.

Jeffrey Shainline (2:27:52.200)

You know that there are these fluctuations

Lex Fridman (2:27:54.200)

that are occurring.

Lex Fridman (2:27:55.120)

What Smolin is arguing is that there is

Lex Fridman (2:27:58.080)

this extensive multiverse, that this universe,

Lex Fridman (2:28:01.400)

what we can measure and interact with

Lex Fridman (2:28:04.520)

is not unique in nature.

Jeffrey Shainline (2:28:07.040)

It's just our residents, it's where we reside.

Lex Fridman (2:28:10.800)

And there are countless, potentially infinity

Jeffrey Shainline (2:28:13.640)

other universes, other entire evolutionary trajectories

Lex Fridman (2:28:17.280)

that have evolved into things like

Lex Fridman (2:28:19.360)

what you were mentioning a second ago

Lex Fridman (2:28:21.000)

with different parameters and different ways

Jeffrey Shainline (2:28:24.080)

of achieving complexity and reproduction

Lex Fridman (2:28:26.000)

and all that stuff.

Lex Fridman (2:28:27.040)

So it's not that the evolutionary process

Lex Fridman (2:28:30.480)

is a funnel towards this end point, not at all.

Jeffrey Shainline (2:28:34.240)

Just like the biological evolutionary process

Lex Fridman (2:28:37.000)

that has occurred within our universe

Jeffrey Shainline (2:28:39.320)

is not a unique route toward achieving

Lex Fridman (2:28:42.800)

one specific chosen kind of species.

Jeffrey Shainline (2:28:45.000)

No, we have extraordinary diversity around us.

Lex Fridman (2:28:49.160)

That's what evolution does.

Lex Fridman (2:28:50.520)

And for any one species like us,

Lex Fridman (2:28:52.200)

you might feel like we're at the center of this process.

Jeffrey Shainline (2:28:54.840)

We're the destination of this process,

Lex Fridman (2:28:57.160)

but we're just one of the many

Jeffrey Shainline (2:28:59.480)

nearly infinite branches of this process.

Lex Fridman (2:29:02.240)

And I suspect it is exactly infinite.

Jeffrey Shainline (2:29:04.240)

I mean, I just can't understand how with this idea,

Lex Fridman (2:29:09.080)

you can ever draw a boundary around it and say,

Jeffrey Shainline (2:29:11.040)

no, the universe, I mean, the multiverse

Lex Fridman (2:29:13.480)

has 10 to the one quadrillion components,

Lex Fridman (2:29:17.720)

but not infinity.

Lex Fridman (2:29:18.880)

I don't know that.

Jeffrey Shainline (2:29:20.200)

Well, yeah, I have cognitively incapable

Lex Fridman (2:29:24.080)

as I think all of us are

Lex Fridman (2:29:25.680)

and truly understanding the concept of infinity.

Lex Fridman (2:29:29.080)

And the concept of nothing as well.

Lex Fridman (2:29:31.000)

And nothing, but also the concept of a lot

Lex Fridman (2:29:34.320)

is pretty difficult.

Jeffrey Shainline (2:29:35.360)

I can just, I can count.

Lex Fridman (2:29:37.880)

I run out of fingers at a certain point

Lex Fridman (2:29:39.920)

and then you're screwed.

Lex Fridman (2:29:40.760)

And when you're wearing shoes

Lex Fridman (2:29:41.760)

and you can't even get down to your toes, it's like.

Lex Fridman (2:29:44.400)

It's like, all right, a thousand fine, a million.

Lex Fridman (2:29:47.040)

Is that what?

Lex Fridman (2:29:48.040)

And then it gets crazier and crazier.

Jeffrey Shainline (2:29:50.040)

Right, right.

Lex Fridman (2:29:51.720)

So this particular, so when we say technology, by the way,

Jeffrey Shainline (2:29:55.720)

I mean, there's some, not to over romanticize the thing,

Lex Fridman (2:30:00.640)

but there is some aspect about this branch of ours

Jeffrey Shainline (2:30:04.160)

that allows us to, for the universe to know itself.

Lex Fridman (2:30:08.040)

Yes, yes.

Lex Fridman (2:30:08.960)

So to be, to have like little conscious cognitive fingers

Lex Fridman (2:30:15.080)

that are able to feel like to scratch the head.

Jeffrey Shainline (2:30:18.320)

Right, right, right.

Lex Fridman (2:30:19.640)

To be able to construct E equals MC squared

Lex Fridman (2:30:22.040)

and to introspect, to start to gain some understanding

Lex Fridman (2:30:25.640)

of the laws that govern it.

Lex Fridman (2:30:27.400)

Isn't that, isn't that kind of amazing?

Lex Fridman (2:30:32.000)

Okay, I'm just human, but it feels like that,

Jeffrey Shainline (2:30:35.600)

if I were to build a system that does this kind of thing,

Lex Fridman (2:30:39.200)

that evolves laws of physics, that evolves life,

Jeffrey Shainline (2:30:42.080)

that evolves intelligence, that my goal would be

Lex Fridman (2:30:45.440)

to come up with things that are able to think about itself.

Jeffrey Shainline (2:30:48.960)

Right, aren't we kind of close to the design specs,

Lex Fridman (2:30:53.360)

the destination?

Jeffrey Shainline (2:30:54.680)

We're pretty close, I don't know.

Lex Fridman (2:30:56.040)

I mean, I'm spending my career designing things

Jeffrey Shainline (2:30:58.360)

that I hope will think about themselves,

Lex Fridman (2:30:59.800)

so you and I aren't too far apart on that one.

Lex Fridman (2:31:02.680)

But then maybe that problem is a lot harder

Lex Fridman (2:31:05.520)

than we imagine.

Jeffrey Shainline (2:31:06.360)

Maybe we need to.

Lex Fridman (2:31:07.800)

Let's not get, let's not get too far

Jeffrey Shainline (2:31:09.480)

because I want to emphasize something that,

Lex Fridman (2:31:12.040)

what you're saying is, isn't it fascinating

Jeffrey Shainline (2:31:14.600)

that the universe evolved something

Lex Fridman (2:31:16.880)

that can be conscious, reflect on itself?

Lex Fridman (2:31:19.880)

But Lee Smolin's idea didn't take us there, remember?

Lex Fridman (2:31:23.760)

It took us to stars.

Jeffrey Shainline (2:31:25.720)

Lee Smolin has argued, I think,

Lex Fridman (2:31:29.160)

right on almost every single way

Jeffrey Shainline (2:31:32.080)

that cosmological natural selection

Lex Fridman (2:31:35.760)

could lead to a universe with rich structure.

Lex Fridman (2:31:38.720)

And he argued that the structure,

Lex Fridman (2:31:41.120)

the physics of our universe is designed

Jeffrey Shainline (2:31:43.160)

to make a lot of stars so that they can make black holes.

Lex Fridman (2:31:46.080)

But that doesn't explain what we're doing here.

Jeffrey Shainline (2:31:48.200)

In order for that to be an explanation of us,

Lex Fridman (2:31:51.320)

what you have to assume is that once you made that universe

Jeffrey Shainline (2:31:55.360)

that was capable of producing stars,

Lex Fridman (2:31:58.040)

life, planets, all these other things,

Jeffrey Shainline (2:32:00.280)

we're along for the ride.

Lex Fridman (2:32:01.280)

They got lucky.

Jeffrey Shainline (2:32:02.160)

We're kind of arising, growing up in the cracks,

Lex Fridman (2:32:05.560)

but the universe isn't here for us.

Jeffrey Shainline (2:32:06.880)

We're still kind of a fluke in that picture.

Lex Fridman (2:32:09.200)

And I can't, I don't necessarily have

Jeffrey Shainline (2:32:12.400)

like a philosophical opposition to that stance.

Lex Fridman (2:32:14.840)

It's just not, okay, so I don't think it's complete.

Lex Fridman (2:32:20.200)

So it seems like whatever we got going on here to you,

Lex Fridman (2:32:22.960)

it seems like whatever we have here on earth

Jeffrey Shainline (2:32:25.640)

seems like a thing you might want to select for

Lex Fridman (2:32:28.440)

in this whole big process.

Jeffrey Shainline (2:32:29.680)

Exactly.

Lex Fridman (2:32:30.520)

So if what you are truly,

Jeffrey Shainline (2:32:32.080)

if your entire evolutionary process

Lex Fridman (2:32:34.840)

only cares about fecundity,

Jeffrey Shainline (2:32:36.760)

it only cares about making offspring universes

Lex Fridman (2:32:39.760)

because then there's gonna be the most of them

Jeffrey Shainline (2:32:41.720)

in that local region of hyperspace,

Lex Fridman (2:32:45.120)

which is the set of all possible universes, let's say.

Jeffrey Shainline (2:32:50.240)

You don't care how those universes are made.

Lex Fridman (2:32:52.800)

You know they have to be made by black holes.

Jeffrey Shainline (2:32:54.840)

This is what inflationary theory tells us.

Lex Fridman (2:32:57.920)

The big bang tells us that black holes make universes.

Lex Fridman (2:33:02.040)

But what if there was a technological means

Lex Fridman (2:33:04.200)

to make universes?

Jeffrey Shainline (2:33:05.920)

Stars require a ton of matter

Lex Fridman (2:33:09.280)

because they're not thinking very carefully

Jeffrey Shainline (2:33:11.640)

about how you make a black hole.

Lex Fridman (2:33:12.800)

They're just using gravity, you know?

Lex Fridman (2:33:16.040)

But if we devise technologies

Lex Fridman (2:33:19.160)

that can efficiently compress matter into a singularity,

Jeffrey Shainline (2:33:23.280)

it turns out that if you can compress about 10 kilograms

Lex Fridman (2:33:26.200)

into a very small volume,

Jeffrey Shainline (2:33:28.560)

that will make a black hole

Lex Fridman (2:33:29.720)

that is likely highly probable to inflate

Jeffrey Shainline (2:33:32.360)

into its own offspring universe.

Lex Fridman (2:33:34.720)

This is according to calculations done by other people

Jeffrey Shainline (2:33:37.080)

who are professional quantum theorists,

Lex Fridman (2:33:38.720)

quantum field theorists,

Lex Fridman (2:33:40.160)

and I hope I am grasping what they're telling me correctly.

Lex Fridman (2:33:44.480)

I am somewhat of a translator here.

Lex Fridman (2:33:47.520)

But so that's the position

Lex Fridman (2:33:50.400)

that is particularly intriguing to me,

Jeffrey Shainline (2:33:52.680)

which is that what might have happened is that,

Lex Fridman (2:33:56.240)

okay, this particular branch on the vast tree of evolution,

Jeffrey Shainline (2:34:01.120)

cosmological evolution that we're talking about,

Lex Fridman (2:34:03.200)

not biological evolution within our universe,

Lex Fridman (2:34:05.680)

but cosmological evolution,

Lex Fridman (2:34:07.760)

went through exactly the process

Jeffrey Shainline (2:34:09.560)

that Elise Mullen described,

Lex Fridman (2:34:10.920)

got to the stage where stars were making lots of black holes

Lex Fridman (2:34:15.800)

but then continued to evolve and somehow bridged that gap

Lex Fridman (2:34:19.480)

and made intelligence and intelligence

Jeffrey Shainline (2:34:22.040)

capable of devising technologies

Lex Fridman (2:34:24.080)

because technologies, intelligent species

Jeffrey Shainline (2:34:27.840)

working in conjunction with technologies

Lex Fridman (2:34:29.640)

could then produce even more.

Jeffrey Shainline (2:34:32.080)

Yeah, more efficiently, more faster and better

Lex Fridman (2:34:35.800)

and more different.

Jeffrey Shainline (2:34:36.840)

Then you start to have different kind of mechanisms

Lex Fridman (2:34:38.800)

and mutation perhaps, all that kind of stuff.

Lex Fridman (2:34:40.800)

And so if you do a simple calculation that says,

Lex Fridman (2:34:43.080)

all right, if I want to,

Jeffrey Shainline (2:34:44.960)

we know roughly how many core collapse supernovae

Lex Fridman (2:34:50.920)

have resulted in black holes in our galaxy

Jeffrey Shainline (2:34:54.440)

since the beginning of the universe

Lex Fridman (2:34:55.960)

and it's something like a billion.

