Dmitry Korkin: Evolution of Proteins, Viruses, Life, and AI
生物与进化音乐与艺术AI 与机器学习技术与编程心理与人性
📋 章节目录
暂无章节信息
🔑 关键词
proteinproteinsviruslearningdonessentiallymachinedomainsstructurefunctioninterestingbookevolutioncalleddataperspectiveideasspikemutationssequence
💬 精彩语录
暂无语录
🎙️ 完整对话（2430 条）
Lex Fridman (00:00.000)
The following is a conversation with Dmitry Korkin,
以下是与德米特里·科尔金的对话，
Lex Fridman (00:02.860)
his second time in the podcast.
他第二次参加播客。
Lex Fridman (00:04.820)
He's a professor of bioinformatics
他是一位生物信息学教授
Lex Fridman (00:06.980)
and computational biology at WPI,
和 WPI 的计算生物学，
Lex Fridman (00:09.740)
where he specializes in bioinformatics of complex disease,
他专门研究复杂疾病的生物信息学，
Dmitry Korkin (00:13.540)
computational genomics, systems biology,
计算基因组学、系统生物学、
Lex Fridman (00:16.300)
and biomedical data analytics.
和生物医学数据分析。
Dmitry Korkin (00:18.540)
He loves biology, he loves computing,
他热爱生物学，他热爱计算，
Lex Fridman (00:22.080)
plus he is Russian and recites a poem in Russian
而且他是俄罗斯人并用俄语背诵一首诗
Dmitry Korkin (00:26.140)
at the end of the podcast.
在播客的最后。
Lex Fridman (00:27.760)
What else could you possibly ask for in this world?
在这个世界上你还能要求什么呢？
Dmitry Korkin (00:31.080)
Quick mention of our sponsors.
快速提及我们的赞助商。
Lex Fridman (00:32.960)
Brave Browser, NetSuite Business Management Software,
Brave 浏览器、NetSuite 商业管理软件、
Dmitry Korkin (00:37.760)
Magic Spoon Low Carb Cereal,
魔勺低碳水化合物谷物，
Lex Fridman (00:40.280)
and 8sleep Self Cooling Mattress.
和 8sleep 自冷床垫。
Lex Fridman (00:42.920)
So the choice is browsing privacy, business success,
所以选择是浏览隐私、商业成功、
Lex Fridman (00:46.360)
healthy diet, or comfortable sleep.
健康的饮食，或舒适的睡眠。
Dmitry Korkin (00:49.180)
Choose wisely, my friends,
明智的选择，我的朋友们，
Lex Fridman (00:50.660)
and if you wish, click the sponsor links below
如果您愿意，请点击下面的赞助商链接
Dmitry Korkin (00:53.640)
to get a discount and to support this podcast.
获得折扣并支持此播客。
Lex Fridman (00:56.440)
As a side note, let me say that to me,
Dmitry Korkin (00:58.600)
the scientists that did the best apolitical,
Lex Fridman (01:01.520)
impactful, brilliant work of 2020
Dmitry Korkin (01:04.000)
are the biologists who study viruses without an agenda,
Lex Fridman (01:09.160)
without much sleep, to be honest,
Dmitry Korkin (01:11.800)
just a pure passion for scientific discovery
Lex Fridman (01:14.460)
and exploration of the mysteries within viruses.
Dmitry Korkin (01:18.400)
Viruses are both terrifying and beautiful.
Lex Fridman (01:21.340)
Terrifying because they can threaten
Dmitry Korkin (01:22.960)
the fabric of human civilization,
Lex Fridman (01:25.120)
both biological and psychological.
Dmitry Korkin (01:27.840)
Beautiful because they give us insights
Lex Fridman (01:30.480)
into the nature of life on Earth
Lex Fridman (01:32.920)
and perhaps even extraterrestrial life
Lex Fridman (01:35.900)
of the not so intelligent variety
Dmitry Korkin (01:37.960)
that might meet us one day
Lex Fridman (01:39.560)
as we explore the habitable planets
Lex Fridman (01:41.520)
and moons in our universe.
Lex Fridman (01:43.760)
If you enjoy this thing, subscribe on YouTube,
Dmitry Korkin (01:45.800)
review it on Apple Podcast, follow on Spotify,
Lex Fridman (01:49.040)
support on Patreon, or connect with me on Twitter
Dmitry Korkin (01:51.740)
at Lex Friedman.
Lex Fridman (01:53.160)
And now here's my conversation with Dmitry Korkin.
Dmitry Korkin (01:57.920)
It's often said that proteins
Lex Fridman (02:00.680)
and the amino acid residues that make them up
Dmitry Korkin (02:04.200)
are the building blocks of life.
Lex Fridman (02:06.400)
Do you think of proteins in this way
Lex Fridman (02:08.000)
as the basic building blocks of life?
Lex Fridman (02:11.160)
Yes and no.
Lex Fridman (02:12.200)
So the proteins indeed is the basic unit,
Lex Fridman (02:16.300)
biological unit that carries out
Dmitry Korkin (02:20.480)
important function of the cell.
Lex Fridman (02:22.800)
However, through studying the proteins
Lex Fridman (02:25.800)
and comparing the proteins across different species,
Lex Fridman (02:29.360)
across different kingdoms,
Dmitry Korkin (02:31.440)
you realize that proteins are actually
Lex Fridman (02:34.640)
much more complicated.
Lex Fridman (02:36.760)
So they have so called modular complexity.
Lex Fridman (02:42.280)
And so what I mean by that is an average protein
Dmitry Korkin (02:47.280)
consists of several structural units.
Lex Fridman (02:54.760)
So we call them protein domains.
Lex Fridman (02:57.440)
And so you can imagine a protein as a string of beads
Lex Fridman (03:02.580)
where each bead is a protein domain.
Lex Fridman (03:05.760)
And in the past 20 years,
Lex Fridman (03:10.240)
scientists have been studying
Dmitry Korkin (03:13.040)
the nature of the protein domains
Lex Fridman (03:15.040)
because we realize that it's the unit.
Lex Fridman (03:19.480)
Because if you look at the functions, right?
Lex Fridman (03:22.120)
So many proteins have more than one function
Lex Fridman (03:25.880)
and those protein functions are often carried out
Lex Fridman (03:29.440)
by those protein domains.
Lex Fridman (03:31.560)
So we also see that in the evolution,
Lex Fridman (03:37.320)
those proteins domains get shuffled.
Lex Fridman (03:40.160)
So they act actually as a unit.
Lex Fridman (03:43.460)
Also from the structural perspective, right?
Lex Fridman (03:45.280)
So some people think of a protein
Lex Fridman (03:50.960)
as a sort of a globular molecule,
Lex Fridman (03:55.320)
but as a matter of fact,
Lex Fridman (03:56.800)
is the globular part of this protein is a protein domain.
Lex Fridman (04:02.500)
So we often have this, again,
Lex Fridman (04:06.000)
the collection of this protein domains
Dmitry Korkin (04:09.600)
align on a string as beads.
Lex Fridman (04:14.800)
And the protein domains are made up of amino acid residue.
Lex Fridman (04:17.880)
So we're talking about.
Lex Fridman (04:18.720)
So this is the basic,
Lex Fridman (04:20.640)
so you're saying the protein domain
Lex Fridman (04:22.600)
is the basic building block of the function
Dmitry Korkin (04:25.640)
that we think about proteins doing.
Lex Fridman (04:28.320)
So of course you can always talk
Dmitry Korkin (04:30.200)
about different building blocks.
Lex Fridman (04:31.520)
It's turtles all the way down.
Lex Fridman (04:32.880)
But there's a point where there is,
Lex Fridman (04:36.280)
at the point of the hierarchy
Dmitry Korkin (04:37.680)
where it's the most, the cleanest element block
Lex Fridman (04:43.640)
based on which you can put them together
Dmitry Korkin (04:46.240)
in different kinds of ways to form complex function.
Lex Fridman (04:49.200)
And you're saying protein domains,
Lex Fridman (04:50.880)
why is that not talked about as often in popular culture?
Lex Fridman (04:55.160)
Well, there are several perspectives on this.
Lex Fridman (04:59.280)
And one of course is the historical perspective, right?
Lex Fridman (05:03.200)
So historically scientists have been able
Dmitry Korkin (05:07.800)
to structurally resolved
Lex Fridman (05:10.120)
to obtain the 3D coordinates of a protein
Dmitry Korkin (05:14.240)
for smaller proteins.
Lex Fridman (05:17.520)
And smaller proteins tend to be a single domain protein.
Lex Fridman (05:21.000)
So we have a protein equal to a protein domain.
Lex Fridman (05:24.000)
And so because of that,
Dmitry Korkin (05:26.040)
the initial suspicion was that the proteins are,
Lex Fridman (05:29.640)
they have globular shapes
Lex Fridman (05:31.680)
and the more of smaller proteins you obtain structurally,
Lex Fridman (05:36.840)
the more you became convinced that that's the case.
Lex Fridman (05:41.720)
And only later when we started having
Lex Fridman (05:47.920)
alternative approaches.
Lex Fridman (05:49.640)
So the traditional ones are X ray crystallography
Lex Fridman (05:55.920)
and NMR spectroscopy.
Lex Fridman (05:57.320)
So this is sort of the two main techniques
Lex Fridman (06:02.000)
that give us the 3D coordinates.
Lex Fridman (06:04.440)
But nowadays there's huge breakthrough
Lex Fridman (06:07.760)
in cryo electron microscopy.
Lex Fridman (06:10.480)
So the more advanced methods that allow us
Lex Fridman (06:13.840)
to get into the 3D shapes of much larger molecules,
Dmitry Korkin (06:21.560)
molecular complexes,
Lex Fridman (06:23.480)
just to give you one of the common examples
Lex Fridman (06:28.120)
for this year, right?
Lex Fridman (06:29.440)
So the first experimental structure
Dmitry Korkin (06:32.760)
of a SARS COVID 2 protein
Lex Fridman (06:35.920)
was the cryo EM structure of the S protein.
Lex Fridman (06:40.160)
So the spike protein.
Lex Fridman (06:41.960)
And so it was solved very quickly.
Lex Fridman (06:46.320)
And the reason for that is the advancement
Lex Fridman (06:49.480)
of this technology is pretty spectacular.
Lex Fridman (06:53.920)
How many domains does the, is it more than one domain?
Lex Fridman (06:57.480)
Oh yes.
Dmitry Korkin (06:58.320)
Oh yes, I mean, so it's a very complex structure.
Lex Fridman (07:01.320)
And we, you know, on top of the complexity
Lex Fridman (07:06.480)
of a single protein, right?
Lex Fridman (07:08.520)
So this structure is actually is a complex, is a trimer.
Lex Fridman (07:13.720)
So it needs to form a trimer in order to function properly.
Lex Fridman (07:17.640)
What's a complex?
Lex Fridman (07:18.720)
So a complex is a glomeration of multiple proteins.
Lex Fridman (07:22.880)
And so we can have the same protein copied in multiple,
Dmitry Korkin (07:29.280)
you know, made up in multiple copies
Lex Fridman (07:32.080)
and forming something that we called a homo oligomer.
Lex Fridman (07:36.160)
Homo means the same, right?
Lex Fridman (07:38.120)
So in this case, so the spike protein is the,
Dmitry Korkin (07:42.800)
is an example of a homo tetram, homo trimer, sorry.
Lex Fridman (07:46.720)
So you need three copies of it?
Dmitry Korkin (07:48.120)
Three copies.
Lex Fridman (07:48.960)
In order to.
Dmitry Korkin (07:50.040)
Exactly.
Lex Fridman (07:50.880)
We have these three chains,
Dmitry Korkin (07:52.760)
the three molecular chains coupled together
Lex Fridman (07:56.800)
and performing the function.
Dmitry Korkin (07:58.480)
That's what, when you look at this protein from the top,
Lex Fridman (08:02.380)
you see a perfect triangle.
Dmitry Korkin (08:03.920)
Yeah.
Lex Fridman (08:04.760)
So, but other, you know,
Lex Fridman (08:07.120)
so other complexes are made up of, you know,
Lex Fridman (08:10.640)
different proteins.
Dmitry Korkin (08:12.840)
Some of them are completely different.
Lex Fridman (08:15.400)
Some of them are similar.
Lex Fridman (08:16.920)
The hemoglobin molecule, right?
Lex Fridman (08:18.880)
So it's actually, it's a protein complex.
Dmitry Korkin (08:21.880)
It's made of four basic subunits.
Lex Fridman (08:25.760)
Two of them are identical to each other.
Dmitry Korkin (08:29.040)
Two other identical to each other,
Lex Fridman (08:30.800)
but they are also similar to each other,
Dmitry Korkin (08:32.820)
which sort of gives us some ideas about the evolution
Lex Fridman (08:36.960)
of this, you know, of this molecule.
Lex Fridman (08:40.640)
And perhaps, so one of the hypothesis is that, you know,
Lex Fridman (08:44.000)
in the past, it was just a homo tetramer, right?
Lex Fridman (08:48.280)
So four identical copies,
Lex Fridman (08:50.840)
and then it became, you know, sort of modified,
Dmitry Korkin (08:55.520)
it became mutated over the time
Lex Fridman (08:58.520)
and became more specialized.
Lex Fridman (09:00.160)
Can we linger on the spike protein for a little bit?
Lex Fridman (09:02.560)
Is there something interesting
Lex Fridman (09:04.940)
or like beautiful you find about it?
Lex Fridman (09:06.960)
I mean, first of all,
Dmitry Korkin (09:07.880)
it's an incredibly challenging protein.
Lex Fridman (09:10.960)
And so we, as a part of our sort of research
Dmitry Korkin (09:16.120)
to understand the structural basis of this virus,
Lex Fridman (09:20.200)
to sort of decode, structurally decode,
Dmitry Korkin (09:22.760)
every single protein in its proteome,
Lex Fridman (09:27.560)
which, you know, we've been working on this spike protein.
Lex Fridman (09:31.800)
And one of the main challenges was that the cryoEM data
Lex Fridman (09:36.800)
allows us to reconstruct or to obtain the 3D coordinates
Dmitry Korkin (09:44.640)
of roughly two thirds of the protein.
Lex Fridman (09:48.040)
The rest of the one third of this protein,
Dmitry Korkin (09:51.960)
it's a part that is buried into the membrane of the virus
Lex Fridman (09:58.400)
and of the viral envelope.
Lex Fridman (10:01.560)
And it also has a lot of unstable structures around it.
Lex Fridman (10:06.920)
So it's chemically interacting somehow
Dmitry Korkin (10:08.640)
with whatever the hex is connecting to.
Lex Fridman (10:10.160)
Yeah, so people are still trying to understand.
Lex Fridman (10:12.800)
So the nature of, and the role of this one third,
Lex Fridman (10:18.600)
because the top part, you know, the primary function
Dmitry Korkin (10:23.120)
is to get attached to the ACE2 receptor, human receptor.
Lex Fridman (10:28.120)
There is also beautiful mechanics
Lex Fridman (10:32.600)
of how this thing happens, right?
Lex Fridman (10:34.720)
So because there are three different copies of this chains,
Lex Fridman (10:39.800)
you know, there are three different domains, right?
Lex Fridman (10:43.480)
So we're talking about domains.
Lex Fridman (10:44.800)
So this is the receptor binding domains, RBDs,
Lex Fridman (10:47.840)
that gets untangled and get ready to get attached
Dmitry Korkin (10:53.840)
to the receptor.
Lex Fridman (10:55.400)
And now they are not necessarily going in a sync mode.
Dmitry Korkin (11:02.760)
As a matter of fact.
Lex Fridman (11:04.080)
It's asynchronous.
Lex Fridman (11:05.240)
So yes, and this is where another level of complexity
Lex Fridman (11:11.000)
comes into play because right now what we see is,
Dmitry Korkin (11:16.000)
we typically see just one of the arms going out
Lex Fridman (11:20.520)
and getting ready to be attached to the ACE2 receptors.
Dmitry Korkin (11:27.560)
However, there was a recent mutation
Lex Fridman (11:30.360)
that people studied in that spike protein.
Lex Fridman (11:35.080)
And very recently, a group from UMass Medical School
Lex Fridman (11:43.560)
will happen to collaborate with groups.
Lex Fridman (11:45.280)
So this is a group of Jeremy Lubin
Lex Fridman (11:47.240)
and a number of other faculty.
Dmitry Korkin (11:51.560)
They actually solve the mutated structure of the spike.
Lex Fridman (11:59.000)
And they showed that actually, because of these mutations,
Dmitry Korkin (12:03.000)
you have more than one arms opening up.
Lex Fridman (12:08.880)
And so now, so the frequency of two arms going up
Dmitry Korkin (12:13.880)
increase quite drastically.
Lex Fridman (12:17.200)
Interesting.
Lex Fridman (12:18.040)
Does that change the dynamics somehow?
Lex Fridman (12:20.120)
It potentially can change the dynamics
Dmitry Korkin (12:22.920)
because now you have two possible opportunities
Lex Fridman (12:27.280)
to get attached to the ACE2 receptor.
Dmitry Korkin (12:30.000)
It's a very complex molecular process, mechanistic process.
Lex Fridman (12:34.120)
But the first step of this process is the attachment
Dmitry Korkin (12:38.280)
of this spike protein, of the spike trimer
Lex Fridman (12:42.560)
to the human ACE2 receptor.
Lex Fridman (12:46.600)
So this is a molecule that sits
Lex Fridman (12:48.880)
on the surface of the human cell.
Lex Fridman (12:51.920)
And that's essentially what initiates,
Lex Fridman (12:54.720)
what triggers the whole process of encapsulation.
Dmitry Korkin (12:58.880)
If this was dating, this would be the first date.
Lex Fridman (13:01.440)
So this is the...
Dmitry Korkin (13:03.160)
In a way.
Lex Fridman (13:04.200)
Yes.
Lex Fridman (13:05.640)
So is it possible to have the spike protein
Lex Fridman (13:07.920)
just like floating about on its own?
Lex Fridman (13:10.640)
Or does it need that interactability with the membrane?
Lex Fridman (13:14.680)
Yeah, so it needs to be attached,
Dmitry Korkin (13:16.920)
at least as far as I know.
Lex Fridman (13:19.040)
But when you get this thing attached on the surface,
Dmitry Korkin (13:23.320)
there is also a lot of dynamics
Lex Fridman (13:25.120)
on how it sits on the surface.
Lex Fridman (13:28.200)
So for example, there was a recent work in,
Lex Fridman (13:32.200)
again, where people use the cryolectron microscopy
Dmitry Korkin (13:35.800)
to get the first glimpse of the overall structure.
Lex Fridman (13:38.960)
It's a very low res, but you still get
Dmitry Korkin (13:41.600)
some interesting details about the surface,
Lex Fridman (13:45.160)
about what is happening inside,
Dmitry Korkin (13:47.040)
because we have literally no clue until recent work
Lex Fridman (13:50.760)
about how the capsid is organized.
Lex Fridman (13:54.520)
What's a capsid?
Lex Fridman (13:55.360)
So a capsid is essentially,
Dmitry Korkin (13:56.720)
it's the inner core of the viral particle
Lex Fridman (14:01.040)
where there is the RNA of the virus,
Lex Fridman (14:05.000)
and it's protected by another protein, N protein,
Lex Fridman (14:10.280)
that essentially acts as a shield.
Lex Fridman (14:13.440)
But now we are learning more and more,
Lex Fridman (14:16.520)
so it's actually, it's not just this shield,
Dmitry Korkin (14:18.600)
it potentially is used for the stability
Lex Fridman (14:21.800)
of the outer shell of the virus.
Lex Fridman (14:25.040)
So it's pretty complicated.
Lex Fridman (14:27.840)
And I mean, understanding all of this is really useful
Dmitry Korkin (14:30.480)
for trying to figure out like developing a vaccine
Lex Fridman (14:33.000)
or some kind of drug to attack,
Lex Fridman (14:34.680)
any aspects of this, right?
Lex Fridman (14:36.040)
So, I mean, there are many different implications to that.
Dmitry Korkin (14:39.640)
First of all, it's important to understand
Lex Fridman (14:43.040)
the virus itself, right?
Lex Fridman (14:44.560)
So in order to understand how it acts,
Lex Fridman (14:51.560)
what is the overall mechanistic process
Dmitry Korkin (14:55.320)
of this virus replication,
Lex Fridman (14:57.320)
of this virus proliferation to the cell, right?
Lex Fridman (15:00.560)
So that's one aspect.
Lex Fridman (15:03.000)
The other aspect is designing new treatments.
Lex Fridman (15:06.480)
So one of the possible treatments
Lex Fridman (15:09.040)
is designing nanoparticles.
Lex Fridman (15:12.480)
And so some nanoparticles that will resemble the viral shape
Lex Fridman (15:17.200)
that would have the spike integrated,
Lex Fridman (15:19.520)
and essentially would act as a competitor to the real virus
Lex Fridman (15:23.680)
by blocking the ACE2 receptors,
Lex Fridman (15:26.680)
and thus preventing the real virus entering the cell.
Lex Fridman (15:30.400)
Now, there are also, you know,
Dmitry Korkin (15:32.920)
there is a very interesting direction
Lex Fridman (15:35.600)
in looking at the membrane,
Dmitry Korkin (15:38.320)
at the envelope portion of the protein
Lex Fridman (15:40.960)
and attacking its M protein.
Lex Fridman (15:44.880)
So there are, you know, to give you a, you know,
Lex Fridman (15:48.320)
sort of a brief overview,
Dmitry Korkin (15:50.120)
there are four structural proteins.
Lex Fridman (15:52.320)
These are the proteins that made up
Dmitry Korkin (15:54.560)
a structure of the virus.
Lex Fridman (15:58.160)
So SPIKE, S protein that acts as a trimer,
Lex Fridman (16:02.920)
so it needs three copies.
Lex Fridman (16:06.080)
E, envelope protein that acts as a pantomime,
Lex Fridman (16:09.720)
so it needs five copies to act properly.
Lex Fridman (16:13.400)
M is a membrane protein, it forms dimers,
Lex Fridman (16:18.600)
and actually it forms beautiful lattice.
Lex Fridman (16:20.560)
And this is something that we've been studying
Lex Fridman (16:22.480)
and we are seeing it in simulations.
