Bioelectricity, Morphogenesis, and Two-Headed Worms | Michael Levin

0:00There's a really deep and interesting

0:01thing about plenarian, but let's talk

0:02about head and tail decisions. In most

0:04textbooks, what you'll see is they show

0:05the worm being cut into thirds. So

0:07there's like two cuts and then they draw

0:09this like gradient and they say, "Well,

0:11this is how you know where the head and

0:12tail is." That's all well and good, but

0:14what they're neglecting to show you is

0:16the simpler but much more challenging

0:18version of one cut. Because if you make

0:21one cut in the middle, the cells on the

0:23left side of the cut are going to make a

0:24tail. Cells on the right side of that

0:26cut are going to make a head. Their

0:27positional information is exactly the

0:29same. they were neighbors until you came

0:30with your scalpel and cut them apart.

0:31And just purely logically, you can't

0:33tell locally whether you should be a

0:35head or a tail. You have to communicate

0:37with the rest of the tissue. I wanted to

0:38have one giant eyeball. I wanted the

0:40whole thing to be one giant eyeball. So,

0:42what I would do is I would go in and

0:43inject like an eight cell embryo. I

0:44would inject every single cell like,

0:46"Okay, this thing's going to be a giant

0:47eye." No, you can get multiple eyes, but

0:49you will never get a bigger eye. There's

0:51something that we haven't cracked yet

0:52where they know what the size is. The

0:54first two-headed worms were seen around

0:561903, made by a completely different

0:58process. Nobody had thought to recut

1:00them for over a hundred years. Why?

1:04Because everyone thought it was obvious

1:05what would happen. This week, Michael

1:07and Shingu spoke with Dr. Michael Levan

1:09about how bioelectric signals guide

1:11development, regeneration, and

1:13collective cellular decision-making.

1:15Levven explains how voltage patterns

1:17help tissues determine body structure,

1:19why cells communicate like networks, and

1:21how manipulating bioelectricity can

1:23induce regeneration, alter anatomy, and

1:25even create two-headed plenarian worms.

1:28We also discuss morphagenesis, memory,

1:30and living systems, and the possibility

1:31that biology operates through high level

1:33information processing far beyond

1:35genetics alone.

1:37>> Cool.

1:38>> Please welcome Mike Leaven.

1:40>> Start was actually really early. I was

1:42interested uh from an extremely early

1:43age in uh engineering in biology. I

1:47spent a lot of time looking at bugs and

1:49insects, you know, when I was a a kid

1:51and um thinking about the difference

1:53between the things I was building uh

1:54with electronics and later computers and

1:57the living things that put themselves

1:58together and have preferences about what

2:00happens and all of that. And so for the

2:01longest time uh I was kind of interested

2:04in this big question of how minds exist

2:06in the physical universe, how they

2:08scale, how they change, what you know

2:10what happens and all of that. And um

2:12what I was always looking for is what I

2:14now call cognitive glue, which is the

2:17policies and mechanisms that take parts

2:19and come together to be a whole that

2:21knows things and and has goals and

2:23preferences that the individual parts

2:24don't have. And I've always wondered

2:26what what that was. And uh we we we kind

2:29of knew what it was in neuroscience,

2:31right? So it's the it's the

2:33electrophysiology of the of the neural

2:34networks and so on that that uh are are

2:37critical for for those kinds of things.

2:38But we didn't really know what it was in

2:41in development. And in 1986,

2:44I was at the Vancouver World's Fair with

2:46my dad and we went into a used bookstore

2:48as we did all the time and we found uh

2:51Robert Becker's The Body Electric. It's

2:53a book and it's it's a book that you

2:54know kind of well well known um in the

2:56field. Now the most interesting thing to

2:58me in that book was a bibliography going

3:02back decades showing that people were

3:04thinking about this already and doing

3:06experiments and there's a whole there's

3:07a whole history of developmental bio

3:09electricity and it occurred to me that

3:10this was this was kind of a perfect uh

3:14you know a perfect merger of of the

3:16things I was interested in because not

3:17only does it answer an evolutionary

3:19question which is where did the brain

3:21learn its amazing tricks right so so

3:23clearly it was much older than that and

3:24and and was here long before neurons and

3:26and brains appeared. But also uh this it

3:30it enabled us to have a kind of uh entry

3:33point to understand how do morphagenetic

3:36systems make decisions. In other words,

3:38beyond just the chemistry of the

3:40mechanisms, the question that we always

3:41ask when we look at embryos like how

3:42does it know? How does it know how many

3:44fingers? How does it know? So we had a

3:45standard story that that basically all

3:48of the parts work together according to

3:50the laws of chemistry. Everything is a

3:52dumb machine. Nothing actually knows

3:53anything. and the resulting outcome is a

3:56sort of emergent consequence of of what

3:58happens. But uh I and many others were

4:00very suspicious of that story and it

4:02seemed to me that um bioelect

4:03electricity by taking uh inspiration

4:06from what we learn in neuroscience about

4:08top- down causation and multiscale kinds

4:10of decision-m and so on. It seemed like

4:12that was a perfect way to address it.