Lex Fridman (2:34:57.720)

So then you would have to estimate

Lex Fridman (2:35:00.600)

that it would be possible for a technological civilization

Jeffrey Shainline (2:35:04.200)

to produce more than a billion black holes

Lex Fridman (2:35:07.360)

with the energy and matter at their disposal.

Lex Fridman (2:35:09.920)

And so one of the calculations in that paper,

Lex Fridman (2:35:12.640)

back of the envelope,

Lex Fridman (2:35:14.000)

but I think revealing nonetheless is that

Lex Fridman (2:35:16.320)

if you take a relatively common asteroid,

Jeffrey Shainline (2:35:20.720)

something that's about a kilometer in diameter,

Lex Fridman (2:35:23.640)

what I'm thinking of is just scrap material

Jeffrey Shainline (2:35:26.680)

laying around in our solar system

Lex Fridman (2:35:28.960)

and break it up into 10 kilogram chunks

Lex Fridman (2:35:31.360)

and turn each of those into a universe,

Lex Fridman (2:35:33.320)

then you would have made at least a trillion black holes

Jeffrey Shainline (2:35:38.080)

outpacing the star production rate

Lex Fridman (2:35:41.480)

by some three orders of magnitude.

Jeffrey Shainline (2:35:43.320)

That's one asteroid.

Lex Fridman (2:35:44.760)

So now if you envision an intelligent species

Jeffrey Shainline (2:35:46.840)

that would potentially have been devised initially

Lex Fridman (2:35:50.920)

by humans, but then based on superconducting

Jeffrey Shainline (2:35:53.120)

optoelectronic networks, no doubt,

Lex Fridman (2:35:55.160)

and they go out and populate,

Jeffrey Shainline (2:35:57.080)

they don't have to fill the galaxy.

Lex Fridman (2:35:58.920)

They just have to get out to the asteroid belt.

Jeffrey Shainline (2:36:01.640)

They could potentially dramatically outpace

Lex Fridman (2:36:05.240)

the rate at which stars are producing offspring universes.

Lex Fridman (2:36:07.840)

And then wouldn't you expect that

Lex Fridman (2:36:10.600)

that's where we came from instead of a star?

Jeffrey Shainline (2:36:13.080)

Yeah, so you have to somehow become masters of gravity,

Lex Fridman (2:36:16.520)

so like, or generate.

Jeffrey Shainline (2:36:17.360)

John, this is really gravity.

Lex Fridman (2:36:18.680)

So stars make black holes with gravity,

Lex Fridman (2:36:20.600)

but any force that can make the energy density

Lex Fridman (2:36:26.160)

can compactify matter to produce

Jeffrey Shainline (2:36:28.320)

a great enough energy density can form a singularity.

Lex Fridman (2:36:31.120)

It doesn't, it would not likely be gravity.

Jeffrey Shainline (2:36:33.480)

It's the weakest force.

Lex Fridman (2:36:34.480)

You're more likely to use something like the technologies

Jeffrey Shainline (2:36:38.480)

that we're developing for fusion, for example.

Lex Fridman (2:36:40.480)

So I don't know, the Large Ignition Facility

Jeffrey Shainline (2:36:44.240)

recently blasted a pellet with 100 really bright lasers

Lex Fridman (2:36:50.520)

and caused that to get dense enough

Jeffrey Shainline (2:36:53.360)

to engage in nuclear fusion.

Lex Fridman (2:36:55.280)

So something more like that,

Jeffrey Shainline (2:36:56.560)

or a tokamak with a really hot plasma, I'm not sure.

Lex Fridman (2:36:59.200)

Something, I don't know exactly how it would be done.

Jeffrey Shainline (2:37:02.160)

I do like the idea of that,

Lex Fridman (2:37:04.560)

especially just been reading a lot about gravitational waves

Lex Fridman (2:37:07.120)

and the fact that us humans with our technological

Lex Fridman (2:37:10.600)

capabilities, one of the most impressive

Jeffrey Shainline (2:37:14.120)

technological accomplishments of human history is LIGO,

Lex Fridman (2:37:17.360)

being able to precisely detect gravitational waves.

Jeffrey Shainline (2:37:20.720)

I'm particularly find appealing the idea

Lex Fridman (2:37:25.000)

that other alien civilizations from very far distances

Jeffrey Shainline (2:37:29.520)

communicate with gravity, with gravitational waves,

Lex Fridman (2:37:34.360)

because as you become greater and greater master of gravity,

Jeffrey Shainline (2:37:37.800)

which seems way out of reach for us right now,

Lex Fridman (2:37:40.600)

maybe that seems like a effective way of sending signals,

Jeffrey Shainline (2:37:44.280)

especially if your job is to manufacture black holes.

Lex Fridman (2:37:48.440)

Right.

Lex Fridman (2:37:49.280)

So that, so let me ask there,

Lex Fridman (2:37:53.360)

whatever, I mean, broadly thinking,

Jeffrey Shainline (2:37:56.440)

because we tend to think other alien civilizations

Lex Fridman (2:37:58.920)

would be very human like,

Lex Fridman (2:38:00.000)

but if we think of alien civilizations out there

Lex Fridman (2:38:04.080)

as basically generators of black holes,

Jeffrey Shainline (2:38:07.640)

however they do it, because they got stars,

Lex Fridman (2:38:10.960)

do you think there's a lot of them

Lex Fridman (2:38:12.800)

in our particular universe out there?

Lex Fridman (2:38:17.760)

In our universe?

Jeffrey Shainline (2:38:20.480)

Well, okay, let me ask, okay, this is great.

Lex Fridman (2:38:23.400)

Let me ask a very generic question

Lex Fridman (2:38:26.680)

and then let's see how you answer it,

Lex Fridman (2:38:29.200)

which is how many alien civilizations are out there?

Jeffrey Shainline (2:38:35.080)

If the hypothesis that I just described

Lex Fridman (2:38:38.400)

is on the right track,

Jeffrey Shainline (2:38:40.840)

it would mean that the parameters of our universe

Lex Fridman (2:38:43.880)

have been selected so that intelligent civilizations

Jeffrey Shainline (2:38:48.480)

will occur in sufficient numbers

Lex Fridman (2:38:51.120)

so that if they reach something

Jeffrey Shainline (2:38:54.720)

like supreme technological maturity,

Lex Fridman (2:38:56.480)

let's define that as the ability to produce black holes,

Jeffrey Shainline (2:39:00.080)

then that's not a highly improbable event.

Lex Fridman (2:39:02.640)

It doesn't need to happen often

Jeffrey Shainline (2:39:05.440)

because as I just described,

Lex Fridman (2:39:06.680)

if you get one of them in a galaxy,

Jeffrey Shainline (2:39:09.200)

you're gonna make more black holes

Lex Fridman (2:39:10.520)

than the stars in that galaxy.

Lex Fridman (2:39:12.720)

But there's also not a super strong motivation,

Lex Fridman (2:39:16.560)

well, it's not obvious that you need them

Jeffrey Shainline (2:39:21.600)

to be ubiquitous throughout the galaxy.

Lex Fridman (2:39:23.920)

Right.

Jeffrey Shainline (2:39:24.760)

One of the things that I try to emphasize in that paper

Lex Fridman (2:39:27.520)

is that given this idea

Jeffrey Shainline (2:39:30.640)

of how our parameters might've been selected,

Lex Fridman (2:39:35.120)

it's clear that it's a series of trade offs, right?

Jeffrey Shainline (2:39:39.280)

If you make, I mean, in order for intelligent life

Lex Fridman (2:39:42.040)

of our variety or anything resembling us to occur,

Jeffrey Shainline (2:39:45.760)

you need a bunch of stuff, you need stars.

Lex Fridman (2:39:47.600)

So that's right back to Smolin's roots of this idea,

Lex Fridman (2:39:51.080)

but you also need water to have certain properties.

Lex Fridman (2:39:54.440)

You need things like the rocky planets,

Jeffrey Shainline (2:39:58.760)

like the Earth to be within the habitable zone,

Lex Fridman (2:40:00.600)

all these things that you start talking about

Jeffrey Shainline (2:40:02.680)

in the field of astrobiology,

Lex Fridman (2:40:06.760)

trying to understand life in the universe,

Lex Fridman (2:40:08.800)

but you can't over emphasize,

Lex Fridman (2:40:10.480)

you can't tune the parameters so precisely

Jeffrey Shainline (2:40:13.600)

to maximize the number of stars

Lex Fridman (2:40:15.160)

or to give water exactly the properties

Jeffrey Shainline (2:40:18.960)

or to make rocky planets like Earth the most numerous.

Lex Fridman (2:40:22.280)

You have to compromise on all these things.

Lex Fridman (2:40:24.480)

And so I think the way to test this idea

Lex Fridman (2:40:27.360)

is to look at what parameters are necessary

Jeffrey Shainline (2:40:30.240)

for each of these different subsystems,

Lex Fridman (2:40:32.760)

and I've laid out a few that I think are promising,

Jeffrey Shainline (2:40:35.080)

there could be countless others,

Lex Fridman (2:40:36.640)

and see how changing the parameters

Jeffrey Shainline (2:40:40.840)

makes it more or less likely that stars would form

Lex Fridman (2:40:43.600)

and have long lifetimes or that rocky planets

Jeffrey Shainline (2:40:46.320)

in the habitable zone are likely to form,

Lex Fridman (2:40:48.280)

all these different things.

Lex Fridman (2:40:49.400)

So we can test how much these things are in a tug of war

Lex Fridman (2:40:53.360)

with each other, and the prediction would be

Jeffrey Shainline (2:40:56.280)

that we kind of sit at this central point

Lex Fridman (2:40:58.160)

where if you move the parameters too much,

Jeffrey Shainline (2:41:02.040)

stars aren't stable, or life doesn't form,

Lex Fridman (2:41:05.680)

or technology's infeasible,

Jeffrey Shainline (2:41:07.720)

because life alone, at least the kind of life

Lex Fridman (2:41:10.760)

that we know of, cannot make black holes.

Jeffrey Shainline (2:41:14.160)

We don't have this, well, I'm speaking for myself,

Lex Fridman (2:41:16.400)

you're a very fit and strong person,

Lex Fridman (2:41:18.560)

but it might be possible for you,

Lex Fridman (2:41:20.600)

but not for me to compress matter.

Lex Fridman (2:41:22.360)

So we need these technologies, but we don't know,

Lex Fridman (2:41:25.960)

we have not been able to quantify yet

Lex Fridman (2:41:28.960)

how finely adjusted the parameters would need to be

Lex Fridman (2:41:33.840)

in order for silicon to have the properties it does.

Jeffrey Shainline (2:41:35.600)

Okay, this is not directly speaking to what you're saying,

Lex Fridman (2:41:37.840)

you're getting to the Fermi paradox,

Jeffrey Shainline (2:41:39.520)

which is where are they, where are the life forms out there,

Lex Fridman (2:41:42.840)

how numerous are they, that sort of thing.

Lex Fridman (2:41:44.560)

What I'm trying to argue is that

Lex Fridman (2:41:46.200)

if this framework is on the right track,

Jeffrey Shainline (2:41:50.800)

a potentially correct explanation for our existence,

Lex Fridman (2:41:53.720)

we, it doesn't necessarily predict

Jeffrey Shainline (2:41:56.320)

that intelligent civilizations are just everywhere,

Lex Fridman (2:41:59.280)

because even if you just get one of them in a galaxy,

Jeffrey Shainline (2:42:02.520)

which is quite rare, it could be enough

Lex Fridman (2:42:05.760)

to dramatically increase the fecundity

Jeffrey Shainline (2:42:08.920)

of the universe as a whole.

Lex Fridman (2:42:10.160)

Yeah, and I wonder, once you start generating

Jeffrey Shainline (2:42:12.440)

the offspring for universes, black holes,

Lex Fridman (2:42:15.400)

how that has effect on the,

Lex Fridman (2:42:18.360)

what kind of effect does it have

Lex Fridman (2:42:19.920)

on the other candidate's civilizations

Lex Fridman (2:42:24.920)

within that universe?