Lex Fridman (16:24.520)
It actually forms a very nice grid
Dmitry Korkin (16:26.920)
or, you know, threads, you know,
Lex Fridman (16:30.600)
of different dimers attached next to each other.
Dmitry Korkin (16:33.600)
Just a bunch of copies of each other,
Lex Fridman (16:34.960)
and they naturally, when you have a bunch of copies
Dmitry Korkin (16:36.960)
of each other, they form an interesting lattice.
Lex Fridman (16:38.960)
Exactly.
Lex Fridman (16:39.800)
And, you know, if you think about this, right?
Lex Fridman (16:42.280)
So this complex, you know, the viral shape
Lex Fridman (16:48.160)
needs to be organized somehow, self organized somehow, right?
Lex Fridman (16:52.160)
So it, you know, if it was a completely random process,
Dmitry Korkin (16:56.160)
you know, you probably wouldn't have the envelope shell
Lex Fridman (17:02.080)
of the ellipsoid shape, you know,
Dmitry Korkin (17:03.920)
you would have something, you know,
Lex Fridman (17:05.880)
pretty random, right, shape.
Lex Fridman (17:07.600)
So there is some, you know, regularity
Lex Fridman (17:10.560)
in how this, you know, how this M dimers
Dmitry Korkin (17:16.720)
get to attach to each other
Lex Fridman (17:18.480)
in a very specific directed way.
Lex Fridman (17:20.520)
Is that understood at all?
Lex Fridman (17:23.080)
It's not understood.
Dmitry Korkin (17:24.280)
We are now, we've been working in the past six months
Lex Fridman (17:28.400)
since, you know, we met, actually,
Dmitry Korkin (17:30.160)
this is where we started working on trying to understand
Lex Fridman (17:33.400)
the overall structure of the envelope
Lex Fridman (17:36.280)
and the key components that made up this, you know,
Lex Fridman (17:40.640)
structure.
Dmitry Korkin (17:41.480)
Wait, does the envelope also have the lattice structure
Lex Fridman (17:43.240)
or no?
Lex Fridman (17:44.080)
So the envelope is essentially is the outer shell
Lex Fridman (17:47.360)
of the viral particle.
Dmitry Korkin (17:48.800)
The N, the nucleocapsid protein,
Lex Fridman (17:51.600)
is something that is inside.
Dmitry Korkin (17:53.960)
Got it.
Lex Fridman (17:54.800)
But get that, the N is likely to interact with M.
Lex Fridman (17:59.520)
Does it go M and E?
Lex Fridman (18:01.480)
Like, where's the E and the M?
Lex Fridman (18:02.880)
So E, those different proteins,
Lex Fridman (18:05.640)
they occur in different copies on the viral particle.
Lex Fridman (18:10.800)
So E, this pentamer complex,
Lex Fridman (18:13.960)
we only have two or three, maybe, per each particle, okay?
Dmitry Korkin (18:18.960)
We have thousand or so of M dimers
Lex Fridman (18:24.520)
that essentially made up,
Dmitry Korkin (18:26.600)
that makes up the entire, you know, outer shell.
Lex Fridman (18:30.920)
So most of the outer shell is the M.
Dmitry Korkin (18:33.680)
M dimer.
Lex Fridman (18:34.520)
And the M protein.
Dmitry Korkin (18:35.640)
When you say particle, that's the virion,
Lex Fridman (18:38.160)
the virus, the individual virus.
Dmitry Korkin (18:40.120)
It's a single, yes.
Lex Fridman (18:40.960)
Single element of the virus, it's a single virus.
Dmitry Korkin (18:43.640)
Single virus, right.
Lex Fridman (18:45.080)
And we have about, you know, roughly 50 to 90 spike trimmers.
Lex Fridman (18:50.840)
Right?
Lex Fridman (18:51.680)
So when you, you know, when you show a...
Dmitry Korkin (18:54.000)
Per virus particle.
Lex Fridman (18:55.040)
Per virus particle.
Lex Fridman (18:56.560)
Sorry, what did you say, 50 to 90?
Lex Fridman (18:58.680)
50 to 90, right?
Lex Fridman (19:00.680)
So this is how this thing is organized.
Lex Fridman (19:04.000)
And so now, typically, right,
Lex Fridman (19:06.400)
so you see these, the antibodies that target,
Lex Fridman (19:11.480)
you know, spike protein,
Dmitry Korkin (19:13.240)
certain parts of the spike protein,
Lex Fridman (19:15.200)
but there could be some, also some treatments, right?
Lex Fridman (19:17.960)
So these are, you know, these are small molecules
Lex Fridman (19:22.000)
that bind strategic parts of these proteins,
Dmitry Korkin (19:27.520)
disrupting its function.
Lex Fridman (19:29.680)
So one of the promising directions,
Dmitry Korkin (19:34.040)
it's one of the newest directions,
Lex Fridman (19:35.600)
is actually targeting the M dimer of the protein.
Dmitry Korkin (19:40.600)
Targeting the proteins that make up this outer shell.
Lex Fridman (19:44.120)
Because if you're able to destroy the outer shell,
Dmitry Korkin (19:47.640)
you're essentially destroying the viral particle itself.
Lex Fridman (19:52.160)
So preventing it from, you know, functioning at all.
Lex Fridman (19:56.720)
So that's, you think is,
Lex Fridman (19:59.160)
from a sort of cyber security perspective,
Lex Fridman (1:00:00.260)
And so I met him, you know,
Lex Fridman (1:00:02.460)
and had a very short conversation, you know.
Lex Fridman (1:00:06.340)
But so I started, you know, reading about Dendral
Lex Fridman (1:00:10.380)
and I was amazed, you know, it's,
Lex Fridman (1:00:12.660)
we're talking about 1960, right?
Lex Fridman (1:00:16.100)
The ideas were so profound.
Lex Fridman (1:00:19.300)
Well, what's the fun about the ideas of it?
Lex Fridman (1:00:21.140)
The reason to make this is even crazier.
Dmitry Korkin (1:00:25.020)
So, Lederberg wanted to make a system
Lex Fridman (1:00:29.860)
that would help him study the extraterrestrial molecules,
Lex Fridman (1:00:38.220)
right?
Lex Fridman (1:00:39.060)
So, the idea was that, you know,
Dmitry Korkin (1:00:40.980)
the way you study the extraterrestrial molecules
Lex Fridman (1:00:43.420)
is you do the mass spec analysis, right?
Lex Fridman (1:00:46.780)
And so the mass spec gives you sort of bits,
Lex Fridman (1:00:49.700)
numbers about essentially gives you the ideas
Dmitry Korkin (1:00:52.620)
about the possible fragments or, you know,
Lex Fridman (1:00:55.900)
atoms, you know, and maybe a little fragments,
Lex Fridman (1:00:59.820)
pieces of this molecule that make up the molecule, right?
Lex Fridman (1:01:03.620)
So now you need to sort of,
Dmitry Korkin (1:01:06.060)
to decompose this information
Lex Fridman (1:01:09.180)
and to figure out what was the hole
Lex Fridman (1:01:12.460)
before it became fragments, bits and pieces, right?
Lex Fridman (1:01:17.660)
So, in order to make this, you know,
Dmitry Korkin (1:01:20.860)
to have this tool, the idea of Lederberg
Lex Fridman (1:01:25.660)
was to connect chemistry, computer science,
Lex Fridman (1:01:32.060)
and to design this so called expert system
Lex Fridman (1:01:36.100)
that looks, that takes into account,
Dmitry Korkin (1:01:38.180)
that takes as an input the mass spec data,
Lex Fridman (1:01:42.180)
the possible database of possible molecules
Lex Fridman (1:01:47.980)
and essentially try to sort of induce the molecule
Lex Fridman (1:01:52.660)
that would correspond to this spectra
Dmitry Korkin (1:01:55.580)
or, you know, essentially what this project ended up being
Lex Fridman (1:02:03.060)
was that, you know, it would provide a list of candidates
Dmitry Korkin (1:02:07.100)
that then a chemist would look at and make final decision.
Lex Fridman (1:02:11.940)
So.
Lex Fridman (1:02:12.780)
But the original idea, I suppose,
Lex Fridman (1:02:13.980)
is to solve the entirety of this problem automatically.
Dmitry Korkin (1:02:16.820)
Yes, yes.
Lex Fridman (1:02:17.660)
So he, you know, so he,
Dmitry Korkin (1:02:21.940)
back then he approached. 60s.
Lex Fridman (1:02:25.180)
Yes, believe that, it's amazing.
Dmitry Korkin (1:02:28.940)
I mean, it still blows my mind, you know, that it's,
Lex Fridman (1:02:32.220)
that's, and this was essentially the origin
Dmitry Korkin (1:02:37.420)
of the modern bioinformatics, cheminformatics,
Lex Fridman (1:02:41.100)
you know, back in 60s.
Lex Fridman (1:02:42.780)
So that's, you know, every time you deal with projects
Lex Fridman (1:02:48.540)
like this, with the, you know, research like this,
Dmitry Korkin (1:02:51.340)
you just, you know, so the power of the, you know,
Lex Fridman (1:02:56.340)
intelligence of this people is just, you know, overwhelming.
Lex Fridman (1:03:01.740)
Do you think about expert systems, is there,
Lex Fridman (1:03:05.420)
and why they kind of didn't become successful,
Dmitry Korkin (1:03:10.380)
especially in the space of bioinformatics,
Lex Fridman (1:03:12.500)
where it does seem like there is a lot of expertise
Dmitry Korkin (1:03:15.380)
in humans, and, you know, it's possible to see
Lex Fridman (1:03:20.060)
that a system like this could be made very useful.
Dmitry Korkin (1:03:23.580)
Right.
Lex Fridman (1:03:24.420)
And be built up.
Lex Fridman (1:03:25.260)
So it's actually, it's a great question,
Lex Fridman (1:03:26.900)
and this is something, so, you know, so, you know,
Dmitry Korkin (1:03:30.460)
at my university, I teach artificial intelligence,
Lex Fridman (1:03:33.900)
and, you know, we start, my first two lectures
Dmitry Korkin (1:03:37.940)
are on the history of AI.
Lex Fridman (1:03:40.140)
And there we, you know, we try to, you know,
Dmitry Korkin (1:03:45.300)
go through the main stages of AI.
Lex Fridman (1:03:48.180)
And so, you know, the question of why expert systems failed
Dmitry Korkin (1:03:54.260)
or became obsolete, it's actually a very interesting one.
Lex Fridman (1:03:58.540)
And there are, you know, if you try to read the, you know,
Dmitry Korkin (1:04:01.980)
the historical perspectives,
Lex Fridman (1:04:03.340)
there are actually two lines of thoughts.
Dmitry Korkin (1:04:05.540)
One is that they were essentially
Lex Fridman (1:04:11.940)
not up to the expectations.
Lex Fridman (1:04:14.820)
And so therefore they were replaced, you know,
Lex Fridman (1:04:18.020)
by other things, right?
Dmitry Korkin (1:04:21.180)
The other one was that completely opposite one,
Lex Fridman (1:04:25.340)
that they were too good.
Lex Fridman (1:04:28.140)
And as a result, they essentially became
Lex Fridman (1:04:31.900)
sort of a household name,
Lex Fridman (1:04:33.180)
and then essentially they got transformed.
Lex Fridman (1:04:37.100)
I mean, in both cases, sort of the outcome was the same.
Lex Fridman (1:04:40.700)
They evolved into something, right?
Lex Fridman (1:04:43.740)
And that's what I, you know, if I look at this, right?
Lex Fridman (1:04:47.700)
So the modern machine learning, right?
Lex Fridman (1:04:50.180)
So.
Lex Fridman (1:04:51.020)
So there's echoes in the modern machine learning.
Lex Fridman (1:04:53.260)
I think so, I think so, because, you know,
Dmitry Korkin (1:04:55.340)
if you think about this, you know, and how we design,
Lex Fridman (1:04:59.140)
you know, the most successful algorithms,
Lex Fridman (1:05:02.500)
including AlphaFold, right?
Lex Fridman (1:05:04.140)
You built in the knowledge about the domain
Lex Fridman (1:05:08.100)
that you study, right?
Lex Fridman (1:05:09.940)
So you built in your expertise.
Lex Fridman (1:05:12.900)
So speaking of AlphaFold,
Lex Fridman (1:05:14.460)
so DeepMind's AlphaFold 2 recently was announced
Dmitry Korkin (1:05:18.260)
to have, quote unquote, solved protein folding.
Lex Fridman (1:05:21.980)
But how exciting is this to you?
Dmitry Korkin (1:05:24.220)
It seems to be one of the,
Lex Fridman (1:05:27.060)
one of the exciting things that have happened in 2020.
Dmitry Korkin (1:05:29.660)
It's an incredible accomplishment from the looks of it.
Lex Fridman (1:05:32.340)
What part of it is amazing to you?
Lex Fridman (1:05:33.860)
What part would you say is over hype
Lex Fridman (1:05:36.300)
or maybe misunderstood?
Dmitry Korkin (1:05:39.020)
It's definitely a very exciting achievement.
Lex Fridman (1:05:41.940)
To give you a little bit of perspective, right?
Lex Fridman (1:05:43.820)
So in bioinformatics, we have several competitions.
Lex Fridman (1:05:50.020)
And so the way, you know, you often hear
Lex Fridman (1:05:53.940)
how those competitions have been explained
Lex Fridman (1:05:56.220)
to sort of to non bioinformaticians is that, you know,
Dmitry Korkin (1:05:59.820)
they call it bioinformatics Olympic games.
Lex Fridman (1:06:01.900)
And there are several disciplines, right?
Lex Fridman (1:06:03.620)
So the historically one of the first one
Lex Fridman (1:06:07.020)
was the discipline in predicting the protein structure,
Dmitry Korkin (1:06:10.300)
predicting the 3D coordinates of the protein.
Lex Fridman (1:06:12.580)
But there are some others.
Lex Fridman (1:06:13.580)
So the predicting protein functions,
Lex Fridman (1:06:16.740)
predicting effects of mutations on protein functions,
Dmitry Korkin (1:06:21.460)
then predicting protein, protein interactions.
Lex Fridman (1:06:24.900)
So the original one was CASP
Dmitry Korkin (1:06:28.100)
or a critical assessment of a protein structure.
Lex Fridman (1:06:32.020)
And the, you know, typically what happens
Dmitry Korkin (1:06:40.020)
during this competitions is, you know, scientists,
Lex Fridman (1:06:43.980)
experimental scientists solve the structures,
Lex Fridman (1:06:48.380)
but don't put them into the protein data bank,
Lex Fridman (1:06:51.700)
which is the centralized database
Dmitry Korkin (1:06:54.700)
that contains all the 3D coordinates.
Lex Fridman (1:06:57.260)
Instead, they hold it and release protein sequences.
Lex Fridman (1:07:02.340)
And now the challenge of the community
Lex Fridman (1:07:05.420)
is to predict the 3D structures of this proteins
Lex Fridman (1:07:10.180)
and then use the experimental results structures
Lex Fridman (1:07:12.940)
to assess which one is the closest one, right?
Lex Fridman (1:07:16.620)
And this competition, by the way,
Lex Fridman (1:07:17.740)
just a bunch of different tangents.
Lex Fridman (1:07:19.540)
And maybe you can also say, what is protein folding?
Lex Fridman (1:07:22.860)
Then this competition, CASP competition
Dmitry Korkin (1:07:25.020)
has become the gold standard.
Lex Fridman (1:07:27.420)
And that's what was used to say
Dmitry Korkin (1:07:29.500)
that protein folding was solved.
Lex Fridman (1:07:32.420)
So just to add a little, just a bunch.
Lex Fridman (1:07:35.300)
So if you could, whenever you say stuff,
Lex Fridman (1:07:37.700)
maybe throw in some of the basics
Dmitry Korkin (1:07:39.380)
for the folks that might be outside of the field.
Lex Fridman (1:07:41.580)
Anyway, sorry.
Dmitry Korkin (1:07:42.740)
So, yeah, so, you know, so the reason it's, you know,
Lex Fridman (1:07:45.900)
it's relevant to our understanding of protein folding
Dmitry Korkin (1:07:50.260)
is because, you know, we've yet to learn
Lex Fridman (1:07:54.180)
how the folding mechanistically works, right?
Lex Fridman (1:07:58.140)
So there are different hypothesis,
Lex Fridman (1:08:00.740)
what happens to this fold?
Dmitry Korkin (1:08:02.780)
For example, there is a hypothesis that the folding happens
Lex Fridman (1:08:07.620)
by, you know, also in the modular fashion, right?
Lex Fridman (1:08:12.660)
So that, you know, we have protein domains
Lex Fridman (1:08:16.220)
that get folded independently
Dmitry Korkin (1:08:17.940)
because their structure is stable.
Lex Fridman (1:08:19.700)
And then the whole protein structure gets formed.
Dmitry Korkin (1:08:23.380)
But, you know, within those domains,
Lex Fridman (1:08:25.380)
we also have a so called secondary structure,
Dmitry Korkin (1:08:27.460)
the small alpha helices, beta schists.
Lex Fridman (1:08:29.820)
So these are, you know, elements that are structurally stable.
Lex Fridman (1:08:34.340)
And so, and the question is, you know,
Lex Fridman (1:08:37.820)
when do they get formed?
Dmitry Korkin (1:08:40.340)
Because some of the secondary structure elements,
Lex Fridman (1:08:42.580)
you have to have, you know, a fragment in the beginning
Lex Fridman (1:08:46.500)
and say the fragment in the middle, right?
Lex Fridman (1:08:49.420)
So you cannot potentially start having the full fold
Lex Fridman (1:08:54.780)
from the get go, right?
Lex Fridman (1:08:57.100)
So it's still, you know, it's still a big enigma,
Lex Fridman (1:09:00.340)
what happens.
Lex Fridman (1:09:01.420)
We know that it's an extremely efficient
Lex Fridman (1:09:04.260)
and stable process, right?
Lex Fridman (1:09:05.660)
So there's this long sequence
Lex Fridman (1:09:07.660)
and the fold happens really quickly.
Lex Fridman (1:09:09.500)
Exactly.
Lex Fridman (1:09:10.340)
So that's really weird, right?
Lex Fridman (1:09:11.180)
And it happens like the same way almost every time.
Dmitry Korkin (1:09:15.340)
Exactly, exactly.
Lex Fridman (1:09:16.380)
That's really weird.
Dmitry Korkin (1:09:17.860)
That's freaking weird.
Lex Fridman (1:09:19.060)
It's, yeah, that's why it's such an amazing thing.
Lex Fridman (1:09:22.900)
But most importantly, right?
Lex Fridman (1:09:24.300)
So it's, you know, so when you see the, you know,
Lex Fridman (1:09:27.460)
the translation process, right?
Lex Fridman (1:09:29.260)
So when you don't have the whole protein translated,
Dmitry Korkin (1:09:36.100)
right, it's still being translated,
Lex Fridman (1:09:37.860)
you know, getting out from the ribosome,
Dmitry Korkin (1:09:41.180)
you already see some structural, you know, fragmentation.
Lex Fridman (1:09:45.780)
So folding starts happening
Lex Fridman (1:09:49.300)
before the whole protein gets produced, right?
Lex Fridman (1:09:52.780)
And so this is obviously, you know,
Dmitry Korkin (1:09:55.060)
one of the biggest questions in, you know,
Lex Fridman (1:09:59.220)
in modern molecular biologists.
Dmitry Korkin (1:10:00.980)
Not like maybe what happens,
Lex Fridman (1:10:04.180)
like that's not as bigger than the question of folding.
Dmitry Korkin (1:10:07.860)
That's the question of like,
Lex Fridman (1:10:09.540)
something like deeper fundamental idea of folding.
Dmitry Korkin (1:10:12.460)
Yes. Behind folding.
Lex Fridman (1:10:13.380)
Exactly, exactly.
Dmitry Korkin (1:10:14.620)
So, you know, so obviously if we are able to predict
Lex Fridman (1:10:21.340)
the end product of protein folding,
Dmitry Korkin (1:10:24.060)
we are one step closer to understanding
Lex Fridman (1:10:27.660)
sort of the mechanisms of the protein folding.
Dmitry Korkin (1:10:30.220)
Because we can then potentially look and start probing
Lex Fridman (1:10:34.700)
what are the critical parts of this process
Lex Fridman (1:10:38.260)
and what are not so critical parts of this process.
Lex Fridman (1:10:41.260)
So we can start decomposing this, you know,
Lex Fridman (1:10:44.420)
so in a way this protein structure prediction algorithm
Lex Fridman (1:10:50.100)
can be used as a tool, right?
Lex Fridman (1:10:53.700)
So you change the, you know, you modify the protein,
Lex Fridman (1:10:59.220)
you get back to this tool, it predicts,
Dmitry Korkin (1:11:02.380)
okay, it's completely unstable.
Lex Fridman (1:11:04.940)
Yeah, which aspects of the input
Lex Fridman (1:11:07.820)
will have a big impact on the output?
Lex Fridman (1:11:09.860)
Exactly, exactly.
Lex Fridman (1:11:11.140)
So what happens is, you know,
Lex Fridman (1:11:13.340)
we typically have some sort of incremental advancement,
Dmitry Korkin (1:11:18.700)
you know, each stage of this CASP competition,
Lex Fridman (1:11:22.580)
you have groups with incremental advancement
Dmitry Korkin (1:11:25.340)
and, you know, historically the top performing groups
Lex Fridman (1:11:29.860)
were, you know, they were not using machine learning.
Dmitry Korkin (1:11:34.420)
They were using a very advanced biophysics
Lex Fridman (1:11:37.700)
combined with bioinformatics,
Dmitry Korkin (1:11:39.620)
combined with, you know, the data mining
Lex Fridman (1:11:43.220)
and that was, you know, that would enable them
Dmitry Korkin (1:11:47.380)
to obtain protein structures of those proteins
Lex Fridman (1:11:52.660)
that don't have any structurally solved relatives
Dmitry Korkin (1:11:57.540)
because, you know, if we have another protein,
Lex Fridman (1:12:01.860)
say the same protein, but coming from a different species,
Dmitry Korkin (1:12:07.500)
we could potentially derive some ideas
Lex Fridman (1:12:10.460)
and that's so called homology or comparative modeling,
Dmitry Korkin (1:12:13.220)
where we'll derive some ideas
Lex Fridman (1:12:15.300)
from the previously known structures
Lex Fridman (1:12:17.540)
and that would help us tremendously
Lex Fridman (1:12:19.860)
in, you know, in reconstructing the 3D structure overall.