Lex Fridman (2:42:26.160)

Maybe it has a destructive aspect,

Jeffrey Shainline (2:42:28.400)

or there could be some arguments

Lex Fridman (2:42:29.840)

about once you have a lot of offspring,

Jeffrey Shainline (2:42:32.040)

that that just quickly accelerates

Lex Fridman (2:42:34.040)

to where the other ones can't even catch up.

Jeffrey Shainline (2:42:35.760)

It could, but I guess if you want me

Lex Fridman (2:42:39.160)

to put my chips on the table or whatever,

Jeffrey Shainline (2:42:42.520)

I think I come down more on the side

Lex Fridman (2:42:46.320)

that intelligent life civilizations are rare.

Lex Fridman (2:42:52.400)

And I guess I follow Max Tegmark here.

Lex Fridman (2:42:57.160)

And also there's a lot of papers coming out recently

Jeffrey Shainline (2:43:01.000)

in the field of astrobiology that are seeming to say,

Lex Fridman (2:43:04.360)

all right, you just work through the numbers

Jeffrey Shainline (2:43:06.000)

on some modified Drake equation or something like that.

Lex Fridman (2:43:09.680)

And it looks like it's not improbable.

Jeffrey Shainline (2:43:13.040)

You shouldn't be surprised that an intelligent species

Lex Fridman (2:43:16.280)

has arisen in our galaxy,

Lex Fridman (2:43:18.040)

but if you think there's one the next solar system over,

Lex Fridman (2:43:20.360)

it's highly improbable.

Lex Fridman (2:43:21.600)

So I can see that the number,

Lex Fridman (2:43:23.960)

the probability of finding a civilization in a galaxy,

Jeffrey Shainline (2:43:28.280)

maybe it's most likely that you're gonna find

Lex Fridman (2:43:31.080)

one to a hundred or something.

Lex Fridman (2:43:32.880)

But okay, now it's really important

Lex Fridman (2:43:34.960)

to put a time window on that, I think,

Jeffrey Shainline (2:43:36.720)

because does that mean in the entire lifetime of the galaxy

Lex Fridman (2:43:40.720)

before it, so for in our case, before we run into Andromeda,

Jeffrey Shainline (2:43:49.760)

I think it's highly probable, I shouldn't say I think,

Lex Fridman (2:43:53.960)

it's tempting to believe that it's highly probable

Jeffrey Shainline (2:43:56.760)

that in that entire lifetime of your galaxy,

Lex Fridman (2:44:00.000)

you're gonna get at least one intelligent species,

Jeffrey Shainline (2:44:02.600)

maybe thousands or something like that.

Lex Fridman (2:44:05.160)

But it's also, I think, a little bit naive to think

Jeffrey Shainline (2:44:10.400)

that they're going to coincide in time

Lex Fridman (2:44:13.200)

and we'll be able to observe them.

Lex Fridman (2:44:14.960)

And also, if you look at the span of life on Earth,

Lex Fridman (2:44:20.080)

the Earth history, it was surprising to me

Jeffrey Shainline (2:44:24.480)

to kind of look at the amount of time,

Lex Fridman (2:44:27.880)

first of all, the short amount of time,

Jeffrey Shainline (2:44:29.680)

there's no life, it's surprising.

Lex Fridman (2:44:31.520)

Life sprang up pretty quickly.

Jeffrey Shainline (2:44:33.440)

It's single cell.

Lex Fridman (2:44:35.280)

But that's the point I'm trying to make

Jeffrey Shainline (2:44:36.960)

is like so much of life on Earth

Lex Fridman (2:44:42.040)

was just like single cell organisms, like most of it.

Jeffrey Shainline (2:44:45.840)

Most of it was like boring bacteria type of stuff.

Lex Fridman (2:44:48.640)

Well, bacteria are fascinating, but I take your point.

Jeffrey Shainline (2:44:50.640)

No, I get it.

Lex Fridman (2:44:51.520)

I mean, no offense to them.

Lex Fridman (2:44:52.880)

But this kind of speaking from the perspective

Lex Fridman (2:44:56.840)

of your paper of something that's able

Jeffrey Shainline (2:44:58.720)

to generate technology as we kind of understand it,

Lex Fridman (2:45:01.440)

that's a very short moment in time

Jeffrey Shainline (2:45:03.400)

relative to that full history of life on Earth.

Lex Fridman (2:45:08.640)

And maybe our universe is just saturated

Jeffrey Shainline (2:45:12.200)

with bacteria like humans.

Lex Fridman (2:45:15.880)

Right.

Lex Fridman (2:45:17.480)

But not the special extra AGI super humans,

Lex Fridman (2:45:24.200)

that those are very rare.

Lex Fridman (2:45:25.560)

And once those spring up, everything just goes to like,

Lex Fridman (2:45:30.400)

it accelerates very quickly.

Jeffrey Shainline (2:45:33.360)

Yeah, we just don't have enough data to really say,

Lex Fridman (2:45:36.520)

but I find this whole subject extremely engaging.

Jeffrey Shainline (2:45:40.000)

I mean, there's this concept,

Lex Fridman (2:45:41.680)

I think it's called the Rare Earth Hypothesis,

Jeffrey Shainline (2:45:43.880)

which is that basically stating that,

Lex Fridman (2:45:46.920)

okay, microbes were here right away

Jeffrey Shainline (2:45:49.040)

after the Hadian era where we were being bombarded.

Lex Fridman (2:45:52.120)

Well, after, yeah, bombarded by comets, asteroids,

Jeffrey Shainline (2:45:54.920)

things like that, and also after the moon formed.

Lex Fridman (2:45:57.080)

So once things settled down a little bit,

Jeffrey Shainline (2:45:59.560)

in a few hundred million years,

Lex Fridman (2:46:02.280)

you have microbes everywhere.

Lex Fridman (2:46:03.680)

And it could have been, we don't know exactly

Lex Fridman (2:46:05.200)

when it could have been remarkably brief that that took.

Lex Fridman (2:46:08.080)

So it does indicate that, okay,

Lex Fridman (2:46:10.200)

life forms relatively easily.

Jeffrey Shainline (2:46:12.160)

I think that alone is sort of a checker on the scale

Lex Fridman (2:46:15.920)

for the argument that the parameters that allow

Jeffrey Shainline (2:46:21.480)

even microbial life to form are not just a fluke.

Lex Fridman (2:46:24.360)

But anyway, that aside, yes,

Jeffrey Shainline (2:46:27.520)

then there was this long dormant period,

Lex Fridman (2:46:29.880)

not dormant, things were happening,

Lex Fridman (2:46:31.600)

but important things were happening

Lex Fridman (2:46:34.120)

for some two and a half billion years or something

Jeffrey Shainline (2:46:37.320)

after the metabolic process

Lex Fridman (2:46:40.160)

that releases oxygen was developed.

Jeffrey Shainline (2:46:42.840)

Then basically the planet's just sitting there,

Lex Fridman (2:46:46.120)

getting more and more oxygenated,

Jeffrey Shainline (2:46:47.560)

more and more oxygenated until it's enough

Lex Fridman (2:46:50.200)

that you can build these large, complex organisms.

Lex Fridman (2:46:54.160)

And so the Rare Earth Hypothesis would argue

Lex Fridman (2:46:56.480)

that the microbes are common everywhere

Jeffrey Shainline (2:47:01.000)

in any planet that's roughly in the habitable zone

Lex Fridman (2:47:04.160)

and has some water on it, it's probably gonna have those.

Lex Fridman (2:47:06.640)

But then getting to this Cambrian explosion

Lex Fridman (2:47:09.320)

that happened some between 500 and 600 million years ago,

Lex Fridman (2:47:13.760)

that's rare, you know?

Lex Fridman (2:47:16.360)

And I buy that, I think that is rare.

Lex Fridman (2:47:19.080)

So if you say how much life is in our galaxy,

Lex Fridman (2:47:21.880)

I think that's probably the right answer

Jeffrey Shainline (2:47:24.120)

is that microbes are everywhere.

Lex Fridman (2:47:26.440)

Cambrian explosion is extremely rare.

Lex Fridman (2:47:29.360)

And then, but the Cambrian explosion kind of went like that

Lex Fridman (2:47:32.760)

where within a couple of tens or a hundred million years,

Jeffrey Shainline (2:47:38.600)

all of these body plans came into existence.

Lex Fridman (2:47:40.960)

And basically all of the body plans

Jeffrey Shainline (2:47:43.120)

that are now in existence on the planet

Lex Fridman (2:47:46.080)

were formed in that brief window

Lex Fridman (2:47:48.720)

and we've just been shuffling around since then.

Lex Fridman (2:47:51.640)

So then what caused humans to pop out of that?

Jeffrey Shainline (2:47:54.840)

I mean, that could be another extremely rare threshold

Lex Fridman (2:48:01.920)

that a planet roughly in the habitable zone with water

Lex Fridman (2:48:06.200)

is not guaranteed to cross, you know?

Lex Fridman (2:48:08.400)

To me, it's fascinating for being humble,

Jeffrey Shainline (2:48:10.200)

like the humans cannot possibly be the most amazing thing

Lex Fridman (2:48:13.080)

that such, if you look at the entirety of the system

Jeffrey Shainline (2:48:15.880)

that Lee Smolin and you paint,

Lex Fridman (2:48:17.800)

that cannot possibly be the most amazing thing

Jeffrey Shainline (2:48:20.080)

that process generates.

Lex Fridman (2:48:21.480)

So like, if you look at the evolution,

Jeffrey Shainline (2:48:23.720)

what's the equivalent in the cosmological evolution

Lex Fridman (2:48:27.040)

and its selection for technology,

Lex Fridman (2:48:29.000)

the equivalent of the human eye or the human brain?

Lex Fridman (2:48:32.440)

Universes that are able to do some like,

Jeffrey Shainline (2:48:35.600)

they don't need the damn stars.

Lex Fridman (2:48:37.760)

They're able to just do some incredible generation

Jeffrey Shainline (2:48:42.120)

of complexity fast, like much more than,

Lex Fridman (2:48:46.640)

if you think about it,

Jeffrey Shainline (2:48:47.520)

it's like most of our universe is pretty freaking boring.

Lex Fridman (2:48:50.920)

There's not much going on, there's a few rocks flying around

Lex Fridman (2:48:53.400)

and there's some like apes

Lex Fridman (2:48:54.520)

that are just like doing podcasts on some weird planet.

Jeffrey Shainline (2:49:00.520)

It just seems very inefficient.

Lex Fridman (2:49:02.840)

If you think about like the amazing thing in the human eye,

Jeffrey Shainline (2:49:05.960)

the visual cortex can do, the brain, the nervous,

Lex Fridman (2:49:09.440)

everything that makes us more powerful

Jeffrey Shainline (2:49:12.840)

than single cell organisms.

Lex Fridman (2:49:15.480)

Like if there's an equivalent of that for universes,

Jeffrey Shainline (2:49:19.320)

like the richness of physics

Lex Fridman (2:49:21.680)

that could be expressed

Jeffrey Shainline (2:49:24.640)

through a particular set of parameters.

Lex Fridman (2:49:26.840)

Like, I mean, like for me,

Jeffrey Shainline (2:49:31.040)

I'm a sort of from a computer science perspective,

Lex Fridman (2:49:33.760)

huge fan of cellular automata,

Jeffrey Shainline (2:49:35.600)

which is a nice sort of pretty visual way

Lex Fridman (2:49:39.320)

to illustrate how different laws

Jeffrey Shainline (2:49:42.160)

can result in drastically different levels of complexity.

Lex Fridman (2:49:46.400)

So like, it's like, yeah, okay.

Lex Fridman (2:49:49.000)

So we're all like celebrating,

Lex Fridman (2:49:50.320)

look, our little cellular automata

Jeffrey Shainline (2:49:52.120)

is able to generate pretty triangles and squares

Lex Fridman (2:49:54.560)

and therefore we achieve general intelligence.

Lex Fridman (2:49:57.600)

And then there'll be like some badass Chuck Norris type,

Lex Fridman (2:50:01.840)

like universal Turing machine type of cellular automata.