Lex Fridman (1:12:25.380)
But what happens when we don't have these relatives?
Lex Fridman (1:12:27.900)
This is when it becomes really, really hard, right?
Lex Fridman (1:12:31.220)
So that's so called de novo, you know,
Lex Fridman (1:12:35.260)
de novo protein structure prediction.
Lex Fridman (1:12:37.500)
And in this case, those methods were traditionally very good.
Lex Fridman (1:12:43.060)
But what happened in the last year,
Dmitry Korkin (1:12:46.300)
the original alpha fold came into
Lex Fridman (1:12:50.640)
and all of a sudden it's much better than everyone else.
Dmitry Korkin (1:12:56.420)
This is 2018.
Lex Fridman (1:12:57.900)
Yeah.
Dmitry Korkin (1:12:58.740)
Oh, and the competition is only every two years, I think.
Lex Fridman (1:13:02.140)
And then, so, you know, it was sort of kind of over shockwave
Dmitry Korkin (1:13:08.060)
to the bioinformatics community that, you know,
Lex Fridman (1:13:10.740)
we have like a state of the art machine learning system
Dmitry Korkin (1:13:15.440)
that does, you know, structure prediction.
Lex Fridman (1:13:18.460)
And essentially what it does, you know,
Lex Fridman (1:13:20.780)
so if you look at this, it actually predicts the context.
Lex Fridman (1:13:26.120)
So, you know, so the process of reconstructing
Dmitry Korkin (1:13:29.460)
the 3D structure starts by predicting the context
Lex Fridman (1:13:34.700)
between the different parts of the protein.
Lex Fridman (1:13:38.860)
And the context essentially is the parts of the proteins
Lex Fridman (1:13:40.980)
that are in a close proximity to each other.
Dmitry Korkin (1:13:43.240)
Right, so actually the machine learning part
Lex Fridman (1:13:45.820)
seems to be estimating, you can correct me if I'm wrong here,
Lex Fridman (1:13:51.080)
but it seems to be estimating the distance matrix,
Lex Fridman (1:13:53.180)
which is like the distance between the different parts.
Dmitry Korkin (1:13:55.900)
Yeah, so we call the contact map.
Lex Fridman (1:13:58.080)
Contact map.
Lex Fridman (1:13:58.920)
So once you have the contact map,
Lex Fridman (1:14:00.580)
the reconstruction is becoming more straightforward, right?
Lex Fridman (1:14:04.860)
But so the contact map is the key.
Lex Fridman (1:14:06.780)
And so, you know, so that what happened.
Lex Fridman (1:14:11.260)
And now we started seeing in this current stage, right?
Lex Fridman (1:14:15.980)
Well, in the most recent one,
Dmitry Korkin (1:14:18.500)
we started seeing the emergence of these ideas
Lex Fridman (1:14:22.020)
in other people works, right?
Lex Fridman (1:14:25.080)
But yet here's, you know, AlphaFold2
Lex Fridman (1:14:29.500)
that again outperforms everyone else.
Lex Fridman (1:14:33.380)
And also by introducing yet another wave
Lex Fridman (1:14:35.780)
of the machine learning ideas.
Dmitry Korkin (1:14:38.620)
Yeah, there don't seem to be also an incorporation.
Lex Fridman (1:14:41.260)
First of all, the paper is not out yet,
Lex Fridman (1:14:43.040)
but there's a bunch of ideas already out.
Lex Fridman (1:14:44.860)
There does seem to be an incorporation of this other thing.
Dmitry Korkin (1:14:48.100)
I don't know if it's something that you could speak to,
Lex Fridman (1:14:50.160)
which is like the incorporation of like other structures,
Dmitry Korkin (1:14:58.200)
like evolutionary similar structures
Lex Fridman (1:15:01.720)
that are used to kind of give you hints.
Dmitry Korkin (1:15:03.820)
Yes, so evolutionary similarity is something
Lex Fridman (1:15:08.360)
that we can detect at different levels, right?
Lex Fridman (1:15:10.740)
So we know, for example,
Lex Fridman (1:15:12.860)
that the structure of proteins
Dmitry Korkin (1:15:17.140)
is more conserved than the sequence.
Lex Fridman (1:15:20.520)
The sequence could be very different,
Lex Fridman (1:15:22.340)
but the structural shape is actually still very conserved.
Lex Fridman (1:15:26.300)
So that's sort of the intrinsic property that, you know,
Dmitry Korkin (1:15:28.880)
in a way related to protein folds,
Lex Fridman (1:15:31.140)
you know, to the evolution of the, you know,
Dmitry Korkin (1:15:34.140)
of the proteins and protein domains, et cetera.
Lex Fridman (1:15:37.820)
But we know that, I mean, there've been multiple studies.
Dmitry Korkin (1:15:41.060)
And, you know, ideally, if you have structures,
Lex Fridman (1:15:45.340)
you know, you should use that information.
Dmitry Korkin (1:15:48.580)
However, sometimes we don't have this information.
Lex Fridman (1:15:51.220)
Instead, we have a bunch of sequences.
Lex Fridman (1:15:53.220)
Sequences, we have a lot, right?
Lex Fridman (1:15:54.860)
So we have, you know, hundreds, thousands
Lex Fridman (1:16:00.340)
of, you know, different organisms sequenced, right?
Lex Fridman (1:16:04.300)
And by taking the same protein,
Lex Fridman (1:16:07.840)
but in different organisms and aligning it,
Lex Fridman (1:16:11.620)
so making it, you know, making the corresponding positions
Dmitry Korkin (1:16:15.980)
aligned, we can actually say a lot
Lex Fridman (1:16:20.500)
about sort of what is conserved in this protein
Lex Fridman (1:16:24.220)
and therefore, you know, structurally more stable,
Lex Fridman (1:16:26.920)
what is diverse in this protein.
Lex Fridman (1:16:28.920)
So on top of that, we could provide sort of the information
Lex Fridman (1:16:32.380)
about the sort of the secondary structure
Dmitry Korkin (1:16:35.100)
of this protein, et cetera, et cetera.
Lex Fridman (1:16:36.420)
So this information is extremely useful
Lex Fridman (1:16:39.940)
and it's already there.
Lex Fridman (1:16:41.300)
So while it's tempting to, you know,
Dmitry Korkin (1:16:44.140)
to do a complete ab initio,
Lex Fridman (1:16:46.060)
so you just have a protein sequence and nothing else,
Dmitry Korkin (1:16:49.540)
the reality is such that we are overwhelmed with this data.
Lex Fridman (1:16:54.220)
So why not use it?
Lex Fridman (1:16:56.500)
And so, yeah, so I'm looking forward
Lex Fridman (1:16:59.220)
to reading this paper.
Dmitry Korkin (1:17:01.500)
It does seem to, like they've,
Lex Fridman (1:17:03.420)
in the previous version of Alpha Fold,
Dmitry Korkin (1:17:05.100)
they didn't, for this evolutionary similarity thing,
Lex Fridman (1:17:09.780)
they didn't use machine learning for that.
Dmitry Korkin (1:17:12.960)
Or rather, they used it as like the input
Lex Fridman (1:17:15.600)
to the entirety of the neural net,
Dmitry Korkin (1:17:17.880)
like the features derived from the similarity.
Lex Fridman (1:17:22.020)
It seems like there's some kind of quote, unquote,
Dmitry Korkin (1:17:24.660)
iterative thing where it seems to be part of the learning
Lex Fridman (1:17:30.500)
process is the incorporation of this evolutionary similarity.
Lex Fridman (1:17:34.260)
Yeah, I don't think there is a bioarchive paper, right?
Lex Fridman (1:17:36.940)
There's nothing.
Dmitry Korkin (1:17:37.780)
No, there's nothing.
Lex Fridman (1:17:38.620)
There's a blog post that's written
Dmitry Korkin (1:17:40.680)
by a marketing team, essentially,
Lex Fridman (1:17:42.600)
which, you know, it has some scientific similarity,
Dmitry Korkin (1:17:48.420)
probably, to the actual methodology used,
Lex Fridman (1:17:51.780)
but it could be, it's like interpreting scripture.
Dmitry Korkin (1:17:55.220)
It could be just poetic interpretations of the actual work
Lex Fridman (1:17:59.660)
as opposed to direct connection to the work.
Lex Fridman (1:18:01.900)
So now, speaking about protein folding, right?
Lex Fridman (1:18:04.260)
So, you know, in order to answer the question
Lex Fridman (1:18:06.820)
whether or not we have solved this, right?
Lex Fridman (1:18:09.460)
So we need to go back to the beginning of our conversation
Dmitry Korkin (1:18:13.580)
with the realization that an average protein
Lex Fridman (1:18:16.100)
is that typically what the CASP has been focusing on
Dmitry Korkin (1:18:22.180)
is this competition has been focusing
Lex Fridman (1:18:25.820)
on the single, maybe two domain proteins
Dmitry Korkin (1:18:29.220)
that are still very compact.
Lex Fridman (1:18:31.060)
And even those ones are extremely challenging to solve.
Lex Fridman (1:18:35.400)
But now we talk about, you know,
Lex Fridman (1:18:37.660)
an average protein that has two, three protein domains.
Dmitry Korkin (1:18:42.420)
If you look at the proteins that are in charge
Lex Fridman (1:18:46.920)
of the, you know, of the process with the neural system,
Dmitry Korkin (1:18:51.120)
right, perhaps one of the most recently evolved
Lex Fridman (1:18:58.500)
sort of systems in an organism, right?
Dmitry Korkin (1:19:03.500)
All of them, well, the majority of them
Lex Fridman (1:19:06.360)
are highly multi domain proteins.
Lex Fridman (1:19:09.000)
So they are, you know, some of them have five, six, seven,
Lex Fridman (1:19:13.520)
you know, and more domains, right?
Dmitry Korkin (1:19:16.840)
And, you know, we are very far away
Lex Fridman (1:19:20.000)
from understanding how these proteins are folded.
Lex Fridman (1:19:22.400)
So the complexity of the protein matters here.
Lex Fridman (1:19:24.440)
The complexity of the protein modules
Dmitry Korkin (1:19:27.920)
or the protein domains.
Lex Fridman (1:19:30.220)
So you're saying solved, so the definition
Dmitry Korkin (1:19:35.220)
of solved here is particularly the CASP competition
Lex Fridman (1:19:38.620)
achieving human level, not human level,
Dmitry Korkin (1:19:41.760)
achieving experimental level performance
Lex Fridman (1:19:45.620)
on these particular sets of proteins
Dmitry Korkin (1:19:48.520)
that have been used in these competitions.
Lex Fridman (1:19:50.300)
Well, I mean, you know, I do think that, you know,
Dmitry Korkin (1:19:54.740)
especially with regards to the alpha fold,
Lex Fridman (1:19:57.500)
you know, it is able to, you know, to solve,
Dmitry Korkin (1:20:03.020)
you know, at the near experimental level,
Lex Fridman (1:20:08.980)
pre big majority of the more compact proteins
Dmitry Korkin (1:20:15.000)
like, or protein domains.
Lex Fridman (1:20:16.360)
Because again, in order to understand
Lex Fridman (1:20:18.740)
how the overall protein, you know,
Lex Fridman (1:20:22.800)
multi domain protein fold, we do need to understand
Dmitry Korkin (1:20:26.220)
the structure of its individual domains.
Lex Fridman (1:20:28.760)
I mean, unlike if you look at alpha zero
Dmitry Korkin (1:20:31.140)
or like even mu zero, if you look at that work,
Lex Fridman (1:20:36.500)
you know, it's nice reinforcement learning
Dmitry Korkin (1:20:39.540)
self playing mechanisms are nice
Lex Fridman (1:20:41.100)
cause it's all in simulation.
Lex Fridman (1:20:42.380)
So you can learn from just huge amounts.
Lex Fridman (1:20:45.920)
Like you don't need data.
Dmitry Korkin (1:20:47.340)
It was like the problem with proteins,
Lex Fridman (1:20:49.740)
like the size, I forget how many 3D structures
Dmitry Korkin (1:20:54.540)
have been mapped, but the training data is very small.
Lex Fridman (1:20:56.980)
No matter what, it's like millions,
Dmitry Korkin (1:20:59.060)
maybe a one or two million or something like that,
Lex Fridman (1:21:01.400)
but it's some very small number,
Lex Fridman (1:21:02.940)
but like, it doesn't seem like that's scalable.
Lex Fridman (1:21:06.820)
There has to be, I don't know,
Dmitry Korkin (1:21:09.380)
it feels like you want to somehow 10 X the data
Lex Fridman (1:21:13.100)
or a hundred X the data somehow.
Dmitry Korkin (1:21:15.700)
Yes, but we also can take advantage of homology models,
Lex Fridman (1:21:20.700)
right, so the models that are of very good quality
Dmitry Korkin (1:21:26.740)
because they are essentially obtained
Lex Fridman (1:21:30.660)
based on the evolutionary information, right?
Lex Fridman (1:21:33.720)
So you can, there is a potential to enhance this information
Lex Fridman (1:21:38.540)
and, you know, use it again to empower the training set.
Lex Fridman (1:21:43.540)
And it's, I think, I am actually very optimistic.
Lex Fridman (1:21:49.780)
I think it's been one of this sort of, you know,
Dmitry Korkin (1:21:58.100)
churning events where you have a system that is,
Lex Fridman (1:22:05.220)
you know, a machine learning system
Dmitry Korkin (1:22:07.300)
that is truly better than the machine learning system.
Lex Fridman (1:22:12.300)
Better than the sort of the more conventional
Dmitry Korkin (1:22:15.740)
biophysics based methods.
Lex Fridman (1:22:17.940)
That's a huge leap.
Dmitry Korkin (1:22:19.380)
This is one of those fun questions,
Lex Fridman (1:22:21.280)
but where would you put it in the ranking
Dmitry Korkin (1:22:26.720)
of the greatest breakthroughs
Lex Fridman (1:22:28.540)
in artificial intelligence history?
Lex Fridman (1:22:31.740)
So like, okay, so let's see who's in the running.
Lex Fridman (1:22:34.940)
Maybe you can correct me.
Lex Fridman (1:22:35.860)
So you got like AlphaZero and AlphaGo
Lex Fridman (1:22:39.900)
beating the world champion at the game of Go.
Dmitry Korkin (1:22:44.500)
Thought to be impossible like 20 years ago.
Lex Fridman (1:22:48.220)
Or at least the AI community was highly skeptical.
Dmitry Korkin (1:22:51.340)
Then you got like also Deep Blue original Kasparov.
Lex Fridman (1:22:55.060)
You have deep learning itself,
Dmitry Korkin (1:22:56.260)
like the maybe, what would you say,
Lex Fridman (1:22:58.280)
the AlexNet, ImageNet moment.
Lex Fridman (1:23:00.940)
So the first neural network
Lex Fridman (1:23:02.780)
achieving human level performance.
Dmitry Korkin (1:23:04.780)
Super, that's not true.
Lex Fridman (1:23:07.860)
Achieving like a big leap in performance
Dmitry Korkin (1:23:10.980)
on the computer vision problem.
Lex Fridman (1:23:14.420)
There is OpenAI, the whole like GPT3,
Dmitry Korkin (1:23:18.980)
that whole space of transformers and language models
Lex Fridman (1:23:23.020)
just achieving this incredible performance
Dmitry Korkin (1:23:27.120)
of application of neural networks to language models.
Lex Fridman (1:23:31.780)
Boston Dynamics, pretty cool.
Dmitry Korkin (1:23:33.540)
Like robotics.
Lex Fridman (1:23:35.220)
People are like, there's no AI.
Dmitry Korkin (1:23:38.200)
No, no, there's no machine learning currently.
Lex Fridman (1:23:41.520)
But AI is much bigger than machine learning.
Lex Fridman (1:23:44.520)
So that just the engineering aspect,
Lex Fridman (1:23:48.860)
I would say it's one of the greatest accomplishments
Dmitry Korkin (1:23:50.780)
in engineering side.
Lex Fridman (1:23:52.860)
Engineering meaning like mechanical engineering
Dmitry Korkin (1:23:56.140)
of robotics ever.
Lex Fridman (1:23:57.980)
Then of course, autonomous vehicles.
Dmitry Korkin (1:23:59.500)
You can argue for Waymo,
Lex Fridman (1:24:01.300)
which is like the Google self driving car.
Dmitry Korkin (1:24:03.580)
Or you can argue for Tesla,
Lex Fridman (1:24:05.460)
which is like actually being used
Dmitry Korkin (1:24:07.860)
by hundreds of thousands of people on the road today,
Lex Fridman (1:24:10.740)
machine learning system.
Lex Fridman (1:24:13.700)
And I don't know if you can, what else is there?
Lex Fridman (1:24:17.560)
But I think that's it.
Lex Fridman (1:24:18.700)
And then AlphaFold, many people are saying
Lex Fridman (1:24:20.900)
is up there, potentially number one.
Lex Fridman (1:24:23.300)
Would you put them at number one?
Lex Fridman (1:24:24.820)
Well, in terms of the impact on the science
Lex Fridman (1:24:29.820)
and on the society beyond, it's definitely,
Lex Fridman (1:24:34.060)
to me would be one of the...
Lex Fridman (1:24:37.460)
Top three?
Lex Fridman (1:24:39.060)
What you want?
Dmitry Korkin (1:24:39.900)
Maybe, I mean, I'm probably not the best person
Lex Fridman (1:24:43.020)
to answer that.
Lex Fridman (1:24:45.460)
But I do have, I remember my,
Lex Fridman (1:24:51.540)
back in, I think 1997, when Deep Blue,
Dmitry Korkin (1:24:56.380)
that Kasparov, it was, I mean, it was a shock.
Lex Fridman (1:25:01.860)
I mean, it was, and I think for the,
Dmitry Korkin (1:25:04.180)
for the pre substantial part of the world,
Lex Fridman (1:25:14.220)
that especially people who have some experience with chess,
Lex Fridman (1:25:21.740)
and realizing how incredibly human this game,
Lex Fridman (1:25:25.660)
how much of a brain power you need
Dmitry Korkin (1:25:30.220)
to reach those levels of grandmasters, right, level.
Lex Fridman (1:25:36.020)
And it's probably one of the first time,
Lex Fridman (1:25:37.920)
and how good Kasparov was.
Lex Fridman (1:25:39.780)
And again, yeah, so Kasparov's arguably
Lex Fridman (1:25:42.300)
one of the best ever, right?
Lex Fridman (1:25:45.580)
And you get a machine that beats him.
Dmitry Korkin (1:25:47.860)
All right, so it's...
Lex Fridman (1:25:48.820)
First time a machine probably beat a human
Dmitry Korkin (1:25:50.740)
at that scale of a thing, of anything.
Lex Fridman (1:25:53.720)
Yes, yes.
Lex Fridman (1:25:54.740)
So that was, to me, that was like, you know,
Lex Fridman (1:25:57.220)
one of the groundbreaking events in the history of AI.
Dmitry Korkin (1:26:00.620)
Yeah, that's probably number one.
Lex Fridman (1:26:02.340)
Probably, like we don't, it's hard to remember.
Dmitry Korkin (1:26:05.460)
It's like Muhammad Ali versus, I don't know,
Lex Fridman (1:26:08.100)
any of the Mike Tyson, something like that.
Dmitry Korkin (1:26:09.900)
It's like, nah, you gotta put Muhammad Ali at number one.
Lex Fridman (1:26:13.660)
Same with Deep Blue,
Dmitry Korkin (1:26:15.300)
even though it's not machine learning based.
Lex Fridman (1:26:19.340)
Still, it uses advanced search,
Lex Fridman (1:26:21.540)
and search is the integral part of AI, right?
Lex Fridman (1:26:24.420)
It's not, people don't think of it that way at this moment.
Dmitry Korkin (1:26:27.660)
In vogue currently, search is not seen
Lex Fridman (1:26:30.900)
as a fundamental aspect of intelligence,
Lex Fridman (1:26:34.220)
but it very well, I mean, it very likely is.
Lex Fridman (1:26:37.700)
In fact, I mean, that's what neural networks are,
Dmitry Korkin (1:26:39.660)
is they're just performing search
Lex Fridman (1:26:41.260)
on the space of parameters, and it's all search.
Dmitry Korkin (1:26:45.540)
All of intelligence is some form of search,
Lex Fridman (1:26:47.740)
and you just have to become cleverer and clever
Dmitry Korkin (1:26:49.660)
at that search problem.
Lex Fridman (1:26:50.900)
And I also have another one that you didn't mention
Dmitry Korkin (1:26:53.980)
that's one of my favorite ones is,
Lex Fridman (1:26:58.260)
so you've probably heard of this,
Dmitry Korkin (1:26:59.860)
it's, I think it's called Deep Rembrandt.
Lex Fridman (1:27:03.420)
It's the project where they trained,
Dmitry Korkin (1:27:06.820)
I think there was a collaboration
Lex Fridman (1:27:08.220)
between the sort of the experts
Dmitry Korkin (1:27:11.580)
in Rembrandt painting in Netherlands,
Lex Fridman (1:27:15.500)
and a group, an artificial intelligence group,
Dmitry Korkin (1:27:18.300)
where they train an algorithm
Lex Fridman (1:27:20.220)
to replicate the style of the Rembrandt,
Lex Fridman (1:27:22.980)
and they actually printed a portrait
Lex Fridman (1:27:26.980)
that never existed before in the style of Rembrandt.
Dmitry Korkin (1:27:32.620)
I think they printed it on a sort of,
Lex Fridman (1:27:36.740)
on the canvas that, you know,
Dmitry Korkin (1:27:38.500)
using pretty much same types of paints and stuff.
Lex Fridman (1:27:42.580)
To me, it was mind blowing.
Dmitry Korkin (1:27:44.060)
Yeah, and the space of art, that's interesting.
Lex Fridman (1:27:46.900)
There hasn't been, maybe that's it,
Lex Fridman (1:27:50.100)
but I think there hasn't been an image in that moment yet
Lex Fridman (1:27:54.580)
in the space of art.
Dmitry Korkin (1:27:56.780)
You haven't been able to achieve
Lex Fridman (1:27:58.620)
superhuman level performance in the space of art,
Dmitry Korkin (1:28:01.420)
even though there's this big famous thing
Lex Fridman (1:28:04.660)
where a piece of art was purchased,
Dmitry Korkin (1:28:07.660)
I guess for a lot of money.
Lex Fridman (1:28:08.700)
Yes.
Dmitry Korkin (1:28:09.540)
Yeah, but it's still, you know,
Lex Fridman (1:28:11.660)
people are like in the space of music at least,
Dmitry Korkin (1:28:15.620)
that's, you know, it's clear that human created pieces
Lex Fridman (1:28:19.740)
are much more popular.
Lex Fridman (1:28:21.700)
So there hasn't been a moment where it's like,
Lex Fridman (1:28:24.420)
oh, this is, we're now,
Dmitry Korkin (1:28:26.700)
I would say in the space of music,
Lex Fridman (1:28:28.780)
what makes a lot of money,
Dmitry Korkin (1:28:30.140)
we're talking about serious money,
Lex Fridman (1:28:32.100)
it's music and movies, or like shows and so on,
Lex Fridman (1:28:35.300)
and entertainment.
Lex Fridman (1:28:36.640)
There hasn't been a moment where AI created,
Dmitry Korkin (1:28:41.280)
AI was able to create a piece of music
Lex Fridman (1:28:44.460)
or a piece of cinema, like Netflix show,
Dmitry Korkin (1:28:49.820)
that is, you know, that's sufficiently popular
Lex Fridman (1:28:53.540)
to make a ton of money.
Dmitry Korkin (1:28:55.260)
Yeah.
Lex Fridman (1:28:56.100)
And that moment would be very, very powerful,
Dmitry Korkin (1:28:58.940)
because that's like, that's an AI system
Lex Fridman (1:29:01.560)
being used to make a lot of money.
Lex Fridman (1:29:03.060)
And like direct, of course, AI tools,
Lex Fridman (1:29:05.480)
like even Premiere, audio editing,
Dmitry Korkin (1:29:07.140)
all the editing, everything I do,
Lex Fridman (1:29:08.780)
to edit this podcast, there's a lot of AI involved.
Dmitry Korkin (1:29:11.660)
Actually, this is a program,
Lex Fridman (1:29:13.260)
I wanna talk to those folks, just cause I wanna nerd out,
Dmitry Korkin (1:29:15.540)
it's called iZotope, I don't know if you're familiar with it.
Lex Fridman (1:29:18.060)
They have a bunch of tools of audio processing,
Lex Fridman (1:29:20.140)
and they have, I think they're Boston based,
Lex Fridman (1:29:23.080)
just, it's so exciting to me to use it,
Dmitry Korkin (1:29:26.380)
like on the audio here,
Lex Fridman (1:29:28.200)
cause it's all machine learning.
Dmitry Korkin (1:29:30.380)
It's not, cause most audio production stuff
Lex Fridman (1:29:35.780)
is like any kind of processing you do,
Dmitry Korkin (1:29:37.540)
it's very basic signal processing,
Lex Fridman (1:29:39.500)
and you're tuning knobs and so on.
Dmitry Korkin (1:29:41.980)
They have all of that, of course,
Lex Fridman (1:29:43.580)
but they also have all of this machine learning stuff,
Dmitry Korkin (1:29:46.020)
like where you actually give it training data,
Lex Fridman (1:29:48.520)
you select parts of the audio you train on,
Dmitry Korkin (1:29:51.740)
you train on it, and it figures stuff out.
Lex Fridman (1:29:56.380)
It's great, it's able to detect,
Dmitry Korkin (1:29:59.020)
like the ability of it to be able
Lex Fridman (1:30:01.380)
to separate voice and music, for example,
Dmitry Korkin (1:30:04.820)
or voice and anything, is incredible.
Lex Fridman (1:30:07.260)
Like it just, it's clearly exceptionally good
Dmitry Korkin (1:30:11.140)
at applying these different neural networks models
Lex Fridman (1:30:14.940)
to just separate the different kinds
Dmitry Korkin (1:30:17.740)
of signals from the audio.
Lex Fridman (1:30:19.180)
That, okay, so that's really exciting.
Dmitry Korkin (1:30:22.260)
Photoshop, Adobe people also use it,
Lex Fridman (1:30:24.580)
but to generate a piece of music
Dmitry Korkin (1:30:28.260)
that will sell millions, a piece of art, yeah.
Lex Fridman (1:30:31.980)
No, I agree, and you know, it's,
Dmitry Korkin (1:30:34.420)
that's, you know, as I mentioned,
Lex Fridman (1:30:39.220)
I offer my AI class, and you know,
Lex Fridman (1:30:41.700)
an integral part of this is the project, right?
Lex Fridman (1:30:44.660)
So it's my favorite, ultimate favorite part,
Dmitry Korkin (1:30:47.340)
because it typically, we have these project presentations
Lex Fridman (1:30:51.380)
the last two weeks of the classes,
Dmitry Korkin (1:30:53.720)
right before, you know, the Christmas break,
Lex Fridman (1:30:56.220)
and it's sort of, it adds this cool excitement,
Lex Fridman (1:31:00.300)
and every time, I mean, I'm amazed, you know,
Lex Fridman (1:31:02.660)
with some projects that people, you know, come up with.
Lex Fridman (1:31:07.660)
And so, and quite a few of them are actually, you know,
Lex Fridman (1:31:12.060)
they have some link to arts.
Dmitry Korkin (1:31:17.060)
I mean, you know, I think last year we had a group
Lex Fridman (1:31:21.260)
who designed an AI producing hokus, Japanese poems.
Dmitry Korkin (1:31:27.660)
Oh, wow.
Lex Fridman (1:31:29.380)
So, and some of them, so, you know,
Dmitry Korkin (1:31:31.820)
it got trained on the English based,
Lex Fridman (1:31:34.780)
haikus, haikus, right?
Dmitry Korkin (1:31:36.460)
So, and some of them, you know,
Lex Fridman (1:31:40.260)
they get to present, like, the top selection.
Dmitry Korkin (1:31:43.460)
They were pretty good.
Lex Fridman (1:31:44.300)
I mean, you know, I mean, of course, I'm not a specialist,
Lex Fridman (1:31:47.020)
but you read them, and you see this is real.
Lex Fridman (1:31:49.700)
It seems profound.
Dmitry Korkin (1:31:50.660)
Yes, yeah, it seems real.
Lex Fridman (1:31:52.780)
So it's kind of cool.
Dmitry Korkin (1:31:55.060)
We also had a couple of projects where people tried
Lex Fridman (1:31:57.940)
to teach AI how to play, like, rock music, classical music.
Dmitry Korkin (1:32:02.940)
I think, and popular music.
Lex Fridman (1:32:05.940)
Yeah.
Dmitry Korkin (1:32:07.820)
Interestingly enough, you know,
Lex Fridman (1:32:10.620)
classical music was among the most difficult ones.
Dmitry Korkin (1:32:14.580)
Oh, sure.
Lex Fridman (1:32:15.420)
And, you know, of course, if you, if, you know,
Dmitry Korkin (1:32:21.780)
you know, if you look at the, you know,
Lex Fridman (1:32:23.780)
the, like, grandmasters of music, like Bach, right?
Lex Fridman (1:32:28.780)
So there is a lot of, there is a lot of,
Lex Fridman (1:32:31.940)
there is a lot of almost math.
Dmitry Korkin (1:32:34.820)
Yeah, well, he's very mathematical.
Lex Fridman (1:32:36.580)
Yeah, exactly.
Lex Fridman (1:32:37.420)
So this is, I would imagine that at least some style
Lex Fridman (1:32:41.500)
of this music could be picked up,
Lex Fridman (1:32:43.820)
but then you have this completely different spectrum
Lex Fridman (1:32:46.980)
of classical composers.
Lex Fridman (1:32:49.260)
And so, you know, it's almost like, you know,
Lex Fridman (1:32:54.140)
you don't have to sort of look at the data.
Dmitry Korkin (1:32:56.780)
You just listen to it and say, nah, that's not it, not yet.
Lex Fridman (1:33:01.140)
That's not it, yeah.
Dmitry Korkin (1:33:02.380)
That's how I feel too.
Lex Fridman (1:33:03.340)
There's OpenAI has, I think, OpenMuse
Dmitry Korkin (1:33:05.820)
or something like that, the system.
Lex Fridman (1:33:07.540)
It's cool, but it's like, eh,
Dmitry Korkin (1:33:09.740)
it's not compelling for some reason.
Lex Fridman (1:33:12.060)
It could be a psychological reason too.
Dmitry Korkin (1:33:14.180)
Maybe we need to have a human being,
Lex Fridman (1:33:17.580)
a tortured soul behind the music.
Dmitry Korkin (1:33:19.660)
I don't know.
Lex Fridman (1:33:20.700)
Yeah, no, absolutely.
Dmitry Korkin (1:33:22.220)
I completely agree.
Lex Fridman (1:33:23.940)
But yeah, whether or not we'll have,
Dmitry Korkin (1:33:26.580)
one day we'll have, you know,
Lex Fridman (1:33:29.140)
a song written by an AI engine
Dmitry Korkin (1:33:33.340)
to be like in top charts, musical charts,
Lex Fridman (1:33:37.980)
I wouldn't be surprised.
Dmitry Korkin (1:33:40.540)
I wouldn't be surprised.
Lex Fridman (1:33:43.380)
I wonder if we already have one
Lex Fridman (1:33:44.700)
and it just hasn't been announced.
Lex Fridman (1:33:48.020)
We wouldn't know.
Lex Fridman (1:33:49.980)
How hard is the multi protein folding problem?
Lex Fridman (1:33:53.940)
Is that kind of something you've already mentioned
Dmitry Korkin (1:33:57.100)
which is baked into this idea of greater
Lex Fridman (1:33:59.180)
and greater complexity of proteins?
Dmitry Korkin (1:34:01.180)
Like multi domain proteins,
Lex Fridman (1:34:03.300)
is that basically become multi protein complexes?
Dmitry Korkin (1:34:08.940)
Yes, you got it right.
Lex Fridman (1:34:10.620)
So it's sort of, it has the components
Dmitry Korkin (1:34:15.980)
of both of protein folding
Lex Fridman (1:34:18.460)
and protein, protein interactions.
Dmitry Korkin (1:34:21.900)
Because in order for these domains,
Lex Fridman (1:34:24.460)
many of these proteins actually,
Dmitry Korkin (1:34:27.260)
they never form a stable structure.
Lex Fridman (1:34:31.140)
One of my favorite proteins,
Lex Fridman (1:34:33.020)
and pretty much everyone who works in the,
Lex Fridman (1:34:37.700)
I know, whom I know, who works with proteins,
Dmitry Korkin (1:34:41.740)
they always have their favorite proteins.
Lex Fridman (1:34:44.660)
Right, so one of my favorite proteins,
Dmitry Korkin (1:34:47.660)
probably my favorite protein,
Lex Fridman (1:34:49.140)
the one that I worked when I was a postdoc
Dmitry Korkin (1:34:51.420)
is so called post synaptic density 95, PSD 95 protein.
Lex Fridman (1:34:56.180)
So it's one of the key actors
Dmitry Korkin (1:35:00.500)
in the majority of neurological processes
Lex Fridman (1:35:03.780)
at the molecular level.
Lex Fridman (1:35:04.700)
So it's a, and essentially it's a key player
Lex Fridman (1:35:11.060)
in the post synaptic density.
Lex Fridman (1:35:13.460)
So this is the crucial part of this synapse
Lex Fridman (1:35:17.180)
where a lot of these chemical processes are happening.
Lex Fridman (1:35:22.420)
So it has five domains, right?
Lex Fridman (1:35:26.220)
So five protein domains.
Lex Fridman (1:35:27.460)
So pretty large proteins, I think 600 something assets.
Lex Fridman (1:35:35.700)
But the way it's organized itself, it's flexible, right?
Lex Fridman (1:35:41.260)
So it acts as a scaffold.
Lex Fridman (1:35:43.820)
So it is used to bring in other proteins.
Lex Fridman (1:35:49.260)
So they start acting in the orchestrated manner, right?
Lex Fridman (1:35:54.260)
So, and the type of the shape of this protein,
Dmitry Korkin (1:35:58.780)
it's in a way, there are some stable parts of this protein,
Lex Fridman (1:36:02.500)
but there are some flexible.
Lex Fridman (1:36:04.420)
And this flexibility is built in into the protein
Lex Fridman (1:36:08.580)
in order to become sort of this multifunctional machine.
Lex Fridman (1:36:13.100)
So do you think that kind of thing is also learnable
Lex Fridman (1:36:16.460)
through the alpha fold two kind of approach?
Dmitry Korkin (1:36:19.340)
I mean, the time will tell.
Lex Fridman (1:36:22.380)
Is it another level of complexity?
Dmitry Korkin (1:36:24.460)
Is it like how big of a jump in complexity
Lex Fridman (1:36:27.300)
is that whole thing?
Dmitry Korkin (1:36:28.140)
To me, it's yet another level of complexity
Lex Fridman (1:36:31.340)
because when we talk about protein, protein interactions,
Lex Fridman (1:36:35.140)
and there is actually a different challenge for this
Lex Fridman (1:36:38.820)
called Capri, and so this, that is focused specifically
Dmitry Korkin (1:36:43.420)
on macromolecular interactions, protein, protein, protein,
Lex Fridman (1:36:47.060)
DNA, et cetera.
Dmitry Korkin (1:36:48.540)
So, but it's, there are different mechanisms
Lex Fridman (1:36:56.020)
that govern molecular interactions
Lex Fridman (1:36:58.740)
and that need to be picked up,
Lex Fridman (1:37:00.740)
say by a machine learning algorithm.
Dmitry Korkin (1:37:03.660)
Interestingly enough, we actually,
Lex Fridman (1:37:06.500)
we participated for a few years in this competition.
Dmitry Korkin (1:37:11.740)
We typically don't participate in competitions,
Lex Fridman (1:37:14.900)
I don't know, don't have enough time,
Dmitry Korkin (1:37:19.820)
because it's very intensive, it's a very intensive process.
Lex Fridman (1:37:23.700)
But we participated back in about 10 years ago or so.
Lex Fridman (1:37:30.580)
And the way we entered this competition,
Lex Fridman (1:37:32.660)
so we design a scoring function, right?
Lex Fridman (1:37:35.420)
So the function that evaluates
Lex Fridman (1:37:37.580)
whether or not your protein, protein interaction
Lex Fridman (1:37:40.540)
is supposed to look like experimentally solved, right?
Lex Fridman (1:37:43.380)
So the scoring function is very critical part
Dmitry Korkin (1:37:45.900)
of the model prediction.
Lex Fridman (1:37:49.820)
So we designed it to be a machine learning one.
Lex Fridman (1:37:52.740)
And so it was one of the first machine learning
Lex Fridman (1:37:56.620)
based scoring function used in Capri.
Lex Fridman (1:38:00.020)
And we essentially learned what should contribute,
Lex Fridman (1:38:06.580)
what are the critical components contributing
Dmitry Korkin (1:38:08.860)
into the protein, protein interaction.
Lex Fridman (1:38:10.540)
So this could be converted into a learning problem
Lex Fridman (1:38:13.340)
and thereby it could be learned?
Lex Fridman (1:38:15.620)
I believe so, yes.
Lex Fridman (1:38:17.020)
Do you think AlphaFold2 or something similar to it
Lex Fridman (1:38:20.460)
from DeepMind or somebody else will be,
Lex Fridman (1:38:24.300)
will result in a Nobel Prize or multiple Nobel Prizes?
Lex Fridman (1:38:28.660)
So like, you know, obviously, maybe not so obviously,
Dmitry Korkin (1:38:33.300)
you can't give a Nobel Prize to a computer program.
Lex Fridman (1:38:38.660)
At least for now, give it to the designers of that program.
Lex Fridman (1:38:42.140)
But do you see one or multiple Nobel Prizes
Lex Fridman (1:38:46.060)
where AlphaFold2 is like a large percentage
Lex Fridman (1:38:51.700)
of what that prize is given for?
Lex Fridman (1:38:54.860)
Would it lead to discoveries at the level of Nobel Prizes?
Dmitry Korkin (1:39:00.540)
I mean, I think we are definitely destined
Lex Fridman (1:39:05.420)
to see the Nobel Prize becoming sort of,
Dmitry Korkin (1:39:08.740)
to be evolving with the evolution of science
Lex Fridman (1:39:12.340)
and the evolution of science as such
Lex Fridman (1:39:14.540)
that it now becomes like really multi facets, right?
Lex Fridman (1:39:17.860)
So where you don't really have like a unique discipline,
Dmitry Korkin (1:39:21.340)
you have sort of the, a lot of cross disciplinary talks
Lex Fridman (1:39:25.660)
in order to achieve sort of, you know,
Dmitry Korkin (1:39:28.740)
really big advancements, you know.
Lex Fridman (1:39:32.380)
So I think, you know, the computational methods
Dmitry Korkin (1:39:39.180)
will be acknowledged in one way or another.
Lex Fridman (1:39:42.500)
And as a matter of fact, you know,
Lex Fridman (1:39:46.860)
they were first acknowledged back in 2013, right?
Lex Fridman (1:39:50.580)
Where, you know, the first three people were, you know,
Dmitry Korkin (1:39:56.500)
awarded the Nobel Prize for study the protein folding,
Lex Fridman (1:40:00.540)
right, the principle.
Dmitry Korkin (1:40:01.500)
And, you know, I think all three of them
Lex Fridman (1:40:03.820)
are computational biophysicists, right?
Dmitry Korkin (1:40:06.940)
So, you know, that I think is unavoidable.
Lex Fridman (1:40:13.260)
You know, it will come with the time.
Dmitry Korkin (1:40:16.580)
The fact that, you know, alpha fold and, you know,
Lex Fridman (1:40:23.540)
similar approaches, because again, it's a matter of time
Dmitry Korkin (1:40:26.340)
that people will embrace this, you know, principle
Lex Fridman (1:40:31.700)
and we'll see more and more such, you know,
Dmitry Korkin (1:40:34.940)
such tools coming into play.
Lex Fridman (1:40:36.940)
But, you know, these methods will be critical
Dmitry Korkin (1:40:41.940)
in a scientific discovery, no doubts about it.
Lex Fridman (1:40:47.700)
On the engineering side, maybe a dark question,
Lex Fridman (1:40:51.500)
but do you think it's possible to use
Lex Fridman (1:40:53.860)
these machine learning methods
Lex Fridman (1:40:55.140)
to start to engineer proteins?
Lex Fridman (1:40:59.020)
And the next question is something quite a few biologists
Dmitry Korkin (1:41:04.660)
are against, some are for, for study purposes,
Lex Fridman (1:41:07.300)
is to engineer viruses.
Lex Fridman (1:41:09.620)
Do you think machine learning, like something like alpha fold
Lex Fridman (1:41:12.620)
could be used to engineer viruses?
Lex Fridman (1:41:14.780)
So to answer the first question, you know,
Lex Fridman (1:41:16.980)
it has been, you know, a part of the research
Dmitry Korkin (1:41:21.660)
in the protein science, the protein design is, you know,
Lex Fridman (1:41:25.500)
is a very prominent areas of research.
Dmitry Korkin (1:41:29.180)
Of course, you know, one of the pioneers is David Baker
Lex Fridman (1:41:32.020)
and Rosetta algorithm that, you know,
Dmitry Korkin (1:41:34.900)
essentially was doing the de novo design and was used
Lex Fridman (1:41:39.740)
to design new proteins, you know.
Lex Fridman (1:41:41.580)
And design of proteins means design of function.
Lex Fridman (1:41:44.220)
So like when you design a protein, you can control,
Dmitry Korkin (1:41:47.300)
I mean, the whole point of a protein
Lex Fridman (1:41:49.100)
with the protein structure comes a function,
Dmitry Korkin (1:41:52.180)
like it's doing something.
Lex Fridman (1:41:53.700)
Correct.
Lex Fridman (1:41:54.540)
So you can design different things.
Lex Fridman (1:41:56.060)
So you can, yeah, so you can, well,
Dmitry Korkin (1:41:58.140)
you can look at the proteins from the functional perspective.
Lex Fridman (1:42:00.700)
You can also look at the proteins
Lex Fridman (1:42:02.700)
from the structural perspective, right?
Lex Fridman (1:42:04.180)
So the structural building blocks.
Lex Fridman (1:42:05.700)
So if you want to have a building block
Lex Fridman (1:42:07.660)
of a certain shape, you can try to achieve it
Dmitry Korkin (1:42:10.540)
by, you know, introducing a new protein sequence
Lex Fridman (1:42:13.140)
and predicting, you know, how it will fold.
Lex Fridman (1:42:17.260)
So with that, I mean, it's a natural,
Lex Fridman (1:42:22.060)
one of the, you know, natural applications
Dmitry Korkin (1:42:25.820)
of these algorithms.
Lex Fridman (1:42:28.220)
Now, talking about engineering a virus.
Dmitry Korkin (1:42:34.140)
With machine learning.
Lex Fridman (1:42:35.140)
With machine learning, right?
Dmitry Korkin (1:42:36.380)
So, well, you know, so luckily for us,
Lex Fridman (1:42:41.740)
I mean, we don't have that much data, right?
Dmitry Korkin (1:42:46.780)
Yeah.
Lex Fridman (1:42:47.580)
We actually, right now, one of the projects
Dmitry Korkin (1:42:50.100)
that we are carrying on in the lab
Lex Fridman (1:42:53.700)
is we're trying to develop a machine learning algorithm
Dmitry Korkin (1:42:56.940)
that determines the,
Lex Fridman (1:42:59.300)
whether or not the current strain is pathogenic.
Lex Fridman (1:43:02.700)
And the current strain of the coronavirus.
Lex Fridman (1:43:04.620)
Of the virus.
Dmitry Korkin (1:43:06.100)
I mean, so there are applications to coronaviruses
Lex Fridman (1:43:08.980)
because we have strains of SARS COVID 2,
Dmitry Korkin (1:43:11.460)
also SARS COVID, MERS that are pathogenic,
Lex Fridman (1:43:14.580)
but we also have strains of other coronaviruses
Dmitry Korkin (1:43:17.620)
that are, you know, not pathogenic.
Lex Fridman (1:43:20.140)
I mean, the common cold viruses and, you know,
Lex Fridman (1:43:24.060)
some other ones, right?