Jeffrey Shainline (2:50:06.560)

They're able to generate other cellular automata

Lex Fridman (2:50:09.560)

that does any arbitrary level of computation off the bat.

Jeffrey Shainline (2:50:14.160)

Like those have to then exist.

Lex Fridman (2:50:16.480)

And then we're just like, we'll be forgotten.

Jeffrey Shainline (2:50:19.840)

This story, this podcast just entertains

Lex Fridman (2:50:23.800)

a few other apes for a few months.

Jeffrey Shainline (2:50:26.880)

Well, I'm kind of surprised to hear your cynicism.

Lex Fridman (2:50:30.240)

No, I'm very up.

Jeffrey Shainline (2:50:32.080)

I usually think of you as like one who celebrates humanity

Lex Fridman (2:50:36.120)

and all its forms and things like that.

Lex Fridman (2:50:37.600)

And I guess I just, I don't,

Lex Fridman (2:50:39.240)

I see it the way you just described.

Jeffrey Shainline (2:50:41.000)

I mean, okay, we've been here for 13.7 billion years

Lex Fridman (2:50:44.520)

and you're saying, gosh, that's a long time.

Jeffrey Shainline (2:50:47.240)

Let's get on with the show already.

Lex Fridman (2:50:48.480)

Some other universe could have kicked our butt by now,

Lex Fridman (2:50:51.520)

but that's putting a characteristic time.

Lex Fridman (2:50:55.360)

I mean, why is 13.7 billion a long time?

Lex Fridman (2:50:58.400)

I mean, compared to what?

Lex Fridman (2:51:00.440)

I guess, so when I look at our universe,

Jeffrey Shainline (2:51:02.320)

I see this extraordinary hierarchy

Lex Fridman (2:51:05.400)

that has developed over that time.

Lex Fridman (2:51:08.080)

So at the beginning, it was a chaotic mess of some plasma

Lex Fridman (2:51:13.720)

and nothing interesting going on there.

Lex Fridman (2:51:16.080)

And even for the first stars to form,

Lex Fridman (2:51:18.880)

that a lot of really interesting evolutionary processes

Jeffrey Shainline (2:51:23.880)

had to occur, by evolutionary in that sense,

Lex Fridman (2:51:26.240)

I just mean taking place over extended periods of time

Lex Fridman (2:51:30.520)

and structures are forming then.

Lex Fridman (2:51:32.000)

And then it took that first generation of stars

Jeffrey Shainline (2:51:34.640)

in order to produce the metals

Lex Fridman (2:51:38.760)

that then can more efficiently produce

Jeffrey Shainline (2:51:41.280)

another generation of stars.

Lex Fridman (2:51:42.440)

We're only the third generation of stars.

Lex Fridman (2:51:44.720)

So we might still be pretty quick to the game here.

Lex Fridman (2:51:47.920)

So, but I don't think, I don't, okay.

Lex Fridman (2:51:51.000)

So then you have these stars

Lex Fridman (2:51:52.440)

and then you have solar systems on those solar systems.

Jeffrey Shainline (2:51:54.640)

You have rocky worlds, you have gas giants,

Lex Fridman (2:51:58.840)

like all this complexity.

Lex Fridman (2:51:59.800)

And then you start getting life

Lex Fridman (2:52:01.120)

and the complexity that's evolved

Jeffrey Shainline (2:52:03.800)

through the evolutionary process in life forms

Lex Fridman (2:52:06.440)

is just, it's not a let down to me.

Jeffrey Shainline (2:52:09.880)

Just seeing that.

Lex Fridman (2:52:10.720)

Some of it is like some of the planets is like icy,

Jeffrey Shainline (2:52:14.600)

it's like different flavors of ice cream.

Lex Fridman (2:52:16.080)

They're icy, but there might be water underneath.

Jeffrey Shainline (2:52:18.560)

All kinds of life forms with some volcanoes,

Lex Fridman (2:52:21.120)

all kinds of weird stuff.

Jeffrey Shainline (2:52:22.320)

No, no, I don't, I think it's beautiful.

Lex Fridman (2:52:24.600)

I think our life is beautiful.

Lex Fridman (2:52:25.960)

And I think it was designed that by design,

Lex Fridman (2:52:29.560)

the scarcity of the whole thing.

Jeffrey Shainline (2:52:31.200)

I think mortality, as terrifying as it is,

Lex Fridman (2:52:33.840)

is fundamental to the whole reason we enjoy everything.

Jeffrey Shainline (2:52:37.280)

No, I think it's beautiful.

Lex Fridman (2:52:38.200)

I just think that all of us conscious beings

Jeffrey Shainline (2:52:42.400)

in the grand scheme of basically every scale

Lex Fridman (2:52:45.520)

will be completely forgotten.

Jeffrey Shainline (2:52:46.960)

Well, that's true.

Lex Fridman (2:52:47.800)

I think everything is transient

Lex Fridman (2:52:49.320)

and that would go back to maybe something more like Lao Tzu,

Lex Fridman (2:52:52.520)

the Tao Te Ching or something where it's like,

Jeffrey Shainline (2:52:55.440)

yes, there is nothing but change.

Lex Fridman (2:52:57.720)

There is nothing but emergence and dissolve and that's it.

Lex Fridman (2:53:00.720)

But I just, in this picture,

Lex Fridman (2:53:03.000)

this hierarchy that's developed,

Jeffrey Shainline (2:53:04.520)

I don't mean to say that now it gets to us

Lex Fridman (2:53:06.600)

and that's the pinnacle.

Jeffrey Shainline (2:53:07.440)

In fact, I think at a high level,

Lex Fridman (2:53:10.520)

the story I'm trying to tease out in my research is about,

Lex Fridman (2:53:14.240)

okay, well, so then what's the next level of hierarchy?

Lex Fridman (2:53:17.000)

And if it's, okay, we're kind of pretty smart.

Jeffrey Shainline (2:53:21.520)

I mean, talking about people like Lee Small

Lex Fridman (2:53:23.840)

and Alan Guth, Max Tegmark, okay, we're really smart.

Jeffrey Shainline (2:53:26.240)

Talking about me, okay, we're kind of,

Lex Fridman (2:53:28.480)

we can find our way to the grocery store or whatever,

Lex Fridman (2:53:30.760)

but what's next?

Lex Fridman (2:53:33.000)

I mean, what if there's another level of hierarchy

Jeffrey Shainline (2:53:36.120)

that grows on top of us

Lex Fridman (2:53:37.760)

that is even more profoundly capable?

Lex Fridman (2:53:40.280)

And I mean, we've talked a lot

Lex Fridman (2:53:42.080)

about superconducting sensors.

Jeffrey Shainline (2:53:43.560)

Imagine these cognitive systems far more capable than us

Lex Fridman (2:53:48.920)

residing somewhere else in the solar system

Jeffrey Shainline (2:53:52.320)

off of the surface of the earth,

Lex Fridman (2:53:53.640)

where it's much darker, much colder,

Jeffrey Shainline (2:53:55.400)

much more naturally suited to them.

Lex Fridman (2:53:57.080)

And they have these sensors that can detect single photons

Jeffrey Shainline (2:54:00.440)

of light from radio waves out to all across the spectrum

Lex Fridman (2:54:04.560)

of the gamma rays and just see the whole universe.

Lex Fridman (2:54:07.400)

And they just live in space

Lex Fridman (2:54:08.960)

with these massive collection optics so that they,

Lex Fridman (2:54:12.960)

what do they do?

Lex Fridman (2:54:13.800)

They just look out and experience that vast array

Jeffrey Shainline (2:54:18.800)

of what's being developed.

Lex Fridman (2:54:22.520)

And if you're such a system,

Jeffrey Shainline (2:54:25.120)

presumably you would do some things for fun.

Lex Fridman (2:54:28.920)

And the kind of fun thing I would do

Jeffrey Shainline (2:54:31.720)

as somebody who likes video games

Lex Fridman (2:54:33.840)

is I would create and maintain

Lex Fridman (2:54:37.040)

and observe something like earth.

Lex Fridman (2:54:42.760)

So in some sense, we're like all what players on a stage

Jeffrey Shainline (2:54:47.320)

for this superconducting cold computing system out there.

Lex Fridman (2:54:54.160)

I mean, all of this is fascinating to think.

Jeffrey Shainline (2:54:56.680)

The fact that you're actually designing systems

Lex Fridman (2:54:59.360)

here on earth that are trying to push this technological

Jeffrey Shainline (2:55:01.560)

at the very cutting edge and also thinking about

Lex Fridman (2:55:04.800)

how does the like the evolution of physical laws

Jeffrey Shainline (2:55:09.760)

lead us to the way we are is fascinating.

Lex Fridman (2:55:14.240)

That coupling is fascinating.

Jeffrey Shainline (2:55:15.920)

It's like the ultimate rigorous application of philosophy

Lex Fridman (2:55:20.800)

to the rigorous application of engineering.

Lex Fridman (2:55:23.680)

So Jeff, you're one of the most fascinating.

Lex Fridman (2:55:26.400)

I'm so glad I did not know much about you

Jeffrey Shainline (2:55:29.000)

except through your work.

Lex Fridman (2:55:30.440)

And I'm so glad we got this chance to talk.

Jeffrey Shainline (2:55:34.200)

You're one of the best explainers

Lex Fridman (2:55:37.800)

of exceptionally difficult concepts.

Lex Fridman (2:55:40.940)

And you're also, speaking of like fractal,

Lex Fridman (2:55:44.600)

you're able to function intellectually

Jeffrey Shainline (2:55:46.680)

at all levels of the stack, which I deeply appreciate.

Lex Fridman (2:55:50.240)

This was really fun.

Jeffrey Shainline (2:55:51.640)

You're a great educator, a great scientist.

Lex Fridman (2:55:53.600)

It's an honor that you would spend

Jeffrey Shainline (2:55:56.080)

your valuable time with me.

Lex Fridman (2:55:57.280)

It's an honor that you would spend your time with me as well.

Jeffrey Shainline (2:56:00.120)

Thanks, Jeff.

Lex Fridman (2:56:01.760)

Thanks for listening to this conversation

Jeffrey Shainline (2:56:03.560)

with Jeff Schoenlein.

Lex Fridman (2:56:05.240)

To support this podcast,

Jeffrey Shainline (2:56:06.680)

please check out our sponsors in the description.

Lex Fridman (2:56:09.580)

And now let me leave you with some words

Jeffrey Shainline (2:56:12.060)

from the great John Carmack,

Lex Fridman (2:56:14.320)

who surely will be a guest on this podcast soon.

Jeffrey Shainline (2:56:18.220)

Because of the nature of Moore's Law,

Lex Fridman (2:56:20.100)

anything that an extremely clever graphics programmer

Jeffrey Shainline (2:56:22.800)

can do at one point can be replicated

Lex Fridman (2:56:26.040)

by a merely competent programmer

Jeffrey Shainline (2:56:27.920)

some number of years later.

Lex Fridman (2:56:30.520)

Thank you for listening and hope to see you next time.

Jeffrey Shainline (30:00.800)

Matrix 4 is coming out, so maybe that's related.

Lex Fridman (30:04.400)

I'm not sure.

Jeffrey Shainline (30:05.240)

I'm dressed for the job.

Lex Fridman (30:06.400)

I was trying to get to become an extra on Matrix 4,

Jeffrey Shainline (30:09.360)

didn't work out.

Lex Fridman (30:10.680)

Anyway, so what's the speed of these packets?

Jeffrey Shainline (30:13.280)

You'll have to find another gig.

Lex Fridman (30:15.000)

I know, I'm sorry.

Lex Fridman (30:16.600)

So the speed of the pack is actually these flux ons,

Lex Fridman (30:19.440)

these sort of pulses of current

Jeffrey Shainline (30:24.300)

that are generated by Joseph's injunctions,

Lex Fridman (30:26.200)

they can actually propagate very close

Jeffrey Shainline (30:28.480)

to the speed of light,

Lex Fridman (30:29.740)

maybe something like a third of the speed of light.

Jeffrey Shainline (30:31.920)

That's quite fast.