Lex Fridman (1:43:25.580)
So, so pathogenic meaning spreading.
Dmitry Korkin (1:43:28.980)
Pathogenic means actually inflicting damage.
Lex Fridman (1:43:33.780)
Correct.
Dmitry Korkin (1:43:35.340)
There are also some, you know,
Lex Fridman (1:43:37.020)
seasonal versus pandemic strains of influenza, right?
Lex Fridman (1:43:41.780)
And determining the, what are the molecular determinant,
Lex Fridman (1:43:45.220)
right?
Lex Fridman (1:43:46.060)
So that are built in, into the protein sequence,
Lex Fridman (1:43:48.300)
into the gene sequence, right?
Dmitry Korkin (1:43:50.700)
So, and whether or not the machine learning
Lex Fridman (1:43:52.980)
can determine those, those components, right?
Dmitry Korkin (1:43:58.420)
Oh, interesting.
Lex Fridman (1:43:59.260)
So like using machine learning to do,
Dmitry Korkin (1:44:00.660)
that's really interesting to, to, to given,
Lex Fridman (1:44:03.940)
give the input is like what the entire,
Dmitry Korkin (1:44:07.380)
the protein sequence and then determine
Lex Fridman (1:44:09.740)
if this thing is going to be able to do damage
Dmitry Korkin (1:44:12.340)
to a biological system.
Lex Fridman (1:44:14.620)
Yeah.
Dmitry Korkin (1:44:15.900)
So, so I mean,
Lex Fridman (1:44:16.740)
It's a good machine learning,
Lex Fridman (1:44:17.580)
you're saying we don't have enough data for that?
Lex Fridman (1:44:19.380)
We, I mean, for, for this specific one, we do.
Dmitry Korkin (1:44:22.620)
We might actually, I have, you know,
Lex Fridman (1:44:24.460)
have to back up on this because we're still in the process.
Dmitry Korkin (1:44:27.260)
There was one work that appeared in bioarchive
Lex Fridman (1:44:31.660)
by Eugene Kunin, who is one of these, you know,
Dmitry Korkin (1:44:34.900)
pioneers in, in, in evolutionary genomics.
Lex Fridman (1:44:39.020)
And they tried to look at this, but, you know,
Dmitry Korkin (1:44:42.820)
the methods were sort of standard, you know,
Lex Fridman (1:44:46.060)
supervised learning methods.
Lex Fridman (1:44:48.620)
And now the question is, you know,
Lex Fridman (1:44:51.340)
can you advance it further by, by using, you know,
Lex Fridman (1:44:56.660)
not so standard methods, you know?
Lex Fridman (1:44:58.620)
So there's obviously a lot of hope in,
Dmitry Korkin (1:45:01.140)
in transfer learning where you can actually try to transfer
Lex Fridman (1:45:05.580)
the information that the machine learning learns about
Lex Fridman (1:45:08.060)
the proper protein sequences, right?
Lex Fridman (1:45:11.340)
And, you know, so, so there is some promise
Dmitry Korkin (1:45:16.140)
in going this direction, but if we have this,
Lex Fridman (1:45:18.740)
it would be extremely useful because then
Dmitry Korkin (1:45:21.060)
we could essentially forecast the potential mutations
Lex Fridman (1:45:24.100)
that would make the current strain
Dmitry Korkin (1:45:26.300)
more or less pathogenic.
Lex Fridman (1:45:27.900)
Anticipate, anticipate them from a vaccine development,
Dmitry Korkin (1:45:31.140)
for the treatment, antiviral drug development.
Lex Fridman (1:45:34.580)
That, that would be a very crucial task.
Lex Fridman (1:45:36.860)
But you could also use that system to then say,
Lex Fridman (1:45:42.180)
how would we potentially modify this virus
Lex Fridman (1:45:45.260)
to make it more pathogenic?
Lex Fridman (1:45:47.940)
This, that's true.
Dmitry Korkin (1:45:49.660)
That's true.
Lex Fridman (1:45:50.500)
And then, you know, the, again,
Lex Fridman (1:45:55.980)
the hope is, well, several things, right?
Lex Fridman (1:45:59.700)
So one is that, you know, it's,
Lex Fridman (1:46:02.140)
even if you design a, you know, a sequence, right?
Lex Fridman (1:46:06.780)
So to carry out the actual experimental biology,
Dmitry Korkin (1:46:12.540)
to ensure that all the components working, you know,
Lex Fridman (1:46:16.820)
is a completely different matter.
Dmitry Korkin (1:46:19.060)
Difficult process.
Lex Fridman (1:46:19.900)
Yes.
Dmitry Korkin (1:46:20.860)
Then the, you know, we've seen in the past,
Lex Fridman (1:46:24.420)
there could be some regulation of the moment
Dmitry Korkin (1:46:27.660)
the scientific community recognizes
Lex Fridman (1:46:30.420)
that it's now becoming no longer a sort of a fun puzzle
Dmitry Korkin (1:46:34.620)
to, you know, for machine learning.
Lex Fridman (1:46:36.660)
Could be open.
Dmitry Korkin (1:46:37.860)
Yeah, so then there might be some regulation.
Lex Fridman (1:46:40.420)
So I think back in, what, 2015, there was, you know,
Dmitry Korkin (1:46:45.780)
there was an issue on regulating the research
Lex Fridman (1:46:49.500)
on influenza strains, right?
Dmitry Korkin (1:46:52.700)
There were several groups, you know,
Lex Fridman (1:46:55.580)
used sort of the mutation analysis
Dmitry Korkin (1:46:58.060)
to determine whether or not this strain will jump
Lex Fridman (1:47:01.820)
from one species to another.
Lex Fridman (1:47:03.300)
And I think there was like a half a year moratorium
Lex Fridman (1:47:06.540)
on the research on the paper published
Dmitry Korkin (1:47:09.780)
until, you know, scientists, you know, analyzed it
Lex Fridman (1:47:13.580)
and decided that it's actually safe.
Dmitry Korkin (1:47:16.440)
I forgot what that's called.
Lex Fridman (1:47:17.620)
Something of function, test of function.
Dmitry Korkin (1:47:20.020)
Gain of function.
Lex Fridman (1:47:20.860)
Gain of function, yeah.
Dmitry Korkin (1:47:22.380)
Gain of function, loss of function, that's right.
Lex Fridman (1:47:24.020)
Sorry.
Dmitry Korkin (1:47:26.420)
It's like, let's watch this thing mutate for a while
Lex Fridman (1:47:29.620)
to see like, to see what kind of things we can observe.
Dmitry Korkin (1:47:33.780)
I guess I'm not so much worried
Lex Fridman (1:47:36.320)
about that kind of research if there's a lot of regulation
Lex Fridman (1:47:38.620)
and if it's done very well and with competence and seriously.
Lex Fridman (1:47:42.780)
I am more worried about kind of this, you know,
Dmitry Korkin (1:47:46.980)
the underlying aspect of this question
Lex Fridman (1:47:49.580)
is more like 50 years from now.
Dmitry Korkin (1:47:52.920)
Speaking to the Drake equation,
Lex Fridman (1:47:54.940)
one of the parameters in the Drake equation
Dmitry Korkin (1:47:57.300)
is how long civilizations last.
Lex Fridman (1:47:59.820)
And that seems to be the most important value actually
Dmitry Korkin (1:48:03.860)
for calculating if there's other alien
Lex Fridman (1:48:06.100)
intelligent civilizations out there.
Dmitry Korkin (1:48:08.040)
That's where there's most variability.
Lex Fridman (1:48:10.960)
Assuming like if life, if that percentage
Dmitry Korkin (1:48:15.060)
that life can emerge is like not zero,
Lex Fridman (1:48:19.380)
like if we're a super unique,
Dmitry Korkin (1:48:21.260)
then it's the how long we last
Lex Fridman (1:48:23.940)
is basically the most important thing.
Lex Fridman (1:48:26.180)
So from a selfish perspective,
Lex Fridman (1:48:29.020)
but also from a Drake equation perspective,
Dmitry Korkin (1:48:32.020)
I'm worried about our civilization lasting.
Lex Fridman (1:48:35.020)
And you kind of think about all the ways
Dmitry Korkin (1:48:37.620)
in which machine learning can be used
Lex Fridman (1:48:39.140)
to design greater weapons of destruction, right?
Lex Fridman (1:48:45.700)
And I mean, one way to ask that
Lex Fridman (1:48:48.620)
if you look sort of 50 years from now,
Dmitry Korkin (1:48:50.580)
a hundred years from now,
Lex Fridman (1:48:52.620)
would you be more worried about natural pandemics
Lex Fridman (1:48:55.780)
or engineered pandemics?
Lex Fridman (1:48:59.440)
Like who's the better designer of viruses,
Lex Fridman (1:49:02.640)
nature or humans if we look down the line?
Lex Fridman (1:49:05.980)
I think in my view, I would still be worried
Dmitry Korkin (1:49:10.140)
about the natural pandemics simply because I mean,
Lex Fridman (1:49:14.300)
the capacity of the nature producing this.
Lex Fridman (1:49:20.740)
It does pretty good job, right?
Lex Fridman (1:49:22.700)
Yes.
Lex Fridman (1:49:23.540)
And the motivation for using virus,
Lex Fridman (1:49:25.280)
engineering viruses as a weapon is a weird one
Dmitry Korkin (1:49:29.020)
because maybe you can correct me on this,
Lex Fridman (1:49:31.480)
but it seems very difficult to target a virus, right?
Dmitry Korkin (1:49:35.620)
The whole point of a weapon, the way a rocket works,
Lex Fridman (1:49:38.400)
if a starting point, you have an end point
Lex Fridman (1:49:40.100)
and you're trying to hit a target,
Lex Fridman (1:49:42.340)
to hit a target with a virus is very difficult.
Lex Fridman (1:49:44.660)
It's basically just, right?
Lex Fridman (1:49:47.980)
The target would be the human species.
Dmitry Korkin (1:49:51.940)
Oh man.
Lex Fridman (1:49:52.860)
Yeah, I have a hope in us.
Dmitry Korkin (1:49:54.780)
I'm forever optimistic that we will not,
Lex Fridman (1:49:58.260)
there's insufficient evil in the world
Dmitry Korkin (1:50:01.620)
to lead to that kind of destruction.
Lex Fridman (1:50:04.560)
Well, I also hope that, I mean, that's what we see.
Dmitry Korkin (1:50:07.780)
I mean, with the way we are getting connected,
Lex Fridman (1:50:11.780)
the world is getting connected.
Dmitry Korkin (1:50:14.460)
I think it helps for the world to become more transparent.
Lex Fridman (1:50:21.660)
Yeah.
Lex Fridman (1:50:22.560)
So the information spread is,
Lex Fridman (1:50:27.100)
I think it's one of the key things for the society
Dmitry Korkin (1:50:31.660)
to become more balanced one way or another.
Lex Fridman (1:50:36.460)
This is something that people disagree with me on,
Lex Fridman (1:50:38.340)
but I do think that the kind of secrecy
Lex Fridman (1:50:41.900)
that governments have.
Lex Fridman (1:50:43.460)
So you're kind of speaking more to the other aspects,
Lex Fridman (1:50:47.060)
like a research community being more open,
Dmitry Korkin (1:50:49.700)
companies are being more open.
Lex Fridman (1:50:52.160)
Government is still like,
Dmitry Korkin (1:50:55.900)
we're talking about like military secrets.
Lex Fridman (1:50:57.860)
I think military secrets of the kind
Dmitry Korkin (1:51:01.380)
that could destroy the world
Lex Fridman (1:51:03.700)
will become also a thing of the 20th century.
Dmitry Korkin (1:51:07.300)
It'll become more and more open.
Lex Fridman (1:51:09.320)
Yeah.
Dmitry Korkin (1:51:10.160)
I think nations will lose power in the 21st century,
Lex Fridman (1:51:13.220)
like lose sufficient power towards secrecies.
Dmitry Korkin (1:51:15.960)
Transparency is more beneficial than secrecy,
Lex Fridman (1:51:18.860)
but of course it's not obvious.
Dmitry Korkin (1:51:21.140)
Let's hope so.
Lex Fridman (1:51:22.180)
Let's hope so that the governments
Dmitry Korkin (1:51:27.180)
will become more transparent.
Lex Fridman (1:51:31.300)
What, so we last talked, I think in March or April,
Lex Fridman (1:51:35.260)
what have you learned?
Lex Fridman (1:51:36.740)
How has your philosophical, psychological,
Lex Fridman (1:51:40.460)
biological worldview changed since then?
Lex Fridman (1:51:43.820)
Or you've been studying it nonstop
Dmitry Korkin (1:51:46.100)
from a computational biology perspective.
Lex Fridman (1:51:48.900)
How has your understanding and thoughts about this virus
Lex Fridman (1:51:51.140)
changed over those months from the beginning to today?
Lex Fridman (1:51:54.460)
One thing that I was really amazed at
Lex Fridman (1:51:58.660)
how efficient the scientific community was.
Lex Fridman (1:52:03.140)
I mean, and even just judging on this very narrow domain
Dmitry Korkin (1:52:10.100)
of protein structure and understanding
Lex Fridman (1:52:13.060)
the structural characterization of this virus
Dmitry Korkin (1:52:17.600)
from the components point of view,
Lex Fridman (1:52:19.860)
whole virus point of view.
Dmitry Korkin (1:52:21.460)
If you look at SARS, something that happened less than 20,
Lex Fridman (1:52:31.020)
but close enough, 20 years ago,
Lex Fridman (1:52:34.980)
and you see what, when it happened,
Lex Fridman (1:52:38.500)
what was sort of the response by the scientific community,
Dmitry Korkin (1:52:42.460)
you see that the structure characterizations did a cure,
Lex Fridman (1:52:47.100)
but it took several years, right?
Dmitry Korkin (1:52:51.660)
Now the things that took several years,
Lex Fridman (1:52:54.940)
it's a matter of months, right?
Lex Fridman (1:52:56.900)
So we see that the research pop up.
Lex Fridman (1:53:01.620)
We are at the unprecedented level
Lex Fridman (1:53:03.940)
in terms of the sequencing, right?
Lex Fridman (1:53:05.980)
Never before we had a single virus sequence so many times,
Lex Fridman (1:53:10.980)
so which allows us to actually to trace very precisely
Lex Fridman (1:53:16.380)
the sort of the evolutionary nature of this virus,
Lex Fridman (1:53:21.380)
what happens, and it's not just this virus independently
Lex Fridman (1:53:24.540)
people, because our genotype influences also
Dmitry Korkin (1:53:27.420)
of everything, it's the sequence of this virus
Lex Fridman (1:53:31.540)
the evolution of this, it's always a host pathogen,
Dmitry Korkin (1:53:32.420)
linked, anchored to the specific geographic place
Lex Fridman (1:53:35.540)
core evolution that, you know,
Dmitry Korkin (1:53:36.420)
to specific
Lex Fridman (1:53:38.540)
it's not just the virus, it's the sequence of this virus,
Dmitry Korkin (1:53:41.540)
it's the sequence of this virus linked to the specific
Lex Fridman (1:53:44.540)
geographic place, it's the sequence of this virus
Dmitry Korkin (1:53:48.540)
linked to the specific geographic place to specific people,
Lex Fridman (1:53:52.540)
that, you know, occurs.
Dmitry Korkin (1:53:55.540)
It'd be cool if we also had a lot more data about,
Lex Fridman (1:53:58.540)
so that the spread of this virus, not maybe,
Dmitry Korkin (1:54:02.540)
well, it'd be nice if we had it for like contact tracing
Lex Fridman (1:54:06.540)
purposes for this virus, but it'd be also nice if we had it
Dmitry Korkin (1:54:09.540)
for the study for future viruses to be able to respond
Lex Fridman (1:54:12.540)
and so on, but it's already nice that we have geographical
Dmitry Korkin (1:54:15.540)
data and the basic data from individual humans, yeah.
Lex Fridman (1:54:18.540)
Exactly, no, I think contact tracing is obviously
Dmitry Korkin (1:54:22.540)
a key component in understanding
Lex Fridman (1:54:26.540)
the spread of this virus.
Lex Fridman (1:54:29.540)
There is also, there is a number of challenges, right?
Lex Fridman (1:54:31.540)
So XPRIZE is one of them, we
Dmitry Korkin (1:54:35.540)
just recently took a part of
Lex Fridman (1:54:39.540)
this competition, it's the prediction of the
Dmitry Korkin (1:54:43.540)
number of infections in different regions.
Lex Fridman (1:54:47.540)
Oh, sure.
Dmitry Korkin (1:54:48.540)
So, you know, obviously the AI
Lex Fridman (1:54:52.540)
is the main topic in those predictions.
Dmitry Korkin (1:54:55.540)
Yeah, but it's still, the data, I mean, that's a competition,
Lex Fridman (1:54:59.540)
but the data is weak
Dmitry Korkin (1:55:03.540)
on the training. Like, it's great,
Lex Fridman (1:55:07.540)
it's much more than probably before, but like, it'd be nice if it was like
Dmitry Korkin (1:55:11.540)
really rich. I talked to Michael Mina from
Lex Fridman (1:55:15.540)
Harvard, I mean, he dreams that the community comes together with like a
Dmitry Korkin (1:55:19.540)
weather map to where viruses, right, like
Lex Fridman (1:55:23.540)
really high resolution sensors on like how
Lex Fridman (1:55:27.540)
from person to person the viruses that travel, all the different kinds of viruses, right?
Lex Fridman (1:55:31.540)
Because there's a ton of them, and then you'd be able to tell
Dmitry Korkin (1:55:35.540)
the story that you've spoken about
Lex Fridman (1:55:39.540)
of the evolution of these viruses, like day to day mutations that
Dmitry Korkin (1:55:43.540)
are occurring. I mean, that'd be fascinating just from a perspective of
Lex Fridman (1:55:47.540)
study and from the perspective of being able to respond to future pandemics.
Dmitry Korkin (1:55:51.540)
That's ultimately what I'm worried about. People love
Lex Fridman (1:55:55.540)
books. Is there some three
Dmitry Korkin (1:55:59.540)
or whatever number of books, technical, fiction, philosophical, that
Lex Fridman (1:56:03.540)
brought you joy in life, had an impact on your life,
Lex Fridman (1:56:07.540)
and maybe some that you would recommend others?
Lex Fridman (1:56:11.540)
I'll give you three very different books, and I also have a special runner up.
Dmitry Korkin (1:56:15.540)
Honorable mention.
Lex Fridman (1:56:19.540)
I mean, it's an audiobook, and that's
Dmitry Korkin (1:56:23.540)
some specific reason behind it. So the first book is
Lex Fridman (1:56:27.540)
something that sort of impacted my earlier
Dmitry Korkin (1:56:31.540)
stage of life, and I'm probably not going to be very original here.
Lex Fridman (1:56:35.540)
It's Bulgakov's Master and Margarita.
Dmitry Korkin (1:56:39.540)
For a Russian, maybe it's not super original,
Lex Fridman (1:56:43.540)
but it's a really powerful book, even in English.
Dmitry Korkin (1:56:47.540)
It is incredibly powerful, and
Lex Fridman (1:56:51.540)
I mean, the way it ends.
Dmitry Korkin (1:56:55.540)
I still have goosebumps when I read
Lex Fridman (1:56:59.540)
the very last sort of, it's called prologue, where
Lex Fridman (1:57:03.540)
it's just so powerful. What impact did it have on you? What ideas?
Lex Fridman (1:57:07.540)
What insights did you get from it? I was just taken by
Dmitry Korkin (1:57:11.540)
the fact that
Lex Fridman (1:57:15.540)
you have those parallel lives
Dmitry Korkin (1:57:19.540)
apart from many centuries, and
Lex Fridman (1:57:23.540)
somehow they got sort of intertwined into
Dmitry Korkin (1:57:27.540)
one story, and that
Lex Fridman (1:57:31.540)
to me was fascinating. And of course
Dmitry Korkin (1:57:35.540)
the romantic part of this book is like
Lex Fridman (1:57:39.540)
it's not just romance, it's like the romance
Lex Fridman (1:57:43.540)
empowered by sort of magic, right?
Lex Fridman (1:57:47.540)
And maybe on top of that, you have some irony,
Dmitry Korkin (1:57:51.540)
which is unavoidable, right? Because it was that
Lex Fridman (1:57:55.540)
Soviet time. But it's very deeply Russian, so that's
Dmitry Korkin (1:57:59.540)
the wit, the humor, the pain, the love,
Lex Fridman (1:58:03.540)
all of that is one of the books that kind of captures
Dmitry Korkin (1:58:07.540)
something about Russian culture that people outside of Russia
Lex Fridman (1:58:11.540)
should probably read. I agree. What's the second one? So the second one
Dmitry Korkin (1:58:15.540)
is again another one that it happened
Lex Fridman (1:58:19.540)
I read it later in my life. I think I read it
Dmitry Korkin (1:58:23.540)
first time when I was a graduate student.
Lex Fridman (1:58:27.540)
And that's the Solzhenitsyn's Cancer Word.
Dmitry Korkin (1:58:31.540)
That is amazingly powerful book.
Lex Fridman (1:58:35.540)
What is it about? It's about, I mean, essentially
Dmitry Korkin (1:58:39.540)
based on Solzhenitsyn was
Lex Fridman (1:58:43.540)
diagnosed with cancer when he was reasonably young, and he
Dmitry Korkin (1:58:47.540)
made a full recovery. So this is
Lex Fridman (1:58:51.540)
about a person who was sentenced
Dmitry Korkin (1:58:55.540)
for life in one of these camps.
Lex Fridman (1:58:59.540)
And he had some cancer,
Lex Fridman (1:59:03.540)
so he was transported back to one of these
Lex Fridman (1:59:07.540)
Soviet republics, I think it was
Dmitry Korkin (1:59:11.540)
South Asian republics. And the
Lex Fridman (1:59:15.540)
book is about
Dmitry Korkin (1:59:19.540)
his experience being a
Lex Fridman (1:59:23.540)
prisoner, being a patient in the
Dmitry Korkin (1:59:27.540)
cancer clinic, in the cancer ward, surrounded
Lex Fridman (1:59:31.540)
by people, many of which die.