Lex Fridman (30:32.880)

So one of the reasons why Joseph's injunctions are appealing

Jeffrey Shainline (30:37.280)

is because their signals can propagate quite fast

Lex Fridman (30:40.600)

and they can also switch very fast.

Lex Fridman (30:43.440)

What I mean by switch is perform that operation

Lex Fridman (30:46.080)

that I described where you add current to the loop.

Jeffrey Shainline (30:49.440)

That can happen within a few tens of picoseconds.

Lex Fridman (30:53.960)

So you can get devices that operate

Jeffrey Shainline (30:56.880)

in the hundreds of gigahertz range.

Lex Fridman (30:58.840)

And by comparison, most processors

Jeffrey Shainline (31:02.080)

in our conventional computers operate closer

Lex Fridman (31:04.960)

to the one gigahertz range, maybe three gigahertz

Jeffrey Shainline (31:08.360)

seems to be kind of where those speeds have leveled out.

Lex Fridman (31:12.960)

The gamers listening to this are getting really excited

Lex Fridman (31:15.600)

to overclock their system to like, what is it?

Lex Fridman (31:18.080)

Like four gigahertz or something,

Jeffrey Shainline (31:19.560)

a hundred sounds incredible.

Lex Fridman (31:21.980)

Can I just as a tiny tangent,

Jeffrey Shainline (31:24.060)

is the physics of this understood well

Lex Fridman (31:26.880)

how to do this stably?

Jeffrey Shainline (31:28.520)

Oh yes, the physics is understood well.

Lex Fridman (31:30.160)

The physics of Joseph's injunctions is understood well.

Jeffrey Shainline (31:32.540)

The technology is understood quite well too.

Lex Fridman (31:34.520)

The reasons why it hasn't displaced

Jeffrey Shainline (31:37.620)

silicon microelectronics in conventional digital computing

Lex Fridman (31:41.720)

I think are more related to what I was alluding to before

Jeffrey Shainline (31:45.040)

about the myriad practical, almost mundane aspects

Lex Fridman (31:49.240)

of silicon that make it so useful.

Jeffrey Shainline (31:52.080)

You can make a transistor ever smaller and smaller

Lex Fridman (31:55.880)

and it will still perform its digital function quite well.

Jeffrey Shainline (31:58.840)

The same is not true of a Joseph's injunction.

Lex Fridman (32:00.780)

You really, they don't, they just,

Jeffrey Shainline (32:02.400)

it's not the same thing that there's this feature

Lex Fridman (32:04.440)

that you can keep making smaller and smaller

Lex Fridman (32:06.280)

and it'll keep performing the same operations.

Lex Fridman (32:08.220)

This loop I described, any Joseph's in circuit,

Jeffrey Shainline (32:11.480)

well, I wanna be careful, I shouldn't say

Lex Fridman (32:13.600)

any Joseph's in circuit, but many Joseph's in circuits,

Jeffrey Shainline (32:17.240)

the way they process information

Lex Fridman (32:19.440)

or the way they perform whatever function it is

Jeffrey Shainline (32:21.280)

they're trying to do,

Lex Fridman (32:22.120)

maybe it's sensing a weak magnetic field,

Jeffrey Shainline (32:24.560)

it depends on an interplay between the junction

Lex Fridman (32:27.480)

and that loop.

Lex Fridman (32:28.800)

And you can't make that loop much smaller.

Lex Fridman (32:30.560)

And it's not for practical reasons

Jeffrey Shainline (32:32.120)

that have to do with lithography.

Lex Fridman (32:33.480)

It's for fundamental physical reasons

Jeffrey Shainline (32:35.680)

about the way the magnetic field interacts

Lex Fridman (32:38.960)

with that superconducting material.

Jeffrey Shainline (32:41.160)

There are physical limits that no matter how good

Lex Fridman (32:44.360)

our technology got, those circuits would,

Jeffrey Shainline (32:47.260)

I think would never be able to be scaled down

Lex Fridman (32:50.360)

to the densities that silicon microelectronics can.

Jeffrey Shainline (32:54.360)

I don't know if we mentioned,

Lex Fridman (32:55.560)

is there something interesting

Jeffrey Shainline (32:56.960)

about the various superconducting materials involved

Lex Fridman (33:00.200)

or is it all?

Jeffrey Shainline (33:01.040)

There's a lot of stuff that's interesting.

Lex Fridman (33:02.640)

And it's not silicon.

Jeffrey Shainline (33:04.440)

It's not silicon, no.

Lex Fridman (33:05.840)

So like it's some materials that also required

Jeffrey Shainline (33:09.520)

to be super cold, four Kelvin and so on.

Lex Fridman (33:12.560)

So let's dissect a couple of those different things.

Jeffrey Shainline (33:15.280)

The super cold part,

Lex Fridman (33:16.280)

let me just mention for your gamers out there

Jeffrey Shainline (33:19.640)

that are trying to clock it at four gigahertz

Lex Fridman (33:21.320)

and would love to go to 400.

Lex Fridman (33:22.160)

What kind of cooling system can achieve four Kelvin?

Lex Fridman (33:24.120)

Four Kelvin, you need liquid helium.

Lex Fridman (33:26.280)

And so liquid helium is expensive.

Lex Fridman (33:29.040)

It's inconvenient.

Jeffrey Shainline (33:29.880)

You need a cryostat that sits there

Lex Fridman (33:32.080)

and the energy consumption of that cryostat

Jeffrey Shainline (33:36.520)

is impracticable for, it's not going in your cell phone.

Lex Fridman (33:40.080)

So you can picture holding your cell phone like this

Lex Fridman (33:42.080)

and then something the size of a keg of beer or something

Lex Fridman (33:46.600)

on your back to cool it.

Jeffrey Shainline (33:47.800)

Like that makes no sense.

Lex Fridman (33:49.520)

So if you're trying to make this in consumer devices,

Jeffrey Shainline (33:54.120)

electronics that are ubiquitous across society,

Lex Fridman (33:57.000)

superconductors are not in the race for that.

Jeffrey Shainline (33:59.280)

For now, but you're saying,

Lex Fridman (34:01.000)

so just to frame the conversation,

Jeffrey Shainline (34:03.240)

maybe the thing we're focused on

Lex Fridman (34:05.520)

is computing systems that serve as servers, like large.

Jeffrey Shainline (34:10.360)

Yes, large systems.

Lex Fridman (34:11.800)

So then you can contrast what's going on in your cell phone

Jeffrey Shainline (34:14.800)

with what's going on at one of the supercomputers.

Lex Fridman (34:19.400)

Colleague Katie Schuman invited us out to Oak Ridge

Jeffrey Shainline (34:22.080)

a few years ago, so we got to see Titan

Lex Fridman (34:24.080)

and that was when they were building Summit.

Lex Fridman (34:26.120)

So these are some high performance supercomputers

Lex Fridman (34:29.240)

out in Tennessee and those are filling entire rooms

Jeffrey Shainline (34:32.480)

the size of warehouses.

Lex Fridman (34:33.920)

So once you're at that level, okay,

Jeffrey Shainline (34:36.280)

there you're already putting a lot of power into cooling.

Lex Fridman (34:39.080)

Cooling is part of your engineering task

Jeffrey Shainline (34:42.240)

that you have to deal with.

Lex Fridman (34:43.600)

So there it's not entirely obvious

Jeffrey Shainline (34:45.600)

that cooling to four Kelvin is out of the question.

Lex Fridman (34:49.520)

It has not happened yet and I can speak to why that is

Jeffrey Shainline (34:53.240)

in the digital domain if you're interested.

Lex Fridman (34:55.520)

I think it's not going to happen.

Jeffrey Shainline (34:57.520)

I don't think superconductors are gonna replace

Lex Fridman (35:01.240)

semiconductors for digital computation.

Jeffrey Shainline (35:05.880)

There are a lot of reasons for that,

Lex Fridman (35:07.640)

but I think ultimately what it comes down to

Jeffrey Shainline (35:09.800)

is all things considered cooling errors,

Lex Fridman (35:13.440)

scaling down to feature sizes, all that stuff,

Jeffrey Shainline (35:16.080)

semiconductors work better at the system level.

Lex Fridman (35:19.400)

Is there some aspect of just curious

Lex Fridman (35:22.720)

about the historical momentum of this?

Lex Fridman (35:25.560)

Is there some power to the momentum of an industry

Lex Fridman (35:28.200)

that's mass manufacturing using a certain material?

Lex Fridman (35:31.200)

Is this like a Titanic shifting?

Jeffrey Shainline (35:33.680)

Like what's your sense when a good idea comes along,

Lex Fridman (35:37.120)

how good does that idea need to be

Lex Fridman (35:39.920)

for the Titanic to start shifting?

Lex Fridman (35:42.640)

That's an excellent question.

Jeffrey Shainline (35:44.200)

That's an excellent way to frame it.

Lex Fridman (35:46.520)

And you know, I don't know the answer to that,

Lex Fridman (35:51.320)

but what I think is, okay,

Lex Fridman (35:53.600)

so the history of the superconducting logic

Jeffrey Shainline (35:56.400)

goes back to the 70s.

Lex Fridman (35:58.000)

IBM made a big push to do

Jeffrey Shainline (35:59.760)

superconducting digital computing in the 70s.

Lex Fridman (36:02.400)

And they made some choices about their devices

Lex Fridman (36:06.080)

and their architectures and things that in hindsight,

Lex Fridman (36:09.440)

were kind of doomed to fail.

Lex Fridman (36:11.000)

And I don't mean any disrespect for the people that did it,

Lex Fridman (36:13.120)

it was hard to see at the time.

Lex Fridman (36:14.280)

But then another generation of superconducting logic

Lex Fridman (36:17.880)

was introduced, I wanna say the 90s,

Jeffrey Shainline (36:22.320)

someone named Lykarev and Seminov,

Lex Fridman (36:25.000)

they proposed an entire family of circuits

Jeffrey Shainline (36:28.280)

based on Joseph's injunctions

Lex Fridman (36:29.920)

that are doing digital computing based on logic gates

Lex Fridman (36:33.440)

and or not these kinds of things.

Lex Fridman (36:37.920)

And they showed how it could go hundreds of times faster

Jeffrey Shainline (36:41.560)

than silicon microelectronics.

Lex Fridman (36:43.200)

And it's extremely exciting.

Jeffrey Shainline (36:45.360)

I wasn't working in the field at that time,

Lex Fridman (36:47.040)

but later when I went back and read the literature,

Jeffrey Shainline (36:49.560)

I was just like, wow, this is so awesome.

Lex Fridman (36:53.040)

And so you might think, well,

Jeffrey Shainline (36:56.000)

the reason why it didn't display silicon

Lex Fridman (36:58.280)

is because silicon already had so much momentum

Jeffrey Shainline (37:00.400)

at that time.

Lex Fridman (37:01.720)

But that was the 90s.

Jeffrey Shainline (37:02.960)

Silicon kept that momentum

Lex Fridman (37:04.320)

because it had the simple way to keep getting better.

Jeffrey Shainline (37:06.960)

You just make features smaller and smaller.

Lex Fridman (37:08.720)

So it would have to be,

Jeffrey Shainline (37:11.800)

I don't think it would have to be that much better

Lex Fridman (37:13.560)

than silicon to displace it.

Lex Fridman (37:15.440)

But the problem is it's just not better than silicon.

Lex Fridman (37:17.800)

It might be better than silicon in one metric,

Jeffrey Shainline (37:19.960)

speed of a switching operation

Lex Fridman (37:21.440)

or power consumption of a switching operation.

Lex Fridman (37:24.400)

But building a digital computer is a lot more

Lex Fridman (37:26.680)

than just that elemental operation.

Jeffrey Shainline (37:28.720)

It's everything that goes into it,

Lex Fridman (37:31.040)

including the manufacturing, including the packaging,

Jeffrey Shainline (37:33.320)

including the various materials aspects of things.

Lex Fridman (37:38.840)

So the reason why,

Lex Fridman (37:40.600)

and even in some of those early papers,

Lex Fridman (37:42.800)

I can't remember which one it was,

Jeffrey Shainline (37:44.120)

Lykarev said something along the lines of,

Lex Fridman (37:47.480)

you can see how we could build an entire family

Jeffrey Shainline (37:49.920)

of digital electronic circuits based on these components.

Lex Fridman (37:52.800)

They could go a hundred or more times faster

Jeffrey Shainline (37:55.000)

than semiconductor logic gates.

Lex Fridman (37:59.320)

But I don't think that's the right way

Jeffrey Shainline (38:00.920)

to use superconducting electronic circuits.

Lex Fridman (38:02.680)

He didn't say what the right way was,

Lex Fridman (38:04.320)

but he basically said digital logic,

Lex Fridman (38:07.480)

trying to steal the show from silicon

Jeffrey Shainline (38:11.280)

is probably not what these circuits

Lex Fridman (38:13.440)

are most suited to accomplish.

Lex Fridman (38:16.360)

So if we can just linger and use the word computation.

Lex Fridman (38:20.840)

When you talk about computation, how do you think about it?

Lex Fridman (38:24.080)

Do you think purely on just the switching,

Lex Fridman (38:28.920)

or do you think something a little bit larger scale,

Jeffrey Shainline (38:31.320)

a circuit taken together,

Lex Fridman (38:32.720)

performing the basic arithmetic operations

Jeffrey Shainline (38:36.940)

that are then required to do the kind of computation

Lex Fridman (38:40.400)

that makes up a computer?

Jeffrey Shainline (38:42.160)

Because when we talk about the speed of computation,

Lex Fridman (38:44.400)

is it boiled down to the basic switching,

Jeffrey Shainline (38:46.960)

or is there some bigger picture

Lex Fridman (38:48.400)

that you're thinking about?

Jeffrey Shainline (38:49.240)

Well, all right, so maybe we should disambiguate.

Lex Fridman (38:52.240)

There are a variety of different kinds of computation.

Jeffrey Shainline (38:55.600)

I don't pretend to be an expert

Lex Fridman (38:57.200)

in the theory of computation or anything like that.

Jeffrey Shainline (39:00.180)

I guess it's important to differentiate though

Lex Fridman (39:02.640)

between digital logic,

Jeffrey Shainline (39:05.800)

which represents information as a series of bits,

Lex Fridman (39:09.800)

binary digits, which you can think of them

Jeffrey Shainline (39:13.000)

as zeros and ones or whatever.

Lex Fridman (39:14.160)

Usually they correspond to a physical system

Jeffrey Shainline (39:17.400)

that has two very well separated states.

Lex Fridman (39:21.240)

And then other kinds of computation,

Jeffrey Shainline (39:22.860)

like we'll get into more the way your brain works,

Lex Fridman (39:25.280)

which it is, I think,

Jeffrey Shainline (39:27.600)

indisputably processing information,

Lex Fridman (39:30.400)

but where the computation begins and ends

Jeffrey Shainline (39:34.080)

is not anywhere near as well defined.

Lex Fridman (39:36.360)

It doesn't depend on these two levels.

Jeffrey Shainline (39:39.680)

Here's a zero, here's a one.

Lex Fridman (39:41.320)

There's a lot of gray area

Jeffrey Shainline (39:42.640)

that's usually referred to as analog computing.

Lex Fridman (39:45.640)

Also in conventional digital computers

Jeffrey Shainline (39:49.860)

or digital computers in general,

Lex Fridman (39:54.240)

you have a concept of what's called arithmetic depth,

Jeffrey Shainline (39:57.300)

which is jargon that basically means

Lex Fridman (39:59.820)

how many sequential operations are performed

Jeffrey Shainline (40:03.860)

to turn an input into an output.

Lex Fridman (40:07.680)

And those kinds of computations in digital systems

Jeffrey Shainline (40:10.900)

are highly serial, meaning that data streams,

Lex Fridman (40:14.500)

they don't branch off too far to the side.

Jeffrey Shainline (40:16.500)

You do, you have to pull some information over there

Lex Fridman (40:18.900)

and access memory from here and stuff like that.

Lex Fridman (40:20.900)

But by and large, the computation proceeds

Lex Fridman (40:24.340)

in a serial manner.

Jeffrey Shainline (40:26.220)

It's not that way in the brain.

Lex Fridman (40:27.740)

In the brain, you're always drawing information

Jeffrey Shainline (40:30.740)

from different places.

Lex Fridman (40:31.580)

It's much more network based computing.

Jeffrey Shainline (40:33.820)

Neurons don't wait for their turn.

Lex Fridman (40:35.680)

They fire when they're ready to fire.

Lex Fridman (40:37.180)

And so it's asynchronous.

Lex Fridman (40:39.220)

So one of the other things about a digital system

Jeffrey Shainline (40:41.680)

is you're performing these operations on a clock.

Lex Fridman (40:44.500)

And that's a crucial aspect of it.

Jeffrey Shainline (40:46.700)

Get rid of a clock in a digital system,

Lex Fridman (40:48.900)

nothing makes sense anymore.

Jeffrey Shainline (40:50.460)

The brain has no clock.

Lex Fridman (40:51.580)

It builds its own timescales based on its internal activity.

Lex Fridman (40:56.580)

So you can think of the brain as kind of like this,

Lex Fridman (40:59.500)

like network computation,

Jeffrey Shainline (41:00.940)

where it's actually really trivial, simple computers,

Lex Fridman (41:05.700)

just a huge number of them and they're networked.

Jeffrey Shainline (41:08.980)

I would say it is complex, sophisticated little processors

Lex Fridman (41:12.940)

and there's a huge number of them.

Jeffrey Shainline (41:14.420)

Neurons are not, are not simple.

Lex Fridman (41:16.180)

I don't mean to offend neurons.

Jeffrey Shainline (41:17.620)

They're very complicated and beautiful and yeah,

Lex Fridman (41:19.780)

but we often oversimplify them.

Jeffrey Shainline (41:21.980)

Yes, they're actually like there's computation happening

Lex Fridman (41:24.820)

within a neuron.

Jeffrey Shainline (41:25.660)

Right, so I would say to think of a transistor

Lex Fridman (41:29.520)

as the building block of a digital computer is accurate.

Jeffrey Shainline (41:32.340)

You use a few transistors to make your logic gates.

Lex Fridman (41:34.660)

You build up more, you build up processors

Jeffrey Shainline (41:37.060)

from logic gates and things like that.

Lex Fridman (41:39.140)

So you can think of a transistor

Jeffrey Shainline (41:40.600)

as a fundamental building block,

Lex Fridman (41:42.300)

or you can think of,

Jeffrey Shainline (41:43.380)

as we get into more highly parallelized architectures,

Lex Fridman (41:46.360)

you can think of a processor

Jeffrey Shainline (41:47.700)

as a fundamental building block.

Lex Fridman (41:49.300)

To make the analogy to the neuro side of things,

Jeffrey Shainline (41:53.180)

a neuron is not a transistor.

Lex Fridman (41:55.320)

A neuron is a processor.

Jeffrey Shainline (41:57.300)

It has synapses, even synapses are not transistors,

Lex Fridman (42:00.220)

but they are more,

Jeffrey Shainline (42:02.180)

they're lower on the information processing hierarchy

Lex Fridman (42:04.820)

in a sense.

Jeffrey Shainline (42:05.660)

They do a bulk of the computation,

Lex Fridman (42:08.180)

but neurons are entire processors in and of themselves

Jeffrey Shainline (42:13.580)

that can take in many different kinds of inputs

Lex Fridman (42:16.300)

on many different spatial and temporal scales

Lex Fridman (42:18.780)

and produce many different kinds of outputs

Lex Fridman (42:20.820)

so that they can perform different computations

Jeffrey Shainline (42:23.820)

in different contexts.

Lex Fridman (42:24.860)

So this is where enters this distinction

Jeffrey Shainline (42:27.440)

between computation and communication.

Lex Fridman (42:30.740)

So you can think of neurons performing computation

Lex Fridman (42:34.140)

and the inter, the networking,

Lex Fridman (42:36.580)

the interconnectivity of neurons

Jeffrey Shainline (42:39.000)

is communication between neurons.

Lex Fridman (42:40.940)

And you see this with very large server systems.

Jeffrey Shainline (42:43.500)

I've been, I mentioned offline,

Lex Fridman (42:45.020)

we've been talking to Jim Keller,

Jeffrey Shainline (42:46.180)

whose dream is to build giant computers

Lex Fridman (42:48.140)

that, you know, the bottom like there

Jeffrey Shainline (42:51.220)

is often the communication

Lex Fridman (42:52.380)

between the different pieces of computing.

Lex Fridman (42:54.700)

So in this paper that we mentioned,

Lex Fridman (42:57.380)

Optoelectronic Intelligence,

Jeffrey Shainline (42:59.660)

you say electrons excel at computation

Lex Fridman (43:03.220)

while light is excellent for communication.

Jeffrey Shainline (43:08.380)

Maybe you can linger and say in this context,

Lex Fridman (43:11.060)

what do you mean by computation and communication?

Lex Fridman (43:13.980)

What are electrons, what is light

Lex Fridman (43:17.420)

and why do they excel at those two tasks?

Jeffrey Shainline (43:20.660)

Yeah, just to first speak to computation

Lex Fridman (43:23.620)

versus communication,

Jeffrey Shainline (43:25.620)

I would say computation is essentially taking in

Lex Fridman (43:30.340)

some information, performing operations

Jeffrey Shainline (43:33.860)

on that information and producing new,

Lex Fridman (43:37.220)

hopefully more useful information.

Lex Fridman (43:39.060)

So for example, imagine you have a picture in front of you

Lex Fridman (43:45.020)

and there is a key in it

Lex Fridman (43:48.020)

and that's what you're looking for,

Lex Fridman (43:48.940)

for whatever reason, you wanna find the key,

Jeffrey Shainline (43:50.700)

we all wanna find the key.

Lex Fridman (43:51.580)

So the input is that entire picture

Lex Fridman (43:56.540)

and the output might be the coordinates where the key is.

Lex Fridman (43:59.060)

So you've reduced the total amount of information you have

Lex Fridman (44:01.540)

but you found the useful information

Lex Fridman (44:03.020)

for you in that present moment,

Jeffrey Shainline (44:04.380)

that's the useful information.

Lex Fridman (44:05.220)

And you think about this computation

Lex Fridman (44:07.180)

as the controlled synchronous sequential?

Lex Fridman (44:10.820)

Not necessarily, it could be,

Jeffrey Shainline (44:12.700)

that could be how your system is performing the computation

Lex Fridman (44:16.220)

or it could be asynchronous,

Jeffrey Shainline (44:19.300)

there are lots of ways to find the key.

Lex Fridman (44:21.420)

It depends on the nature of the data,

Jeffrey Shainline (44:23.700)

it depends on, that's a very simplified example,

Lex Fridman (44:27.540)

a picture with a key in it,

Lex Fridman (44:28.700)

what about if you're in the world

Lex Fridman (44:30.540)

and you're trying to decide the best way

Lex Fridman (44:32.500)

to live your life?

Lex Fridman (44:35.940)

It might be interactive,

Jeffrey Shainline (44:37.020)

it might be there might be some recurrence

Lex Fridman (44:38.580)

or some weird asynchrony, I got it.

Lex Fridman (44:41.340)

But there's an input and there's an output

Lex Fridman (44:43.260)

and you do some stuff in the middle

Jeffrey Shainline (44:44.460)

that actually goes from the input to the output.

Lex Fridman (44:46.020)

You've taken in information

Lex Fridman (44:47.340)

and output different information,

Lex Fridman (44:49.100)

hopefully reducing the total amount of information

Lex Fridman (44:51.820)

and extracting what's useful.

Lex Fridman (44:53.820)

Communication is then getting that information

Jeffrey Shainline (44:57.780)

from the location at which it's stored

Lex Fridman (44:59.460)

because information is physical as Landauer emphasized

Lex Fridman (45:02.660)

and so it is in one place

Lex Fridman (45:04.940)

and you need to get that information to another place

Lex Fridman (45:07.860)

so that something else can use it

Lex Fridman (45:10.100)

for whatever computation it's working on.