Lex Fridman (1:59:35.540)
But in the way
Lex Fridman (1:59:39.540)
it reads, first of all, later on I
Dmitry Korkin (1:59:43.540)
read the accounts of the doctors
Lex Fridman (1:59:47.540)
who describe the experiences
Dmitry Korkin (1:59:51.540)
in the book by the
Lex Fridman (1:59:55.540)
patient as incredibly accurate.
Lex Fridman (1:59:59.540)
So I read that there was some doctor saying that
Lex Fridman (20:01.440)
virus security perspective,
Lex Fridman (20:02.960)
that's the best attack vector?
Lex Fridman (20:05.160)
Is, or like, that's a promising attack vector?
Dmitry Korkin (20:08.440)
I would say, yeah.
Lex Fridman (20:09.280)
So, I mean, there's still tons of research needs to be,
Dmitry Korkin (20:12.680)
you know, to be done.
Lex Fridman (20:14.000)
But yes, I think, you know, so.
Dmitry Korkin (20:16.560)
There's more attack surface, I guess.
Lex Fridman (20:18.880)
More attack surface.
Dmitry Korkin (20:19.880)
But, you know, from our analysis,
Lex Fridman (20:22.280)
from other evolutionary analysis,
Dmitry Korkin (20:24.200)
this protein is evolutionarily more stable
Lex Fridman (20:28.000)
compared to the, say, to the spike protein.
Lex Fridman (20:31.200)
Oh, and stable means a more static target?
Lex Fridman (20:35.520)
Well, yeah, so it doesn't change.
Dmitry Korkin (20:38.400)
It doesn't evolve from the evolutionary perspective
Lex Fridman (20:42.120)
so drastically as, for example, the spike protein.
Dmitry Korkin (20:46.000)
There's a bunch of stuff in the news
Lex Fridman (20:47.960)
about mutations of the virus in the United Kingdom.
Dmitry Korkin (20:51.400)
I also saw in South Africa something.
Lex Fridman (20:54.160)
Maybe that was yesterday.
Dmitry Korkin (20:56.360)
You just kind of mentioned about stability and so on.
Lex Fridman (21:00.200)
Which aspects of this are mutatable
Lex Fridman (21:02.800)
and which aspects, if mutated, become more dangerous?
Lex Fridman (21:07.600)
And maybe even zooming out,
Lex Fridman (21:09.280)
what are your thoughts and knowledge and ideas
Lex Fridman (21:12.080)
about the way it's mutated,
Lex Fridman (21:13.680)
all the news that we've been hearing?
Lex Fridman (21:15.360)
Are you worried about it from a biological perspective?
Lex Fridman (21:18.520)
Are you worried about it from a human perspective?
Lex Fridman (21:21.280)
So, I mean, you know, mutations are sort of a general way
Lex Fridman (21:26.320)
for these viruses to evolve, right?
Lex Fridman (21:28.640)
So, it's, you know, it's essentially,
Dmitry Korkin (21:32.680)
this is the way they evolve.
Lex Fridman (21:34.760)
This is the way they were able to jump
Dmitry Korkin (21:38.680)
from one species to another.
Lex Fridman (21:42.080)
We also see some recent jumps.
Dmitry Korkin (21:46.800)
There were some incidents of this virus jumping
Lex Fridman (21:50.000)
from human to dogs.
Dmitry Korkin (21:51.880)
So, you know, there is some danger in those jumps
Lex Fridman (21:55.880)
because every time it jumps, it also mutates, right?
Dmitry Korkin (21:59.520)
So, when it jumps to the species
Lex Fridman (22:04.400)
and jumps back, right?
Dmitry Korkin (22:06.160)
So, it acquires some mutations
Lex Fridman (22:08.320)
that are sort of driven by the environment
Lex Fridman (22:14.360)
of a new host, right?
Lex Fridman (22:16.360)
And it's different from the human environment.
Lex Fridman (22:19.280)
And so, we don't know whether the mutations
Lex Fridman (22:21.480)
that are acquired in the new species
Dmitry Korkin (22:24.920)
are neutral with respect to the human host
Lex Fridman (22:28.160)
or maybe, you know, maybe damaging.
Dmitry Korkin (22:32.080)
Yeah, change is always scary, but so are you worried about,
Lex Fridman (22:36.560)
I mean, it seems like because the spread is,
Dmitry Korkin (22:38.960)
during winter now, seems to be exceptionally high
Lex Fridman (22:43.560)
and especially with a vaccine just around the corner
Dmitry Korkin (22:46.760)
already being actually deployed,
Lex Fridman (22:49.160)
is there some worry that this puts evolutionary pressure,
Lex Fridman (22:53.000)
selective pressure on the virus for it to mutate?
Lex Fridman (22:59.000)
Is that a source of worry?
Dmitry Korkin (23:00.440)
Well, I mean, there is always this thought
Lex Fridman (23:03.440)
in the scientist's mind, you know, what will happen, right?
Dmitry Korkin (23:08.720)
So, I know there've been discussions
Lex Fridman (23:12.520)
about sort of the arms race between the ability
Dmitry Korkin (23:17.600)
of the humanity to get vaccinated faster
Lex Fridman (23:22.600)
than the virus, you know, essentially, you know,
Dmitry Korkin (23:27.600)
it becomes, you know, resistant to the vaccine.
Lex Fridman (23:34.200)
I mean, I don't worry that much simply because,
Dmitry Korkin (23:40.920)
you know, there is not that much evidence to that.
Lex Fridman (23:44.920)
To aggressive mutation around the vaccine.
Dmitry Korkin (23:47.440)
Exactly, you know, obviously there are mutations
Lex Fridman (23:49.960)
around the vaccine, so the reason we get vaccinated
Lex Fridman (23:56.080)
every year against the seasonal mutations, right?
Lex Fridman (24:01.280)
But, you know, I think it's important to study it.
Lex Fridman (24:06.120)
No doubts, right?
Lex Fridman (24:07.120)
So, I think one of the, you know, to me,
Lex Fridman (24:10.120)
and again, I might be biased because, you know,
Lex Fridman (24:14.120)
we've been trying to do that as well,
Dmitry Korkin (24:17.120)
so, but one of the critical directions
Lex Fridman (24:20.120)
in understanding the virus is to understand its evolution
Dmitry Korkin (24:23.920)
in order to sort of understand the mechanisms,
Lex Fridman (24:27.480)
the key mechanisms that lead the virus to jump,
Dmitry Korkin (24:30.960)
you know, the Nordic viruses to jump from species,
Lex Fridman (24:34.240)
from species to another, that the mechanisms
Dmitry Korkin (24:37.440)
that lead the virus to become resistant to vaccines,
Lex Fridman (24:42.480)
also to treatments, right?
Lex Fridman (24:44.680)
And hopefully that knowledge will enable us
Lex Fridman (24:48.520)
to sort of forecast the evolutionary traces,
Dmitry Korkin (24:52.520)
the future evolutionary traces of this virus.
Lex Fridman (24:55.160)
I mean, what, from a biological perspective,
Dmitry Korkin (24:58.080)
this might be a dumb question,
Lex Fridman (24:59.320)
but is there parts of the virus that if souped up,
Dmitry Korkin (25:05.080)
like through mutation, could make it more effective
Lex Fridman (25:09.080)
at doing its job?
Dmitry Korkin (25:09.920)
We're talking about this specific coronavirus
Lex Fridman (25:12.520)
because we were talking about the different, like,
Dmitry Korkin (25:14.880)
the membrane, the M protein, the E protein,
Lex Fridman (25:18.440)
the N and the S, the spike, is there some?
Lex Fridman (25:24.080)
And there are 20 or so more in addition to that.
Lex Fridman (25:27.880)
But is that a dumb way to look at it?
Dmitry Korkin (25:29.840)
Like, which of these, if mutated,
Lex Fridman (25:34.520)
could have the greatest impact, potentially damaging impact,
Lex Fridman (25:39.640)
on the effectiveness of the virus?
Lex Fridman (25:41.520)
So it's actually, it's a very good question
Dmitry Korkin (25:44.520)
because, and the short answer is, we don't know yet.
Lex Fridman (25:48.120)
But of course there is capacity of this virus
Dmitry Korkin (25:51.560)
to become more efficient.
Lex Fridman (25:53.560)
The reason for that is, you know,
Lex Fridman (25:56.680)
so if you look at the virus, I mean, it's a machine, right?
Lex Fridman (25:59.760)
So it's a machine that does a lot of different functions,
Lex Fridman (26:03.520)
and many of these functions are sort of nearly perfect,
Lex Fridman (26:06.520)
but they're not perfect.
Lex Fridman (26:07.840)
And those mutations can have the greatest impact
Lex Fridman (26:11.360)
and make those functions more perfect.
Dmitry Korkin (26:14.120)
For example, the attachment to ACE2 receptor, right,
Lex Fridman (26:18.240)
of the spike, right?
Dmitry Korkin (26:19.400)
So, you know, has this virus reached the efficiency
Lex Fridman (26:28.360)
in which the attachment is carried out?
Dmitry Korkin (26:31.560)
Or there are some mutations that still to be discovered,
Lex Fridman (26:36.080)
right, that will make this attachment sort of stronger,
Dmitry Korkin (26:41.920)
or, you know, something more, in a way more efficient
Lex Fridman (26:48.560)
from the point of view of this virus functioning.
Dmitry Korkin (26:51.880)
That's sort of the obvious example.
Lex Fridman (26:54.640)
But if you look at each of these proteins,
Dmitry Korkin (26:57.480)
I mean, it's there for a reason,
Lex Fridman (26:58.840)
it performs certain function.
Lex Fridman (27:00.760)
And it could be that certain mutations will, you know,
Lex Fridman (27:07.120)
enhance this function.
Dmitry Korkin (27:08.480)
It could be that some mutations will make this function
Lex Fridman (27:11.560)
much less efficient, right?
Lex Fridman (27:13.720)
So that's also the case.
Lex Fridman (27:16.200)
Let's, since we're talking about the evolutionary history
Dmitry Korkin (27:18.880)
of a virus, let's zoom back out
Lex Fridman (27:22.720)
and look at the evolution of proteins.
Dmitry Korkin (27:25.240)
I glanced at this 2010 Nature paper
Lex Fridman (27:29.960)
on the quote, ongoing expansion of the protein universe.
Lex Fridman (27:34.320)
And then, you know, it kind of implies and talks about
Lex Fridman (27:39.480)
that proteins started with a common ancestor,
Dmitry Korkin (27:42.520)
which is, you know, kind of interesting.
Lex Fridman (27:44.720)
It's interesting to think about like,
Dmitry Korkin (27:45.960)
even just like the first organic thing
Lex Fridman (27:49.840)
that started life on Earth.
Lex Fridman (27:51.840)
And from that, there's now, you know, what is it?
Lex Fridman (27:56.000)
3.5 billion years later, there's now millions of proteins.
Lex Fridman (27:59.880)
And they're still evolving.
Lex Fridman (28:01.320)
And that's, you know, in part,
Dmitry Korkin (28:02.960)
one of the things that you're researching.
Lex Fridman (28:05.000)
Is there something interesting to you about the evolution
Lex Fridman (28:09.200)
of proteins from this initial ancestor to today?
Lex Fridman (28:14.600)
Is there something beautiful and insightful
Lex Fridman (28:16.280)
about this long story?
Lex Fridman (28:18.120)
So I think, you know, if I were to pick a single keyword
Dmitry Korkin (28:24.120)
about protein evolution, I would pick modularity,
Lex Fridman (28:29.120)
something that we talked about in the beginning.
Lex Fridman (28:32.720)
And that's the fact that the proteins are no longer
Lex Fridman (28:36.960)
considered as, you know, as a sequence of letters.
Dmitry Korkin (28:41.280)
There are hierarchical complexities
Lex Fridman (28:45.880)
in the way these proteins are organized.
Lex Fridman (28:48.240)
And these complexities are actually going
Lex Fridman (28:51.720)
beyond the protein sequence.
Dmitry Korkin (28:53.920)
It's actually going all the way back to the gene,
Lex Fridman (28:57.720)
to the nucleotide sequence.
Lex Fridman (29:00.000)
And so, you know, again, these protein domains,
Lex Fridman (29:04.840)
they are not only functional building blocks,
Dmitry Korkin (29:07.840)
they are also evolutionary building blocks.
Lex Fridman (29:09.920)
And so what we see in the sort of,
Dmitry Korkin (29:12.560)
in the later stages of evolution,
Lex Fridman (29:15.120)
I mean, once this stable structurally
Lex Fridman (29:18.720)
and functionally building blocks were discovered,
Lex Fridman (29:22.040)
they essentially, they stay, those domains stay as such.
Lex Fridman (29:28.040)
So that's why if you start comparing different proteins,
Lex Fridman (29:31.560)
you will see that many of them will have similar fragments.
Lex Fridman (29:37.400)
And those fragments will correspond to something
Lex Fridman (29:39.640)
that we call protein domain families.
Lex Fridman (29:42.280)
And so they are still different
Lex Fridman (29:44.040)
because you still have mutations and, you know,
Dmitry Korkin (29:48.520)
the, you know, different mutations are attributed to,
Lex Fridman (29:53.200)
to, you know, diversification of the function
Dmitry Korkin (29:56.200)
of this, you know, protein domains.
Lex Fridman (29:58.840)
However, you don't, you very rarely see, you know,
Dmitry Korkin (2:00:03.540)
every single doctor should read this book to understand
Lex Fridman (2:00:07.540)
what the patient feels. But
Dmitry Korkin (2:00:11.540)
again, as many of the Solzhenitsyn's
Lex Fridman (2:00:15.540)
books, it has multiple levels of complexity.
Lex Fridman (2:00:19.540)
And obviously if you look above
Lex Fridman (2:00:23.540)
the cancer and the patient, the
Dmitry Korkin (2:00:27.540)
tumor that was growing and then disappeared
Lex Fridman (2:00:31.540)
in his
Dmitry Korkin (2:00:35.540)
body with some consequences, this is
Lex Fridman (2:00:39.540)
allegorically the
Dmitry Korkin (2:00:43.540)
Soviet, and he actually
Lex Fridman (2:00:47.540)
when he was asked, he said that this is what made him
Dmitry Korkin (2:00:51.540)
think about this, how to combine these experiences.
Lex Fridman (2:00:55.540)
Him being a part of the Soviet regime,
Dmitry Korkin (2:00:59.540)
also being a part of the
Lex Fridman (2:01:03.540)
someone sent to Gulag camp,
Lex Fridman (2:01:07.540)
and also someone who experienced cancer
Lex Fridman (2:01:11.540)
in his life. The Gulag Archipelago
Lex Fridman (2:01:15.540)
and this book, these are the works that actually made him
Lex Fridman (2:01:19.540)
receive a Nobel Prize. But to me
Dmitry Korkin (2:01:23.540)
I've read
Lex Fridman (2:01:27.540)
other books by Solzhenitsyn.
Dmitry Korkin (2:01:31.540)
This one to me is the most powerful one.
Lex Fridman (2:01:35.540)
And by the way, both this one and the previous one you read in Russian?
Dmitry Korkin (2:01:39.540)
Yes. So now there is the third book is an English book
Lex Fridman (2:01:43.540)
and it's completely different. So we're switching the gears
Dmitry Korkin (2:01:47.540)
completely. So this is the book which, it's not even
Lex Fridman (2:01:51.540)
a book, it's an essay by
Dmitry Korkin (2:01:55.540)
Jonathan Neumann called The Computer and the Brain.
Lex Fridman (2:01:59.540)
And that was the book he was writing
Dmitry Korkin (2:02:03.540)
knowing that he was dying of cancer.
Lex Fridman (2:02:07.540)
So the book was released back, it's a very thin book.
Lex Fridman (2:02:11.540)
But the power,
Lex Fridman (2:02:15.540)
the intellectual power in this book, in this essay
Dmitry Korkin (2:02:19.540)
is incredible. I mean you probably know that von Neumann
Lex Fridman (2:02:23.540)
is considered to be one of the biggest
Dmitry Korkin (2:02:27.540)
thinkers. So his intellectual power was incredible.
Lex Fridman (2:02:31.540)
And you can actually feel this power
Dmitry Korkin (2:02:35.540)
in this book where the person is writing knowing that he will be,
Lex Fridman (2:02:39.540)
he will die. The book actually got published only after his
Dmitry Korkin (2:02:43.540)
death back in 1958. He died in 1957.
Lex Fridman (2:02:47.540)
So he tried to put as many
Dmitry Korkin (2:02:51.540)
ideas that he still
Lex Fridman (2:02:55.540)
hadn't realized.
Lex Fridman (2:02:59.540)
So this book is very difficult
Lex Fridman (2:03:03.540)
to read because every single paragraph
Dmitry Korkin (2:03:07.540)
is just compact, is
Lex Fridman (2:03:11.540)
filled with these ideas. And the ideas are incredible.
Dmitry Korkin (2:03:15.540)
Even nowadays, so he tried
Lex Fridman (2:03:19.540)
to put the parallels between the brain
Dmitry Korkin (2:03:23.540)
computing power, the neural system, and the computers
Lex Fridman (2:03:27.540)
as they were understood. Do you remember what year he was working on this?
Dmitry Korkin (2:03:31.540)
57. 57. So that was right during his,
Lex Fridman (2:03:35.540)
when he was diagnosed with cancer and he was essentially...
Dmitry Korkin (2:03:39.540)
Yeah, he's one of those, there's a few folks people mention,
Lex Fridman (2:03:43.540)
I think Ed Witten is another that like
Dmitry Korkin (2:03:47.540)
everyone that meets them, they say he's just an intellectual powerhouse.
Lex Fridman (2:03:51.540)
Yes. Okay, so who's the honorable mention?
Lex Fridman (2:03:55.540)
And this is, I mean, the reason I put it sort of in a separate section
Lex Fridman (2:03:59.540)
because this is a book that I recently
Dmitry Korkin (2:04:03.540)
listened to. So it's an audio book.
Lex Fridman (2:04:07.540)
And this is a book called Lab Girl by Hope Jarron.
Lex Fridman (2:04:11.540)
So Hope Jarron, she is a
Lex Fridman (2:04:15.540)
scientist, she's a geochemist that essentially
Dmitry Korkin (2:04:19.540)
studies the
Lex Fridman (2:04:23.540)
fossil plants. And so she uses
Dmitry Korkin (2:04:27.540)
this fossil plant, the chemical analysis to understand
Lex Fridman (2:04:31.540)
what was the climate back in
Dmitry Korkin (2:04:35.540)
a thousand years, hundreds of thousands of years ago.
Lex Fridman (2:04:39.540)
And so something that incredibly
Dmitry Korkin (2:04:43.540)
touched me by this book, it was narrated by the author.
Lex Fridman (2:04:47.540)
Nice. And it's an incredibly
Dmitry Korkin (2:04:51.540)
personal story, incredibly. So
Lex Fridman (2:04:55.540)
certain parts of the book, you could actually hear the author crying.
Lex Fridman (2:04:59.540)
And that to me, I mean, I never experienced
Lex Fridman (2:05:03.540)
anything like this, reading the book, but it was like
Dmitry Korkin (2:05:07.540)
the connection between you and the author.
Lex Fridman (2:05:11.540)
And I think this is really
Dmitry Korkin (2:05:15.540)
a must read, but even better, a must listen
Lex Fridman (2:05:19.540)
to audio book for anyone who
Dmitry Korkin (2:05:23.540)
wants to learn about sort of
Lex Fridman (2:05:27.540)
academia, science, research in general, because it's
Dmitry Korkin (2:05:31.540)
a very personal account about her becoming
Lex Fridman (2:05:35.540)
a scientist. So
Dmitry Korkin (2:05:39.540)
we're just before New Year's.
Lex Fridman (2:05:43.540)
We talked a lot about some difficult topics of viruses and so on.
Lex Fridman (2:05:47.540)
Do you have some exciting things you're looking forward
Lex Fridman (2:05:51.540)
to in 2021? Some New Year's resolutions,
Dmitry Korkin (2:05:55.540)
maybe silly or fun, or
Lex Fridman (2:05:59.540)
something very important and fundamental to
Lex Fridman (2:06:03.540)
the world of science or something completely unimportant?
Lex Fridman (2:06:07.540)
Well, I'm definitely looking forward to
Dmitry Korkin (2:06:11.540)
towards things becoming normal.
Lex Fridman (2:06:15.540)
So yes, I really miss traveling.
Dmitry Korkin (2:06:19.540)
Every summer I go
Lex Fridman (2:06:23.540)
to an international summer school. It's called
Dmitry Korkin (2:06:27.540)
the School for Molecular and Theoretical Biology. It's held in Europe.
Lex Fridman (2:06:31.540)
It's organized by very good friends of mine. And this is
Dmitry Korkin (2:06:35.540)
the school for gifted kids from all over the world, and
Dmitry Korkin (2:06:39.540)
they're incredibly bright. It's like every time I go there, it's like, you know,
Dmitry Korkin (2:06:43.540)
it's a highlight of the year. And
Lex Fridman (2:06:47.540)
we couldn't make it this August, so we
Dmitry Korkin (2:06:51.540)
did this school remotely, but it's different.
Lex Fridman (2:06:55.540)
So I am definitely looking forward to next August
Dmitry Korkin (2:06:59.540)
coming there. One of
Lex Fridman (2:07:03.540)
my personal resolutions, I realized that
Dmitry Korkin (2:07:07.540)
being in the house and working from home,
Lex Fridman (2:07:11.540)
I realized that actually
Dmitry Korkin (2:07:15.540)
I apparently missed a lot
Lex Fridman (2:07:19.540)
spending time with my family,
Dmitry Korkin (2:07:23.540)
believe it or not. So you typically, with all the
Lex Fridman (2:07:27.540)
research and teaching and
Dmitry Korkin (2:07:31.540)
everything related to the academic life,
Lex Fridman (2:07:35.540)
I mean, you get distracted. And so
Dmitry Korkin (2:07:39.540)
you don't feel that
Lex Fridman (2:07:43.540)
the fact that you are away from your family doesn't affect you
Dmitry Korkin (2:07:47.540)
because you are naturally distracted by other things.
Lex Fridman (2:07:51.540)
So this time I realized that
Dmitry Korkin (2:07:55.540)
that's so important, right? Spending your time with
Lex Fridman (2:07:59.540)
the family, with your kids. And so that
Dmitry Korkin (2:08:03.540)
would be my new year resolution and actually trying to
Lex Fridman (2:08:07.540)
spend as much time as possible. Even when the world opens up.
Dmitry Korkin (2:08:11.540)
Yeah, that's a beautiful message. That's a beautiful reminder.
Lex Fridman (2:08:15.540)
I asked you if there's a Russian poem
Dmitry Korkin (2:08:19.540)
that I could read, that I could force you to read, and you said, okay, fine, sure.
Lex Fridman (2:08:23.540)
Do you mind reading?
Lex Fridman (2:08:27.540)
And you said that no paper needed.
Lex Fridman (2:08:31.540)
So this poem was written by my namesake,
Dmitry Korkin (2:08:35.540)
another Dmitry, Dmitry Kemerefeld.
Lex Fridman (2:08:39.540)
It's a recent poem and it's
Dmitry Korkin (2:08:43.540)
called Sorceress, Vyadma,
Lex Fridman (2:08:47.540)
in Russian, or actually
Dmitry Korkin (2:08:51.540)
Koldunya. So that's sort of another sort of connotation of
Lex Fridman (2:08:55.540)
sorceress or witch. And I really like it
Lex Fridman (2:08:59.540)
and it's one of just a handful poems I actually
Lex Fridman (2:09:03.540)
can recall by heart. I also have a very strong
Dmitry Korkin (2:09:07.540)
association when I read this poem with Master and
Lex Fridman (2:09:11.540)
Margarita, the main female character,
Dmitry Korkin (2:09:15.540)
Margarita. And also it's
Lex Fridman (2:09:19.540)
about, it's happening about the same time we're talking
Dmitry Korkin (2:09:23.540)
now, so around New Year,
Lex Fridman (2:09:27.540)
around Christmas. Do you mind reading it in Russian?
Dmitry Korkin (2:09:31.540)
I'll give it a try.
Lex Fridman (2:10:01.540)
So you narrowed your eyes,
Dmitry Korkin (2:10:05.540)
that anyone who was blessed
Lex Fridman (2:10:09.540)
was ready to give their soul to the devil
Dmitry Korkin (2:10:13.540)
for this witch's connection.
Lex Fridman (2:10:17.540)
And I, without prejudice,
Dmitry Korkin (2:10:21.540)
ran out to feel your
Lex Fridman (2:10:25.540)
amazing breath on your lips,
Dmitry Korkin (2:10:29.540)
to remember how you flew above the earth
Lex Fridman (2:10:33.540)
in a white view,
Dmitry Korkin (2:10:37.540)
in a white haze, in a white mist.
Lex Fridman (2:10:41.540)
That's beautiful. I love how it captures a moment of longing
Lex Fridman (2:10:45.540)
and maybe love even.
Lex Fridman (2:10:49.540)
Yes. To me it has a lot of meaning about
Dmitry Korkin (2:10:53.540)
this something that is happening,
Lex Fridman (2:10:57.540)
something that is far away, but still very close to you.
Lex Fridman (2:11:01.540)
And yes, it's the winter.
Lex Fridman (2:11:05.540)
There's something magical about winter, isn't there?
Dmitry Korkin (2:11:09.540)
I don't know how to translate it, but a kiss in winter
Lex Fridman (2:11:13.540)
is interesting. Lips in winter and all that kind of stuff.
Dmitry Korkin (2:11:17.540)
It's beautiful. Russian has a way. It has a reason, Russian poetry
Lex Fridman (2:11:21.540)
is just, I'm a fan of poetry in both languages, but English
Dmitry Korkin (2:11:25.540)
doesn't capture some of the magic that Russian seems to, so
Lex Fridman (2:11:29.540)
thank you for doing that. That was awesome. Dmitry,
Dmitry Korkin (2:11:33.540)
it's great to talk to you again. It's contagious
Lex Fridman (2:11:37.540)
how much you love what you do, how much you love life, so I really appreciate
Dmitry Korkin (2:11:41.540)
you taking the time to talk today. And thank you for having me.
Dmitry Korkin (2:11:45.540)
Thanks for listening to this conversation with Dmitry Korkin, and thank you to our
Dmitry Korkin (2:11:49.540)
sponsors. Brave Browser, NetSuite Business Management
Lex Fridman (2:11:53.540)
Software, Magic Spoon Low Carb Cereal, and
Dmitry Korkin (2:11:57.540)
Asleep Self Cooling Mattress. So the choice is
Lex Fridman (2:12:01.540)
browsing privacy, business success, healthy diet, or comfortable
Dmitry Korkin (2:12:05.540)
sleep. Choose wisely, my friends. And if you wish,
Lex Fridman (2:12:09.540)
click the sponsor links below to get a discount and to support this podcast.
Lex Fridman (2:12:13.540)
And now, let me leave you with some words from Jeffrey Eugenides.
Lex Fridman (2:12:17.540)
Biology gives you a brain.
Dmitry Korkin (2:12:21.540)
Life turns it into a mind. Thank you for listening,
Lex Fridman (2:12:25.540)
and hope to see you next time.
Dmitry Korkin (30:03.520)
the evolutionary events that would split
Lex Fridman (30:07.840)
this domain into fragments because,
Lex Fridman (30:10.520)
and it's, you know, once you have the domain split,
Lex Fridman (30:17.240)
you actually, you, you know,
Dmitry Korkin (30:20.240)
you can completely cancel out its function
Lex Fridman (30:24.000)
or at the very least you can reduce it.
Lex Fridman (30:26.600)
And that's not, you know, efficient from the point of view
Lex Fridman (30:29.640)
of the, you know, of the cell functioning.
Dmitry Korkin (30:32.880)
So, so the, the, the protein domain level
Lex Fridman (30:37.240)
is a very important one.
Lex Fridman (30:39.160)
Now, on top of that, right?
Lex Fridman (30:42.040)
So if you look at the proteins, right,
Lex Fridman (30:44.120)
so you have this structural units
Lex Fridman (30:46.360)
and they carry out the function,
Lex Fridman (30:48.200)
but then much less is known about things
Lex Fridman (30:51.880)
that connect this protein domains,
Dmitry Korkin (30:54.400)
something that we call linkers.
Lex Fridman (30:56.400)
And those linkers are completely flexible, you know,
Dmitry Korkin (31:00.760)
parts of the protein that nevertheless
Lex Fridman (31:03.520)
carry out a lot of function.
Lex Fridman (31:06.360)
So it's like little tails, little heads.
Lex Fridman (31:08.040)
So, so, so we do have tails.
Lex Fridman (31:09.840)
So they're called termini, C and N termini.
Lex Fridman (31:12.320)
So these are things right on the, on, on, on one
Lex Fridman (31:17.160)
and another ends of the protein sequence.
Lex Fridman (31:20.040)
So they are also very important.
Lex Fridman (31:22.560)
So they, they attributed to very specific interactions
Lex Fridman (31:26.320)
between the proteins.
Dmitry Korkin (31:27.720)
So.
Lex Fridman (31:28.560)
But you're referring to the links between domains.
Dmitry Korkin (31:30.800)
That connect the domains.
Lex Fridman (31:32.600)
And, you know, apart from the, just the,
Dmitry Korkin (31:36.160)
the simple perspective, if you have, you know,
Lex Fridman (31:39.840)
a very short domain, you have, sorry, a very short linker,
Dmitry Korkin (31:43.720)
you have two domains next to each other.
Lex Fridman (31:45.880)
They are forced to be next to each other.
Dmitry Korkin (31:47.560)
If you have a very long one,
Lex Fridman (31:49.080)
you have the domains that are extremely flexible
Lex Fridman (31:52.040)
and they carry out a lot of sort of
Lex Fridman (31:54.320)
spatial reorganization, right?
Dmitry Korkin (31:56.880)
That's awesome.
Lex Fridman (31:58.120)
But on top of that, right, just this linker itself,
Dmitry Korkin (32:01.960)
because it's so flexible, it actually can adapt
Lex Fridman (32:05.760)
to a lot of different shapes.
Lex Fridman (32:07.480)
And therefore it's a, it's a very good interactor
Lex Fridman (32:11.080)
when it comes to interaction between this protein
Lex Fridman (32:14.000)
and other protein, right?
Lex Fridman (32:15.720)
So these things also evolve, you know,
Lex Fridman (32:18.920)
and they in a way have different sort of laws of
Lex Fridman (32:25.480)
the driving laws that underlie the evolution
Dmitry Korkin (32:30.600)
because they no longer need to,
Lex Fridman (32:33.400)
to preserve certain structure, right?
Dmitry Korkin (32:37.120)
Unlike protein domains.
Lex Fridman (32:38.880)
And so on top of that,
Dmitry Korkin (32:41.480)
you have something that is even less studied.
Lex Fridman (32:45.840)
And this is something that attribute to,
Dmitry Korkin (32:49.640)
to the concept of alternative splicing.
Lex Fridman (32:53.240)
So alternative splicing.
Lex Fridman (32:54.480)
So it's a, it's a very cool concept.
Lex Fridman (32:56.920)
It's something that we've been fascinated about for,
Dmitry Korkin (33:00.840)
you know, over a decade in my lab
Lex Fridman (33:03.520)
and trying to do research with that.
Lex Fridman (33:05.520)
But so, you know, so typically, you know,
Lex Fridman (33:08.080)
a simplistic perspective is that one gene
Lex Fridman (33:12.480)
is equal one protein product, right?
Lex Fridman (33:16.040)
So you have a gene, you know,
Dmitry Korkin (33:18.320)
you transcribe it and translate it
Lex Fridman (33:21.120)
and it becomes a protein.
Dmitry Korkin (33:24.600)
In reality, when we talk about eukaryotes,
Lex Fridman (33:28.360)
especially sort of more recent eukaryotes
Dmitry Korkin (33:32.320)
that are very complex,
Lex Fridman (33:33.800)
the gene is no longer equal to one protein.
Dmitry Korkin (33:40.200)
It actually can produce multiple functionally,
Lex Fridman (33:47.040)
you know, active protein products.
Lex Fridman (33:50.280)
And each of them is, you know,
Lex Fridman (33:52.720)
is called an alternatively spliced product.
Dmitry Korkin (33:57.040)
The reason it happens is that if you look at the gene,
Lex Fridman (34:00.960)
it actually has, it has also blocks.
Lex Fridman (34:05.560)
And the blocks, some of which,
Lex Fridman (34:08.320)
and it's essentially, it goes like this.
Lex Fridman (34:10.680)
So we have a block that will later be translated.
Lex Fridman (34:13.880)
We call it exon.
Dmitry Korkin (34:15.040)
Then we'll have a block that is not translated, cut out.
Lex Fridman (34:19.240)
We call it intron.
Lex Fridman (34:20.400)
So we have exon, intron, exon, intron,
Lex Fridman (34:22.840)
et cetera, et cetera, et cetera, right?
Lex Fridman (34:24.120)
So sometimes you can have, you know,
Lex Fridman (34:26.920)
dozens of these exons and introns.
Lex Fridman (34:29.880)
So what happens is during the process
Lex Fridman (34:32.680)
when the gene is converted to RNA,
Dmitry Korkin (34:37.320)
we have things that are cut out,
Lex Fridman (34:41.280)
the introns that are cut out,
Lex Fridman (34:43.240)
and exons that now get assembled together.
Lex Fridman (34:47.160)
And sometimes we will throw out some of the exons
Lex Fridman (34:52.320)
and the remaining protein product will become
Lex Fridman (34:54.600)
still be the same.
Dmitry Korkin (34:55.440)
Different.
Lex Fridman (34:56.280)
Oh, different.
Lex Fridman (34:57.120)
So now you have fragments of the protein
Lex Fridman (34:59.960)
that no longer there.
Dmitry Korkin (35:01.360)
They were cut out with the introns.
Lex Fridman (35:03.800)
Sometimes you will essentially take one exon
Lex Fridman (35:07.520)
and replace it with another one, right?
Lex Fridman (35:09.840)
So there's some flexibility in this process.
Lex Fridman (35:12.600)
So that creates a whole new level of complexity.
Lex Fridman (35:17.200)
Cause now.
Lex Fridman (35:18.040)
Is this random though?
Lex Fridman (35:18.880)
Is it random?
Dmitry Korkin (35:19.720)
It's not random.
Lex Fridman (35:20.840)
We, and this is where I think now the appearance
Dmitry Korkin (35:24.480)
of this modern single cell
Lex Fridman (35:27.360)
and before that tissue level sequencing,
Dmitry Korkin (35:31.240)
next generation sequencing techniques such as RNA seed
Lex Fridman (35:34.280)
allows us to see that these are the events
Dmitry Korkin (35:38.200)
that often happen in response.
Lex Fridman (35:41.040)
It's a dynamic event that happens in response
Dmitry Korkin (35:44.560)
to disease or in response
Lex Fridman (35:48.320)
to certain developmental stage of a cell.
Lex Fridman (35:51.800)
And this is an incredibly complex layer
Lex Fridman (35:56.840)
that also undergoes, I mean,
Lex Fridman (35:59.800)
because it's at the gene level, right?
Lex Fridman (36:01.560)
So it undergoes certain evolution, right?
Lex Fridman (36:05.440)
And now we have this interplay
Lex Fridman (36:08.680)
between what is happening in the protein world
Lex Fridman (36:12.720)
and what is happening in the gene and RNA world.
Lex Fridman (36:17.720)
And for example, it's often that we see
Dmitry Korkin (36:22.720)
that the boundaries of this exons coincide
Lex Fridman (36:28.200)
with the boundaries of the protein domains, right?
Lex Fridman (36:32.160)
So there is this close interplay to that.
Lex Fridman (36:36.520)
It's not always, I mean, otherwise it would be too simple,
Lex Fridman (36:39.280)
right?
Lex Fridman (36:40.120)
But we do see the connection
Dmitry Korkin (36:41.880)
between those sort of machineries.
Lex Fridman (36:45.000)
And obviously the evolution will pick up this complexity
Dmitry Korkin (36:49.760)
and, you know.
Lex Fridman (36:51.800)
Select for whatever is successful,
Dmitry Korkin (36:53.480)
whatever is interesting function.
Lex Fridman (36:55.040)
We see that complexity in play
Lex Fridman (36:57.560)
and makes this question more complex, but more exciting.
Lex Fridman (37:02.560)
Small detour, I don't know if you think about this
Dmitry Korkin (37:05.440)
into the world of computer science.
Lex Fridman (37:07.540)
There's a Douglas Hostetter, I think,
Dmitry Korkin (37:11.240)
came up with the name of Quine,
Lex Fridman (37:14.360)
which are, I don't know if you're familiar
Dmitry Korkin (37:16.180)
with these things, but it's computer programs
Lex Fridman (37:18.880)
that have, I guess, exon and intron,
Lex Fridman (37:22.160)
and they copy, the whole purpose of the program
Lex Fridman (37:24.800)
is to copy itself.
Lex Fridman (37:26.240)
So it prints copies of itself,
Lex Fridman (37:28.480)
but can also carry information inside of it.
Lex Fridman (37:30.980)
So it's a very kind of crude, fun exercise of,
Lex Fridman (37:36.420)
can we sort of replicate these ideas from cells?
Dmitry Korkin (37:40.000)
Can we have a computer program that when you run it,
Lex Fridman (37:42.940)
just print itself, the entirety of itself,
Lex Fridman (37:47.080)
and does it in different programming languages and so on.
Lex Fridman (37:50.040)
I've been playing around and writing them.
Dmitry Korkin (37:51.960)
It's a kind of fun little exercise.
Lex Fridman (37:53.720)
You know, when I was a kid, so you know,
Dmitry Korkin (37:55.720)
it was essentially one of the sort of main stages
Lex Fridman (38:02.860)
in informatics Olympiads that you have to reach
Dmitry Korkin (38:08.280)
in order to be any so good,
Lex Fridman (38:10.880)
is you should be able to write a program
Dmitry Korkin (38:14.400)
that replicates itself.
Lex Fridman (38:16.680)
And so the task then becomes even sort of more complicated.
Lex Fridman (38:20.920)
So what is the shortest program?
Lex Fridman (38:24.040)
And of course, it's a function of a programming language,
Lex Fridman (38:27.480)
but yeah, I remember a long, long, long time ago
Lex Fridman (38:30.940)
when we tried to make it short and short
Lex Fridman (38:34.840)
and find the shortcut.
Lex Fridman (38:36.520)
There's actually on a stack exchange, there's a entire site
Dmitry Korkin (38:41.720)
called CodeGolf, I think,
Lex Fridman (38:44.160)
where the entirety is just the competition.
Dmitry Korkin (38:46.560)
People just come up with whatever task, I don't know,
Lex Fridman (38:50.380)
like write code that reports the weather today.
Lex Fridman (38:54.680)
And the competition is about whatever programming language,
Lex Fridman (38:58.680)
what is the shortest program?
Lex Fridman (39:00.440)
And it makes you actually, people should check it out
Lex Fridman (39:02.280)
because it makes you realize
Dmitry Korkin (39:03.640)
there's some weird programming languages out there.
Lex Fridman (39:07.160)
But just to dig on that a little deeper,
Lex Fridman (39:12.640)
do you think, in computer science,
Lex Fridman (39:16.100)
we don't often think about programs,
Dmitry Korkin (39:19.280)
just like the machine learning world now,
Lex Fridman (39:22.280)
that's still kind of basic programs.
Lex Fridman (39:26.280)
And then there's humans that replicate themselves, right?
Lex Fridman (39:29.600)
And there's these mutations and so on.
Lex Fridman (39:31.440)
Do you think we'll ever have a world
Lex Fridman (39:34.520)
where there's programs that kind of
Lex Fridman (39:37.760)
have an evolutionary process?
Lex Fridman (39:40.640)
So I'm not talking about evolutionary algorithms,
Lex Fridman (39:42.640)
but I'm talking about programs that kind of
Lex Fridman (39:44.640)
mate with each other and evolve
Lex Fridman (39:46.480)
and like on their own replicate themselves.
Lex Fridman (39:49.600)
So this is kind of the idea here is,
Dmitry Korkin (39:54.640)
that's how you can have a runaway thing.
Lex Fridman (39:57.140)
So we think about machine learning as a system
Dmitry Korkin (39:59.240)
that gets smarter and smarter and smarter and smarter.
Lex Fridman (40:01.320)
At least the machine learning systems of today are like,
Dmitry Korkin (40:05.240)
it's a program that you can like turn off,
Lex Fridman (40:09.000)
as opposed to throwing a bunch of little programs out there
Lex Fridman (40:12.680)
and letting them like multiply and mate
Lex Fridman (40:15.560)
and evolve and replicate.
Lex Fridman (40:17.320)
Do you ever think about that kind of world,
Lex Fridman (40:20.400)
when we jump from the biological systems
Lex Fridman (40:23.360)
that you're looking at to artificial ones?
Lex Fridman (40:27.160)
I mean, it's almost like you take the sort of the area
Lex Fridman (40:32.440)
of intelligent agents, right?
Lex Fridman (40:34.360)
Which are essentially the independent sort of codes
Lex Fridman (40:38.640)
that run and interact and exchange the information, right?
Lex Fridman (40:42.480)
So I don't see why not.
Dmitry Korkin (40:45.120)
I mean, it could be sort of a natural evolution
Lex Fridman (40:48.760)
in this area of computer science.
Dmitry Korkin (40:52.880)
I think it's kind of an interesting possibility.
Lex Fridman (40:54.640)
It's terrifying too,
Lex Fridman (40:55.880)
but I think it's a really powerful tool.
Lex Fridman (40:58.360)
Like to have like agents that, you know,
Dmitry Korkin (41:00.680)
we have social networks with millions of people
Lex Fridman (41:02.800)
and they interact.
Dmitry Korkin (41:03.840)
I think it's interesting to inject into that,
Lex Fridman (41:05.720)
was already injected into that bots, right?
Lex Fridman (41:08.380)
But those bots are pretty dumb.
Lex Fridman (41:11.240)
You know, they're probably pretty dumb algorithms.
Dmitry Korkin (41:15.680)
You know, it's interesting to think
Lex Fridman (41:17.480)
that there might be bots that evolve together with humans.
Lex Fridman (41:20.440)
And there's the sea of humans and robots
Lex Fridman (41:23.960)
that are operating first in the digital space.
Lex Fridman (41:26.520)
And then you can also think, I love the idea.
Lex Fridman (41:29.080)
Some people worked, I think at Harvard, at Penn,
Dmitry Korkin (41:32.600)
there's robotics labs that, you know,
Lex Fridman (41:37.560)
take as a fundamental task to build a robot
Dmitry Korkin (41:40.600)
that given extra resources can build another copy of itself,
Lex Fridman (41:44.920)
like in the physical space,
Dmitry Korkin (41:46.560)
which is super difficult to do, but super interesting.
Lex Fridman (41:50.900)
I remember there's like research on robots
Dmitry Korkin (41:54.020)
that can build a bridge.
Lex Fridman (41:55.240)
So they make a copy of themselves
Lex Fridman (41:56.880)
and they connect themselves
Lex Fridman (41:57.960)
and the sort of like self building bridge
Dmitry Korkin (42:00.560)
based on building blocks.
Lex Fridman (42:02.380)
You can imagine like a building that self assembles.
Lex Fridman (42:05.640)
So it's basically self assembling structures
Lex Fridman (42:07.560)
from robotic parts.
Lex Fridman (42:10.620)
But it's interesting to, within that robot,
Lex Fridman (42:13.880)
add the ability to mutate
Lex Fridman (42:15.920)
and do all the interesting like little things
Lex Fridman (42:21.320)
that you're referring to in evolution
Dmitry Korkin (42:23.200)
to go from a single origin protein building block
Lex Fridman (42:26.320)
to like this weird complex.
Lex Fridman (42:28.920)
And if you think about this, I mean, you know,
Lex Fridman (42:30.960)
the bits and pieces are there, you know.
Lex Fridman (42:34.600)
So you mentioned the evolution algorithm, right?
Lex Fridman (42:37.040)
You know, so this is sort of,
Lex Fridman (42:38.520)
and maybe sort of the goal is in a way different, right?
Lex Fridman (42:43.520)
So the goal is to, you know, to essentially,
Lex Fridman (42:46.720)
to optimize your search, right?
Lex Fridman (42:50.080)
So, but sort of the ideas are there.