Jeffrey Shainline (45:12.020)

Maybe it's part of the same network

Lex Fridman (45:13.460)

and you're all trying to solve the same problem

Lex Fridman (45:15.020)

but neuron A over here just deduced something

Lex Fridman (45:20.500)

based on its inputs

Lex Fridman (45:21.620)

and it's now sending that information across the network

Lex Fridman (45:25.180)

to another location

Lex Fridman (45:26.460)

so that would be the act of communication.

Lex Fridman (45:28.420)

Can you linger on Landauer

Lex Fridman (45:29.900)

and saying information is physical?

Lex Fridman (45:31.820)

Rolf Landauer, not to be confused with Lev Landauer.

Jeffrey Shainline (45:35.340)

Yeah, and he made huge contributions

Lex Fridman (45:38.020)

to our understanding of the reversibility of information

Lex Fridman (45:42.980)

and this concept that energy has to be dissipated

Lex Fridman (45:46.700)

in computing when the computation is irreversible

Lex Fridman (45:50.100)

but if you can manage to make it reversible

Lex Fridman (45:52.140)

then you don't need to expend energy

Lex Fridman (45:55.060)

but if you do expend energy to perform a computation

Lex Fridman (45:59.660)

there's sort of a minimal amount that you have to do

Lex Fridman (46:02.300)

and it's KT log two.

Lex Fridman (46:04.460)

And it's all somehow related

Jeffrey Shainline (46:05.900)

to the second law of thermodynamics

Lex Fridman (46:07.660)

and that the universe is an information process

Lex Fridman (46:09.620)

and then we're living in a simulation.

Lex Fridman (46:11.500)

So okay, sorry, sorry for that tangent.

Lex Fridman (46:13.980)

So that's the defining the distinction

Lex Fridman (46:17.140)

between computation and communication.

Jeffrey Shainline (46:19.580)

Let me say one more thing just to clarify.

Lex Fridman (46:21.940)

Communication ideally does not change the information.

Jeffrey Shainline (46:27.100)

It moves it from one place to another

Lex Fridman (46:28.900)

but it is preserved.

Jeffrey Shainline (46:30.940)

Got it, okay.

Lex Fridman (46:32.500)

All right, that's beautiful.

Lex Fridman (46:33.700)

So then the electron versus light distinction

Lex Fridman (46:38.620)

and why are electrons good at computation

Lex Fridman (46:42.380)

and light good at communication?

Lex Fridman (46:44.540)

Yes, there's a lot that goes into it I guess

Lex Fridman (46:48.820)

but just try to speak to the simplest part of it.

Lex Fridman (46:54.100)

Electrons interact strongly with one another.

Jeffrey Shainline (46:56.980)

They're charged particles.

Lex Fridman (46:58.340)

So if I pile a bunch of them over here

Jeffrey Shainline (47:02.020)

they're feeling a certain amount of force

Lex Fridman (47:03.860)

and they wanna move somewhere else.

Jeffrey Shainline (47:05.700)

They're strongly interactive.

Lex Fridman (47:06.900)

You can also get them to sit still.

Jeffrey Shainline (47:08.900)

You can, an electron has a mass

Lex Fridman (47:10.660)

so you can cause it to be spatially localized.

Lex Fridman (47:15.860)

So for computation that's useful

Lex Fridman (47:18.100)

because now I can make these little devices

Jeffrey Shainline (47:20.140)

that put a bunch of electrons over here

Lex Fridman (47:21.940)

and then I change the state of a gate

Jeffrey Shainline (47:25.620)

like I've been describing,

Lex Fridman (47:26.500)

put a different voltage on this gate

Lex Fridman (47:28.380)

and now I move the electrons over here.

Lex Fridman (47:29.980)

Now they're sitting somewhere else.

Jeffrey Shainline (47:31.220)

I have a physical mechanism

Lex Fridman (47:33.980)

with which I can represent information.

Jeffrey Shainline (47:36.020)

It's spatially localized and I have knobs

Lex Fridman (47:38.140)

that I can adjust to change where those electrons are

Jeffrey Shainline (47:41.220)

or what they're doing.

Lex Fridman (47:42.380)

Light by contrast, photons of light

Jeffrey Shainline (47:45.220)

which are the discrete packets of energy

Lex Fridman (47:48.100)

that were identified by Einstein,

Jeffrey Shainline (47:50.740)

they do not interact with each other

Lex Fridman (47:54.540)

especially at low light levels.

Jeffrey Shainline (47:56.260)

If you're in a medium and you have a bright high light level

Lex Fridman (48:00.380)

you can get them to interact with each other

Jeffrey Shainline (48:02.540)

through the interaction with that medium that they're in

Lex Fridman (48:05.340)

but that's a little bit more exotic.

Lex Fridman (48:07.780)

And for the purposes of this conversation

Lex Fridman (48:10.340)

we can assume that photons don't interact with each other.

Lex Fridman (48:13.180)

So if you have a bunch of them

Lex Fridman (48:16.100)

all propagating in the same direction

Jeffrey Shainline (48:17.580)

they don't interfere with each other.

Lex Fridman (48:19.140)

If I wanna send, if I have a communication channel

Lex Fridman (48:22.900)

and I put one more photon on it,

Lex Fridman (48:24.460)

it doesn't screw up with those other ones.

Jeffrey Shainline (48:26.020)

It doesn't change what those other ones were doing at all.

Lex Fridman (48:29.060)

So that's really useful for communication

Jeffrey Shainline (48:31.260)

because that means you can sort of allow

Lex Fridman (48:33.660)

a lot of these photons to flow

Jeffrey Shainline (48:37.060)

without disruption of each other

Lex Fridman (48:38.820)

and they can branch really easily and things like that.

Lex Fridman (48:41.060)

But it's not good for computation

Lex Fridman (48:42.700)

because it's very hard for this packet of light

Jeffrey Shainline (48:46.340)

to change what this packet of light is doing.

Lex Fridman (48:48.700)

They pass right through each other.

Lex Fridman (48:50.180)

So in computation you want to change information

Lex Fridman (48:53.260)

and if photons don't interact with each other

Jeffrey Shainline (48:55.700)

it's difficult to get them to change the information

Lex Fridman (48:58.020)

represented by the others.

Lex Fridman (48:59.380)

So that's the fundamental difference.

Lex Fridman (49:01.580)

Is there also something about the way they travel

Jeffrey Shainline (49:04.780)

through different materials

Lex Fridman (49:07.460)

or is that just a particular engineering?

Jeffrey Shainline (49:10.700)

No, it's not, that's deep physics I think.

Lex Fridman (49:12.580)

So this gets back to electrons interact with each other

Lex Fridman (49:17.060)

and photons don't.

Lex Fridman (49:18.140)

So say I'm trying to get a packet of information

Jeffrey Shainline (49:22.380)

from me to you and we have a wire going between us.

Lex Fridman (49:25.820)

In order for me to send electrons across that wire

Jeffrey Shainline (49:29.020)

I first have to raise the voltage on my end of the wire

Lex Fridman (49:32.180)

and that means putting a bunch of charges on it

Lex Fridman (49:34.580)

and then that charge packet has to propagate along the wire

Lex Fridman (49:39.140)

and it has to get all the way over to you.

Jeffrey Shainline (49:41.260)

That wire is gonna have something that's called capacitance

Lex Fridman (49:44.380)

which basically tells you how much charge

Jeffrey Shainline (49:46.940)

you need to put on the wire

Lex Fridman (49:48.060)

in order to raise the voltage on it

Lex Fridman (49:49.980)

and the capacitance is gonna be proportional

Lex Fridman (49:52.500)

to the length of the wire.

Lex Fridman (49:54.060)

So the longer the length of the wire is

Lex Fridman (49:56.900)

the more charge I have to put on it

Lex Fridman (49:59.140)

and the energy required to charge up that line

Lex Fridman (50:03.060)

and move those electrons to you

Jeffrey Shainline (50:04.980)

is also proportional to the capacitance

Lex Fridman (50:06.860)

and goes as the voltage squared.

Lex Fridman (50:08.500)

So you get this huge penalty if you wanna send electrons

Lex Fridman (50:13.780)

across a wire over appreciable distances.

Lex Fridman (50:16.620)

So distance is an important thing here

Lex Fridman (50:19.140)

when you're doing communication.

Jeffrey Shainline (50:20.780)

Distance is an important thing.

Lex Fridman (50:22.140)

So is the number of connections I'm trying to make.

Jeffrey Shainline (50:25.340)

Me to you, okay one, that's not so bad.

Lex Fridman (50:27.620)

If I want to now send it to 10,000 other friends

Jeffrey Shainline (50:31.420)

then all of those wires are adding tons

Lex Fridman (50:34.380)

of extra capacitance.

Jeffrey Shainline (50:35.460)

Now not only does it take forever

Lex Fridman (50:37.660)

to put the charge on that wire

Lex Fridman (50:39.540)

and raise the voltage on all those lines

Lex Fridman (50:41.820)

but it takes a ton of power

Lex Fridman (50:43.540)

and the number 10,000 is not randomly chosen.

Lex Fridman (50:46.980)

That's roughly how many connections

Jeffrey Shainline (50:49.100)

each neuron in your brain makes.

Lex Fridman (50:50.620)

So a neuron in your brain needs to send 10,000 messages

Jeffrey Shainline (50:55.020)

every time it has something to say.

Lex Fridman (50:56.780)

You can't do that if you're trying to drive electrons

Jeffrey Shainline (51:00.100)

from here to 10,000 different places.

Lex Fridman (51:02.060)

The brain does it in a slightly different way

Jeffrey Shainline (51:03.660)

which we can discuss.

Lex Fridman (51:04.860)

How can light achieve the 10,000 connections

Lex Fridman (51:07.020)

and why is it better?

Lex Fridman (51:09.340)

In terms of like the energy use required

Jeffrey Shainline (51:12.580)

to use light for the communication of the 10,000 connections.

Lex Fridman (51:15.260)

Right, right.

Lex Fridman (51:16.100)

So now instead of trying to send electrons

Lex Fridman (51:17.700)

from me to you, I'm trying to send photons.

Lex Fridman (51:19.380)

So I can make what's called a wave guide

Lex Fridman (51:21.540)

which is just a simple piece of a material.

Jeffrey Shainline (51:25.140)

It could be glass like an optical fiber

Lex Fridman (51:27.060)

or silicon on a chip.

Lex Fridman (51:29.860)

And I just have to inject photons into that wave guide

Lex Fridman (51:34.140)

and independent of how long it is,

Jeffrey Shainline (51:35.800)

independent of how many different connections I'm making,

Lex Fridman (51:39.620)

it doesn't change the voltage or anything like that

Jeffrey Shainline (51:43.140)

that I have to raise up on the wire.

Lex Fridman (51:45.460)

So if I have one more connection,

Jeffrey Shainline (51:47.940)

if I add additional connections,

Lex Fridman (51:49.820)

I need to add more light to the wave guide

Jeffrey Shainline (51:51.760)

because those photons need to split

Lex Fridman (51:53.280)

and go to different paths.

Jeffrey Shainline (51:55.040)

That makes sense but I don't have a capacitive penalty.

Lex Fridman (51:58.860)

Sometimes these are called wiring parasitics.

Jeffrey Shainline (52:01.300)

There are no parasitics associated with light

Lex Fridman (52:03.460)

in that same sense.

Lex Fridman (52:04.420)

So this might be a dumb question

Lex Fridman (52:07.500)

but how do I catch a photon on the other end?

Lex Fridman (52:11.380)

Is it material?

Lex Fridman (52:12.500)

Is it the polymer stuff you were talking about

Lex Fridman (52:15.020)

for a different application for photolithography?

Lex Fridman (52:19.380)

Like how do you catch a photon?

Jeffrey Shainline (52:20.980)

There's a lot of ways to catch a photon.

Lex Fridman (52:22.500)

It's not a dumb question.

Jeffrey Shainline (52:23.620)

It's a deep and important question

Lex Fridman (52:25.940)

that basically defines a lot of the work

Jeffrey Shainline (52:29.180)

that goes on in our group at NIST.

Lex Fridman (52:31.380)

One of my group leaders, Seywoon Nam,

Jeffrey Shainline (52:34.260)

has built his career around

Lex Fridman (52:35.780)

these superconducting single photon detectors.

Lex Fridman (52:38.420)

So if you're going to try to sort of reach a lower limit

Lex Fridman (52:42.460)

and detect just one particle of light,

Jeffrey Shainline (52:45.260)

superconductors come back into our conversation

Lex Fridman (52:47.660)

and just picture a simple device

Jeffrey Shainline (52:50.140)

where you have current flowing

Lex Fridman (52:51.600)

through a superconducting wire and...

Lex Fridman (52:54.580)

A loop again or no?

Lex Fridman (52:56.560)

Let's say yes, you have a loop.

Lex Fridman (52:57.820)

So you have a superconducting wire

Lex Fridman (52:59.580)

that goes straight down like this

Lex Fridman (53:01.000)

and on your loop branch, you have a little ammeter,

Lex Fridman (53:04.260)

something that measures current.

Jeffrey Shainline (53:05.800)

There's a resistor up there too.

Lex Fridman (53:07.980)

Go with me here.

Lex Fridman (53:09.020)

So your current biasing this,

Lex Fridman (53:12.020)

so there's current flowing

Jeffrey Shainline (53:13.060)

through that superconducting branch.

Lex Fridman (53:14.400)

Since there's a resistor over here,

Jeffrey Shainline (53:16.560)

all the current goes through the superconducting branch.

Lex Fridman (53:18.940)

Now a photon comes in, strikes that superconductor.

Jeffrey Shainline (53:22.280)

We talked about this superconducting

Lex Fridman (53:24.300)

macroscopic quantum state.

Jeffrey Shainline (53:25.760)

That's going to be destroyed by the energy of that photon.

Lex Fridman (53:28.480)

So now that branch of the circuit is resistive too.

Lex Fridman (53:32.080)

And you've properly designed your circuit

Lex Fridman (53:33.780)

so that the resistance on that superconducting branch

Jeffrey Shainline (53:36.700)

is much greater than the other resistance.

Lex Fridman (53:38.420)

Now all of your current's going to go that way.

Jeffrey Shainline (53:40.780)

Your ammeter says, oh, I just got a pulse of current.

Lex Fridman (53:43.260)

That must mean I detected a photon.

Jeffrey Shainline (53:45.140)

Then where you broke that superconductivity

Lex Fridman (53:47.220)

in a matter of a few nanoseconds,

Jeffrey Shainline (53:49.100)

it cools back off, dissipates that energy

Lex Fridman (53:51.100)

and the current flows back

Jeffrey Shainline (53:52.820)

through that superconducting branch.

Lex Fridman (53:54.300)

This is a very powerful superconducting device

Jeffrey Shainline (53:59.260)

that allows us to understand quantum states of light.

Lex Fridman (54:02.300)

I didn't realize a loop like that

Jeffrey Shainline (54:04.880)

could be sensitive to a single photon.

Lex Fridman (54:07.160)

I mean, that seems strange to me because,

Jeffrey Shainline (54:13.060)

I mean, so what happens when you just barrage it

Lex Fridman (54:15.700)

with photons?

Jeffrey Shainline (54:16.620)

If you put a bunch of photons in there,

Lex Fridman (54:18.340)

essentially the same thing happens.

Jeffrey Shainline (54:19.860)

You just drive it into the normal state,

Lex Fridman (54:21.660)

it becomes resistive and it's not particularly interesting.

Lex Fridman (54:25.420)

So you have to be careful how many photons you send.

Lex Fridman (54:27.980)

Like you have to be very precise with your communication.

Jeffrey Shainline (54:30.100)

Well, it depends.

Lex Fridman (54:31.220)

So I would say that that's actually in the application

Jeffrey Shainline (54:34.220)

that we're trying to use these detectors for.

Lex Fridman (54:37.060)

That's a feature because what we want is for,

Jeffrey Shainline (54:41.140)

if a neuron sends one photon to a synaptic connection

Lex Fridman (54:46.660)

and one of these superconducting detectors is sitting there,

Jeffrey Shainline (54:49.860)

you get this pulse of current.

Lex Fridman (54:51.200)

And that synapse says event,

Jeffrey Shainline (54:54.000)

then I'm gonna do what I do when there's a synapse event,

Lex Fridman (54:56.020)

I'm gonna perform computations, that kind of thing.

Lex Fridman (54:58.660)

But if accidentally you send two there or three or five,

Lex Fridman (55:02.100)

it does the exact same.

Jeffrey Shainline (55:03.380)

Got it.

Lex Fridman (55:04.200)

And so this is how in the system that we're devising here,

Jeffrey Shainline (55:10.040)

communication is entirely binary.

Lex Fridman (55:12.740)

And that's what I tried to emphasize a second ago.

Jeffrey Shainline (55:15.020)

Communication should not change the information.

Lex Fridman (55:17.780)

You're not saying, oh, I got this kind of communication

Jeffrey Shainline (55:21.300)

event for photons.

Lex Fridman (55:22.280)

No, we're not keeping track of that.

Jeffrey Shainline (55:23.700)

This neuron fired, this synapse says that neuron fired,

Lex Fridman (55:26.580)

that's it.

Lex Fridman (55:27.420)

So that's a noise filtering property of those detectors.

Lex Fridman (55:31.460)

However, there are other applications

Jeffrey Shainline (55:33.140)

where you'd rather know the exact number of photons

Lex Fridman (55:36.140)

that can be very useful in quantum computing with light.

Lex Fridman (55:39.300)

And our group does a lot of work

Lex Fridman (55:41.820)

around another kind of superconducting sensor

Jeffrey Shainline (55:44.580)

called a transition edge sensor that Adrian Alita

Lex Fridman (55:48.260)

in our group does a lot of work on that.

Lex Fridman (55:49.940)

And that can tell you based on the amplitude

Lex Fridman (55:53.980)

of the current pulse you divert exactly how many photons

Jeffrey Shainline (55:58.180)

were in that pulse.

Lex Fridman (56:00.900)

What's that useful for?

Jeffrey Shainline (56:02.500)

One way that you can encode information

Lex Fridman (56:04.700)

in quantum states of light is in the number of photons.

Jeffrey Shainline (56:07.460)

You can have what are called number states

Lex Fridman (56:09.300)

and a number state will have a well defined number

Jeffrey Shainline (56:11.980)

of photons and maybe the output of your quantum computation

Lex Fridman (56:16.400)

encodes its information in the number of photons

Jeffrey Shainline (56:19.780)

that are generated.

Lex Fridman (56:20.620)

So if you have a detector that is sensitive to that,

Jeffrey Shainline (56:23.020)

it's extremely useful.

Lex Fridman (56:24.300)

Can you achieve like a clock with photons

Lex Fridman (56:29.240)

or is that not important?

Lex Fridman (56:30.260)

Is there a synchronicity here?

Jeffrey Shainline (56:33.300)

In general, it can be important.

Lex Fridman (56:36.880)

Clock distribution is a big challenge

Jeffrey Shainline (56:39.380)

in especially large computational systems.

Lex Fridman (56:43.300)

And so yes, optical clocks, optical clock distribution

Jeffrey Shainline (56:47.940)

is a very powerful technology.

Lex Fridman (56:51.140)

I don't know the state of that field right now,

Lex Fridman (56:53.180)

but I imagine that if you're trying to distribute a clock

Lex Fridman (56:55.620)

across any appreciable size computational system,

Jeffrey Shainline (56:58.960)

you wanna use light.

Lex Fridman (57:00.340)

Yeah, I wonder how these giant systems work,

Jeffrey Shainline (57:04.300)

especially like supercomputers.

Lex Fridman (57:07.380)

Do they need to do clock distribution

Jeffrey Shainline (57:09.380)

or are they doing more ad hoc parallel

Lex Fridman (57:14.260)

like concurrent programming?

Jeffrey Shainline (57:15.540)

Like there's some kind of locking mechanisms or something.

Lex Fridman (57:18.140)

That's a fascinating question,

Lex Fridman (57:19.320)

but let's zoom in at this very particular question

Lex Fridman (57:23.900)

of computation on a processor

Lex Fridman (57:28.200)

and communication between processors.

Lex Fridman (57:31.560)

So what does this system look like

Lex Fridman (57:36.440)

that you're envisioning?

Lex Fridman (57:38.220)

One of the places you're envisioning it

Jeffrey Shainline (57:40.100)

is in the paper on optoelectronic intelligence.

Lex Fridman (57:43.140)

So what are we talking about?

Jeffrey Shainline (57:44.740)

Are we talking about something

Lex Fridman (57:46.300)

that starts to look a lot like the human brain

Lex Fridman (57:48.740)

or does it still look a lot like a computer?

Lex Fridman (57:51.300)

What are the size of this thing?

Jeffrey Shainline (57:52.980)

Is it going inside a smartphone or as you said,

Lex Fridman (57:55.140)

does it go inside something that's more like a house?

Lex Fridman (57:58.580)

Like what should we be imagining?

Lex Fridman (58:01.180)

What are you thinking about

Lex Fridman (58:02.260)

when you're thinking about these fundamental systems?

Lex Fridman (58:05.460)

Let me introduce the word neuromorphic.

Jeffrey Shainline (58:07.380)

There's this concept of neuromorphic computing

Lex Fridman (58:09.960)

where what that broadly refers to

Jeffrey Shainline (58:12.580)

is computing based on the information processing principles

Lex Fridman (58:17.740)

of the brain.

Lex Fridman (58:19.260)

And as digital computing seems to be pushing

Lex Fridman (58:23.820)

towards some fundamental performance limits,

Jeffrey Shainline (58:26.480)

people are considering architectural advances,

Lex Fridman (58:29.180)

drawing inspiration from the brain,

Jeffrey Shainline (58:30.860)

more distributed parallel network kind of architectures

Lex Fridman (58:33.660)

and stuff.

Lex Fridman (58:34.500)

And so there's this continuum of neuromorphic

Lex Fridman (58:37.420)

from things that are pretty similar to digital computers,

Lex Fridman (58:42.740)

but maybe there are more cores

Lex Fridman (58:45.720)

and the way they send messages is a little bit more

Jeffrey Shainline (58:49.180)

like the way brain neurons send spikes.

Lex Fridman (58:52.780)

But for the most part, it's still digital electronics.

Lex Fridman (58:56.100)

And then you have some things in between

Lex Fridman (58:58.700)

where maybe you're using transistors,

Lex Fridman (59:02.060)

but now you're starting to use them

Lex Fridman (59:03.220)

instead of in a digital way, in an analog way.

Lex Fridman (59:06.140)

And so you're trying to get those circuits

Lex Fridman (59:08.180)

to behave more like neurons.

Lex Fridman (59:10.320)

And then that's a little bit,

Lex Fridman (59:12.140)

quite a bit more on the neuromorphic side of things.

Jeffrey Shainline (59:14.940)

You're trying to get your circuits,

Lex Fridman (59:17.080)

although they're still based on silicon,

Jeffrey Shainline (59:19.140)

you're trying to get them to perform operations

Lex Fridman (59:22.480)

that are highly analogous to the operations in the brain.

Lex Fridman (59:24.660)

And that's where a great deal of work is

Lex Fridman (59:26.700)

in neuromorphic computing,

Jeffrey Shainline (59:27.700)

people like Giacomo Indoveri and Gert Kauenberg,

Lex Fridman (59:30.920)

Jennifer Hasler, countless others.

Jeffrey Shainline (59:32.980)

It's a rich and exciting field going back to Carver Mead

Lex Fridman (59:36.900)

in the late 1980s.

Lex Fridman (59:39.460)

And then all the way on the other extreme of the continuum

Lex Fridman (59:44.240)

is where you say, I'll give up anything related

Jeffrey Shainline (59:48.580)

to transistors or semiconductors or anything like that.

Lex Fridman (59:51.420)

I'm not starting with the assumption

Jeffrey Shainline (59:53.620)

that I'm gonna use any kind

Lex Fridman (59:55.060)

of conventional computing hardware.

Lex Fridman (59:57.020)

And instead, what I wanna do is try and understand

Lex Fridman (59:59.300)

what makes the brain powerful