Lex Fridman (42:53.060)
So people recognize that, you know,
Lex Fridman (42:55.080)
that the recombination events lead to global changes
Dmitry Korkin (43:01.160)
in the search trajectories, the mutations event
Lex Fridman (43:04.440)
is a more refined, you know, step in the search.
Dmitry Korkin (43:09.080)
Then you have, you know, other sort of
Lex Fridman (43:14.220)
nature inspired algorithm, right?
Lex Fridman (43:16.480)
So one of the reasons that, you know,
Lex Fridman (43:19.580)
I think it's one of the funnest one
Lex Fridman (43:21.940)
is the slime based algorithm, right?
Lex Fridman (43:24.820)
So it's, I think the first was introduced
Dmitry Korkin (43:28.220)
by the Japanese group,
Lex Fridman (43:30.220)
where it was able to solve some pre complex problems.
Lex Fridman (43:35.220)
So that's, and then I think there are still a lot of things
Lex Fridman (43:43.340)
we've yet to, you know, borrow from the nature, right?
Lex Fridman (43:48.960)
So there are a lot of sort of ideas
Lex Fridman (43:52.020)
that nature, you know, gets to offer us that, you know,
Dmitry Korkin (43:56.740)
it's up to us to grab it and to, you know,
Lex Fridman (44:01.020)
get the best use of it.
Dmitry Korkin (44:02.140)
Including neural networks, you know, we have a very crude
Lex Fridman (44:06.380)
inspiration from nature on neural networks.
Dmitry Korkin (44:08.300)
Maybe there's other inspirations to be discovered
Lex Fridman (44:10.920)
in the brain or other aspects of the various systems,
Dmitry Korkin (44:16.280)
even like the immune system, the way it interplays.
Lex Fridman (44:20.140)
I recently started to understand that the,
Dmitry Korkin (44:22.580)
like the immune system has something to do
Lex Fridman (44:24.360)
with the way the brain operates.
Dmitry Korkin (44:26.020)
Like there's multiple things going on in there,
Lex Fridman (44:28.360)
which all of which are not modeled
Dmitry Korkin (44:30.500)
in artificial neural networks.
Lex Fridman (44:32.140)
And maybe if you throw a little bit of that biological spice
Dmitry Korkin (44:35.380)
in there, you'll come up with something, something cool.
Lex Fridman (44:39.020)
I'm not sure if you're familiar with the Drake equation
Dmitry Korkin (44:43.740)
that estimate, I just did a video on it yesterday
Lex Fridman (44:46.740)
because I wanted to give my own estimate of it.
Dmitry Korkin (44:49.280)
It's an equation that combines a bunch of factors
Lex Fridman (44:52.340)
to estimate how many alien civilizations are in the galaxy.
Dmitry Korkin (44:56.980)
I've heard about it, yes.
Lex Fridman (44:58.500)
So one of the interesting parameters, you know,
Dmitry Korkin (45:01.340)
it's like how many stars are born every year,
Lex Fridman (45:05.980)
how many planets are on average per star for this,
Lex Fridman (45:11.700)
how many habitable planets are there.
Lex Fridman (45:14.260)
And then the one that starts being really interesting
Dmitry Korkin (45:18.660)
is the probability that life emerges on a habitable planet.
Lex Fridman (45:24.740)
So like, I don't know if you think about,
Dmitry Korkin (45:27.900)
you certainly think a lot about evolution,
Lex Fridman (45:29.720)
but do you think about the thing
Dmitry Korkin (45:31.060)
which evolution doesn't describe,
Lex Fridman (45:32.520)
which is like the beginning of evolution, the origin of life.
Dmitry Korkin (45:36.620)
I think I put the probability of life developing
Lex Fridman (45:39.320)
in a habitable planet at 1%.
Dmitry Korkin (45:41.800)
This is very scientifically rigorous.
Lex Fridman (45:44.440)
Okay, well, first at a high level for the Drake equation,
Lex Fridman (45:48.740)
what would you put that percent at on earth?
Lex Fridman (45:51.660)
And in general, do you have something,
Lex Fridman (45:55.100)
do you have thoughts about how life might've started,
Lex Fridman (45:58.220)
you know, like the proteins being the first kind of,
Lex Fridman (46:01.100)
one of the early jumping points?
Lex Fridman (46:02.940)
Yeah, so I think back in 2018,
Dmitry Korkin (46:07.500)
there was a very exciting paper published in Nature
Lex Fridman (46:10.460)
where they found one of the simplest amino acids,
Dmitry Korkin (46:18.320)
glycine, in a comet dust.
Lex Fridman (46:23.320)
So this is, and I apologize if I don't pronounce,
Dmitry Korkin (46:29.440)
it's a Russian named comet,
Lex Fridman (46:31.840)
it's I think Chugryumov Gerasimenko.
Dmitry Korkin (46:34.760)
This is the comet where, and there was this mission
Lex Fridman (46:40.000)
to get close to this comet and get the stardust
Dmitry Korkin (46:46.320)
from its tail.
Lex Fridman (46:48.160)
And when scientists analyzed it,
Dmitry Korkin (46:50.620)
they actually found traces of, you know, of glycine,
Lex Fridman (46:56.640)
which, you know, makes up, you know,
Dmitry Korkin (46:59.400)
it's one of the basic, one of the 20 basic amino acids
Lex Fridman (47:04.180)
that makes up proteins, right?
Lex Fridman (47:06.400)
So that was kind of very exciting, right?
Lex Fridman (47:10.960)
But, you know, the question is very interesting, right?
Lex Fridman (47:14.220)
So what, you know, if there is some alien life,
Lex Fridman (47:18.540)
is it gonna be made of proteins, right?
Lex Fridman (47:22.940)
Or maybe RNAs, right?
Lex Fridman (47:24.340)
So we see that, you know, the RNA viruses are certainly,
Dmitry Korkin (47:29.120)
you know, very well established sort of, you know,
Lex Fridman (47:35.020)
group of molecular machines, right?
Dmitry Korkin (47:37.820)
So, yeah, it's a very interesting question.
Lex Fridman (47:42.140)
What probability would you put?
Lex Fridman (47:43.580)
Like, how hard is this job?
Lex Fridman (47:45.260)
Like, how unlikely just on Earth do you think
Lex Fridman (47:48.740)
this whole thing is that we got going?
Lex Fridman (47:51.600)
Like, are we really lucky or is it inevitable?
Dmitry Korkin (47:54.620)
Like, what's your sense when you sit back
Lex Fridman (47:56.240)
and think about life on Earth?
Lex Fridman (47:58.820)
Is it higher or lower than 1%?
Lex Fridman (48:00.980)
Well, because 1% is pretty low, but it still is like,
Dmitry Korkin (48:03.420)
damn, that's a pretty good chance.
Lex Fridman (48:05.060)
Yes, it's a pretty good chance.
Dmitry Korkin (48:06.600)
I mean, I would, personally, but again, you know,
Lex Fridman (48:10.580)
I'm, you know, probably not the best person
Dmitry Korkin (48:14.140)
to do such estimations, but I would, you know,
Lex Fridman (48:19.340)
intuitively, I would probably put it lower.
Lex Fridman (48:23.100)
But still, I mean, you know, given.
Lex Fridman (48:24.820)
So we're really lucky here on Earth.
Dmitry Korkin (48:27.980)
I mean.
Lex Fridman (48:28.820)
Or the conditions are really good.
Dmitry Korkin (48:30.500)
It's, you know, I think that there was,
Lex Fridman (48:32.340)
everything was right in a way, right?
Lex Fridman (48:35.460)
So we still, it's not, the conditions were not like ideal
Lex Fridman (48:39.720)
if you try to look at, you know, what was, you know,
Dmitry Korkin (48:44.060)
several billions years ago when the life emerged.
Lex Fridman (48:48.340)
So there is something called the Rare Earth Hypothesis
Dmitry Korkin (48:52.020)
that, you know, in counter to the Drake Equation says
Lex Fridman (48:55.740)
that the, you know, the conditions of Earth,
Dmitry Korkin (49:00.240)
if you actually were to describe Earth,
Lex Fridman (49:03.260)
it's quite a special place.
Lex Fridman (49:05.700)
So special it might be unique in our galaxy
Lex Fridman (49:09.120)
and potentially, you know, close to unique
Dmitry Korkin (49:11.780)
in the entire universe.
Lex Fridman (49:12.860)
Like it's very difficult to reconstruct
Dmitry Korkin (49:14.740)
those same conditions.
Lex Fridman (49:16.380)
And what the Rare Earth Hypothesis argues
Dmitry Korkin (49:19.580)
is all those different conditions are essential for life.
Lex Fridman (49:23.100)
And so that's sort of the counter, you know,
Dmitry Korkin (49:26.180)
like all the things we, you know,
Lex Fridman (49:29.220)
thinking that Earth is pretty average.
Dmitry Korkin (49:31.740)
I mean, I can't really, I'm trying to remember
Lex Fridman (49:34.340)
to go through all of them, but just the fact
Dmitry Korkin (49:36.140)
that it is shielded from a lot of asteroids,
Lex Fridman (49:41.140)
the, obviously the distance to the sun,
Lex Fridman (49:43.820)
but also the fact that it's like a perfect balance
Lex Fridman (49:48.220)
between the amount of water and land
Lex Fridman (49:52.180)
and all those kinds of things.
Lex Fridman (49:53.660)
I don't know, there's a bunch of different factors
Dmitry Korkin (49:55.180)
that I don't remember, there's a long list.
Lex Fridman (49:57.520)
But it's fascinating to think about if in order
Dmitry Korkin (50:01.260)
for something like proteins and then DNA and RNA
Lex Fridman (50:05.020)
to emerge, you need, and basic living organisms,
Dmitry Korkin (50:10.020)
you need to be very close to an Earth like planet,
Lex Fridman (50:14.960)
which will be sad or exciting, I don't know which.
Dmitry Korkin (50:19.740)
If you ask me, I, you know, in a way I put a parallel
Lex Fridman (50:23.220)
between, you know, between our own research.
Lex Fridman (50:28.380)
And I mean, from the intuitive perspective,
Lex Fridman (50:33.820)
you know, you have those two extremes
Lex Fridman (50:36.700)
and the reality is never very rarely falls
Lex Fridman (50:40.820)
into the extremes.
Dmitry Korkin (50:41.900)
It's always the optimus always reached somewhere in between.
Lex Fridman (50:46.500)
So, and that's what I tend to think.
Dmitry Korkin (50:50.060)
I think that, you know, we're probably somewhere in between.
Lex Fridman (50:54.020)
So they were not unique, unique, but again,
Dmitry Korkin (50:58.220)
the chances are, you know, reasonably small.
Lex Fridman (51:01.940)
The problem is we don't know the other extreme
Dmitry Korkin (51:04.180)
is like, I tend to think that we don't actually understand
Lex Fridman (51:08.060)
the basic mechanisms of like what this is all originated
Dmitry Korkin (51:11.900)
from, like, it seems like we think of life
Lex Fridman (51:15.060)
as this distinct thing, maybe intelligence
Dmitry Korkin (51:17.100)
is a distinct thing, maybe the physics that,
Lex Fridman (51:20.380)
from which planets and suns are born is a distinct thing.
Lex Fridman (51:24.380)
But that could be a very, it's like the Stephen Wolfram
Lex Fridman (51:27.140)
thing, it's like the, from simple rules emerges
Dmitry Korkin (51:29.420)
greater and greater complexity.
Lex Fridman (51:31.020)
So, you know, I tend to believe that just life finds a way.
Dmitry Korkin (51:36.100)
Like, we don't know the extreme of how common life is
Lex Fridman (51:39.540)
because it could be life is like everywhere.
Dmitry Korkin (51:44.980)
Like, so everywhere that it's almost like laughable,
Lex Fridman (51:49.420)
like that we're such idiots to think who are you?
Dmitry Korkin (51:52.140)
Like, it's like ridiculous to even like think,
Lex Fridman (51:56.260)
it's like ants thinking that their little colony
Dmitry Korkin (51:59.460)
is the unique thing and everything else doesn't exist.
Lex Fridman (52:03.220)
I mean, it's also very possible that that's the extreme
Lex Fridman (52:07.540)
and we're just not able to maybe comprehend
Lex Fridman (52:09.900)
the nature of that life.
Dmitry Korkin (52:12.860)
Just to stick on alien life for just a brief moment more,
Lex Fridman (52:16.580)
there is some signs of life on Venus in gaseous form.
Dmitry Korkin (52:22.260)
There's hope for life on Mars, probably extinct.
Lex Fridman (52:27.260)
We're not talking about intelligent life.
Dmitry Korkin (52:29.220)
Although that has been in the news recently.
Lex Fridman (52:32.340)
We're talking about basic like, you know, bacteria.
Dmitry Korkin (52:36.300)
Yeah, and then also, I guess, there's a couple moons.
Lex Fridman (52:40.820)
Europe.
Dmitry Korkin (52:41.660)
Yeah, Europa, which is Jupiter's moon.
Lex Fridman (52:45.100)
I think there's another one.
Dmitry Korkin (52:46.580)
Are you, is that exciting or is it terrifying to you
Lex Fridman (52:50.380)
that we might find life?
Lex Fridman (52:52.140)
Do you hope we find life?
Lex Fridman (52:53.580)
I certainly do hope that we find life.
Dmitry Korkin (52:56.020)
I mean, it was very exciting to hear about this news
Lex Fridman (53:05.380)
about the possible life on Venus.
Dmitry Korkin (53:09.260)
It'd be nice to have hard evidence of something with,
Lex Fridman (53:12.540)
which is what the hope is for Mars and Europa.
Lex Fridman (53:17.140)
But do you think those organisms
Lex Fridman (53:18.420)
will be similar biologically
Dmitry Korkin (53:20.780)
or would they even be sort of carbon based
Lex Fridman (53:23.940)
if we do find them?
Dmitry Korkin (53:25.740)
I would say they would be carbon based.
Lex Fridman (53:28.940)
How similar, it's a big question, right?
Lex Fridman (53:31.820)
So it's the moment we discover things outside Earth, right?
Lex Fridman (53:39.540)
Even if it's a tiny little single cell.
Dmitry Korkin (53:43.260)
I mean, there is so much.
Lex Fridman (53:45.380)
Just imagine that, that would be so.
Dmitry Korkin (53:47.660)
I think that that would be another turning point
Lex Fridman (53:50.700)
for the science, you know?
Dmitry Korkin (53:52.540)
Especially if it's different in some very new way.
Lex Fridman (53:56.220)
That's exciting.
Dmitry Korkin (53:57.060)
Because that says, that's a definitive statement,
Lex Fridman (53:59.700)
not a definitive, but a pretty strong statement
Dmitry Korkin (54:01.780)
that life is everywhere in the universe.
Lex Fridman (54:05.420)
To me at least, that's really exciting.
Dmitry Korkin (54:08.940)
You brought up Joshua Lederberg in an offline conversation.
Lex Fridman (54:13.460)
I think I'd love to talk to you about Alpha Fold
Lex Fridman (54:15.780)
and this might be an interesting way
Lex Fridman (54:17.220)
to enter that conversation because,
Lex Fridman (54:19.580)
so he won the 1958 Nobel Prize in Physiology and Medicine
Lex Fridman (54:24.500)
for discovering that bacteria can mate and exchange genes.
Lex Fridman (54:29.020)
But he also did a ton of other stuff,
Lex Fridman (54:32.220)
like we mentioned, helping NASA find life on Mars
Lex Fridman (54:37.740)
and the...
Lex Fridman (54:40.980)
Dendro. Dendro.
Dmitry Korkin (54:42.580)
The chemical expert system.
Lex Fridman (54:45.260)
Expert systems, remember those?
Lex Fridman (54:46.860)
What do you find interesting about this guy
Lex Fridman (54:51.380)
and his ideas about artificial intelligence in general?
Lex Fridman (54:54.980)
So I have a kind of personal story to share.
Lex Fridman (55:00.180)
So I started my PhD in Canada back in 2000.
Lex Fridman (55:05.140)
And so essentially my PhD was,
Lex Fridman (55:07.740)
so we were developing sort of a new language
Dmitry Korkin (55:10.100)
for symbolic machine learning.
Lex Fridman (55:12.540)
So it's different from the feature based machine learning.
Lex Fridman (55:15.100)
And one of the sort of cleanest applications
Lex Fridman (55:19.820)
of this approach, of this formalism
Dmitry Korkin (55:23.980)
was to cheminformatics and computer aided drug design.
Lex Fridman (55:28.820)
So essentially we were, as a part of my research,
Dmitry Korkin (55:33.820)
I developed a system that essentially looked
Lex Fridman (55:37.380)
at chemical compounds of say the same therapeutic category,
Lex Fridman (55:42.380)
you know, male hormones, right?
Lex Fridman (55:45.540)
And try to figure out the structural fragments
Dmitry Korkin (55:51.740)
that are the structural building blocks
Lex Fridman (55:54.420)
that are important that define this class
Dmitry Korkin (55:58.140)
versus structural building blocks
Lex Fridman (55:59.780)
that are there just because, you know,
Dmitry Korkin (56:02.700)
to complete the structure.
Lex Fridman (56:04.260)
But they are not essentially the ones
Dmitry Korkin (56:06.060)
that make up the chemical, the key chemical properties
Lex Fridman (56:10.020)
of this therapeutic category.
Dmitry Korkin (56:12.780)
And, you know, for me, it was something new.
Lex Fridman (56:16.900)
I was trained as an applied mathematicians, you know,
Dmitry Korkin (56:20.580)
as with some machine learning background,
Lex Fridman (56:22.980)
but, you know, computer aided drug design
Dmitry Korkin (56:25.060)
was a completely new territory.
Lex Fridman (56:27.660)
So because of that, I often find myself
Dmitry Korkin (56:31.500)
asking lots of questions on one of these
Lex Fridman (56:34.340)
sort of central forums.
Dmitry Korkin (56:36.940)
Back then, there were no Facebooks or stuff like that.
Lex Fridman (56:40.420)
There was a forum, you know, it's a forum.
Dmitry Korkin (56:43.620)
It's essentially, it's like a bulletin board.
Lex Fridman (56:45.780)
Yeah.
Dmitry Korkin (56:46.620)
On the internet.
Lex Fridman (56:47.460)
Yeah, so you essentially, you have a bunch of people
Lex Fridman (56:50.300)
and you post a question and you get, you know,
Lex Fridman (56:52.900)
an answer from, you know, different people.
Lex Fridman (56:55.300)
And back then, just like one of the most popular forums
Lex Fridman (56:59.300)
was CCL, I think Computational Chemistry Library,
Dmitry Korkin (57:04.300)
not library, but something like that,
Lex Fridman (57:07.100)
but CCL, that was the forum.
Lex Fridman (57:09.820)
And there, I, you know, I...
Lex Fridman (57:12.780)
Asked a lot of dumb questions.
Dmitry Korkin (57:14.060)
Yes, I asked questions.
Lex Fridman (57:15.500)
Also shared some, you know, some information
Dmitry Korkin (57:19.340)
about how formal it is and how we do
Lex Fridman (57:21.460)
and whether whatever we do makes sense.
Lex Fridman (57:25.100)
And so, you know, and I remember that one of these posts,
Lex Fridman (57:29.180)
I mean, I still remember, you know,
Dmitry Korkin (57:31.420)
I would call it desperately looking
Lex Fridman (57:35.340)
for a chemist advice, something like that, right?
Lex Fridman (57:40.740)
And so I post my question, I explained, you know,
Lex Fridman (57:43.980)
how formalism is, what it does
Lex Fridman (57:49.220)
and what kind of applications I'm planning to do.
Lex Fridman (57:53.180)
And, you know, and it was, you know,
Dmitry Korkin (57:55.020)
in the middle of the night and I went back to bed.
Lex Fridman (57:59.660)
And next morning, have a phone call from my advisor
Dmitry Korkin (58:04.780)
who also looked at this forum.
Lex Fridman (58:06.900)
It's like, you won't believe who replied to you.
Lex Fridman (58:11.020)
And it's like, who?
Lex Fridman (58:13.900)
And he said, well, you know, there is a message
Dmitry Korkin (58:16.300)
to you from Joshua Lederberg.
Lex Fridman (58:19.140)
And my reaction was like, who is Joshua Lederberg?
Dmitry Korkin (58:22.660)
Your advisor hung up. So, and essentially, you know,
Lex Fridman (58:29.660)
Joshua wrote me that we had conceptually similar ideas
Dmitry Korkin (58:34.060)
in the dendrial project.
Lex Fridman (58:36.660)
You may wanna look it up.
Lex Fridman (58:39.300)
And we should also, sorry, and it's a side comment,
Lex Fridman (58:42.620)
say that even though he won the Nobel Prize
Dmitry Korkin (58:45.940)
at a really young age, in 58, but so he was,
Lex Fridman (58:49.820)
I think he was what, 33.
Dmitry Korkin (58:52.860)
It's just crazy.
Lex Fridman (58:53.980)
So anyway, so that's, so hence in the 90s,
Dmitry Korkin (58:57.660)
responding to young whippersnappers on the CCL forum.
Lex Fridman (59:02.100)
Okay.
Lex Fridman (59:02.940)
And so back then he was already very senior.
Lex Fridman (59:05.820)
I mean, he unfortunately passed away back in 2008,
Dmitry Korkin (59:09.580)
but, you know, back in 2001, he was, I mean,
Lex Fridman (59:12.580)
he was a professor emeritus at Rockefeller University.
Dmitry Korkin (59:15.980)
And, you know, that was actually, believe it or not,
Lex Fridman (59:18.460)
one of the reasons I decided to join, you know,
Dmitry Korkin (59:25.460)
as a postdoc, the group of Andre Salle,
Lex Fridman (59:28.140)
who was at Rockefeller University,
Dmitry Korkin (59:30.820)
with the hope that, you know, that I could actually,
Lex Fridman (59:33.460)
you know, have a chance to meet Joshua in person.
Lex Fridman (59:38.060)
And I met him very briefly, right?
Lex Fridman (59:42.140)
Just because he was walking, you know,
Dmitry Korkin (59:45.380)
there's a little bridge that connects the,
Lex Fridman (59:47.860)
sort of the research campus with the,
Dmitry Korkin (59:51.940)
with the sort of skyscraper that Rockefeller owns,
Lex Fridman (59:55.500)
the where, you know, postdocs and faculty
Lex Fridman (59:58.780)
and graduate students live.
🔗 相关节目