4:13And so that was that that's that's

4:15pretty much when when my plan really

4:17crystallized around what I wanted to do.

4:19But but the interest goes back much

4:21earlier than that.

4:22>> So this was pre-undergraduate studies.

4:23Did you already go into your

4:24undergraduate studies thinking that you

4:26wanted to in high school that you wanted

4:28to combine electronics that you were

4:30playing around with with uh biological

4:32questions or even more so um uh

4:36questions of consciousness potentially?

4:38>> Yeah, even even long before that as a as

4:40a teenager I I was doing behavioral

4:42experiments with you know worms and and

4:44and tadpoles and things like that. I'm

4:46trying to understand um how they see

4:48their world and things like that. And

4:50this is this is this that that became my

4:52dream is to uh sort of advance the

4:54science in that direction. I didn't

4:56really think it was realistic or

4:57possible, but I figured I would do as

4:59much as I could and you know and see and

5:00see how far I could I could push it. And

5:02what was then the first experiment that

5:04you designed and did yourself that was

5:05in this direction?

5:08>> The well as a as as an undergrad I I I

5:10didn't have access to any uh you know

5:13serious equipment and things like that.

5:15But um what we did what I did do was to

5:18look at the effects of exogenous fields.

5:20So in that in that case electromagnetic

5:21fields on sea urchin development. So so

5:23there are two limitations to that

5:25method. The first limitation is that

5:26this is exogenous stimulation. So

5:28whatever you find uh you don't know that

5:31this has anything to do with how the

5:32system normally controls itself. What

5:34you found is perhaps a cool engineering

5:36trick and and maybe you've maybe you've

5:38learned something no doubt but it wasn't

5:39the same as asking um how does the

5:41system control its own parts? How do

5:43parts come together? How is information

5:44integrated in the in the system? How are

5:46decisions made? So external stimulation

5:49is not sufficient for that. Also

5:51external stimulation is is a blunt tool.

5:53In other words, for for sure you can do

5:55some amazing things and people have, you

5:56know, the the the kind of my heroes in

5:59this field have done incredible work on

6:00on showing all kinds of phenotypes that

6:02made it clear that something important

6:04was going on. But but the limitation of

6:06it is that after when you've applied a

6:08field, you don't know exactly what

6:09you've done mechanistically. In other

6:11words, biological tissue, the impedance

6:13is very complex. And so where do the

6:14currents go? What happens to the cells?

6:17You know, how are they affected and all

6:19of that? It's very hard especially with

6:20magnetic fields because that goes every

6:22you know the ELFs go everywhere and so

6:24you can't you have no real spatial

6:26control over it. Um yeah, I wanted to uh

6:29this was part of my strategy afterwards

6:31uh as a uh as a postto develop the first

6:35molecular tools to really do to to

6:37really address endogenous native

6:40biological signaling, no external

6:42stimulation, no magnets, no no magnetic

6:44component just to understand um how the

6:46system controls itself. And um I suppose

6:49it was quite obvious then from the

6:50beginning that you would focus on gap

6:52junctions and membrane potentials or ion

6:54channels. So how to manipulate them or

6:57how did you create your toolbox?

6:59>> Yeah. Well, it wasn't obvious at all.

7:00These were very hard choices because

7:02because until now um or until then most

7:04people were studying uh electric fields

7:07and some people were studying ion

7:08fluxes. So um Lionel Jaffy and Rich

7:12Nutelli uh designed this thing called

7:13the the vibrating probe. Yeah. At

7:15Purdue. Yeah. They designed this thing

7:16called the vibrating probe which allowed

7:18you to uh characterize ion fluxes and

7:21and those were very important but I

7:23really thought that um for a number of

7:25reasons I thought that what we should be

7:26looking at is resting potential voltage

7:28gradients which is which is quite

7:29different right from from either of

7:31those and so so I focused on two things

7:33I focused on ways to characterize the

7:36voltage potential non-invasively until

7:38then you could do it with electrodes so

7:39standard electrophysiology you poke the

7:41cell with little glass needles but the

7:42problem is it it's invasive you only get

7:44one cell at a time if the cell moves or

7:46divides uh you know you can forget it

7:48and you can't get the whole the whole

7:49embryo right so so what I wanted to do

7:52was to look at uh voltage sensitive dyes

7:54fluorescent dies that allow you to

7:56non-invasively soak the whole the whole

7:57thing and get um images and videos of of

8:00what's going on so that's the that's the

8:01side of reading the information and then

8:03in terms of writing the information I

8:05wanted to manipulate um the way that the

8:08cells endogenously create and change

8:11that potential which meant ion channels

8:13and so ion channels ion pumps and So as

8:15a as a posttock I began to assemble this

8:18this toolkit and what I would do is uh I

8:21would I would um contact various

8:23neuroscientists in the field and I would

8:24ask them for plasmids for their ion

8:26channels. So so at first at first uh I I

8:28told people why I wanted it. Uh in other

8:30words I said I was going to try to

8:31misexpress these in other places in the

8:33embryo to have uh control over the

8:36resting potential and thus study by

8:37electricity. I stopped I stopped doing

8:39that because my boss Mark Mola my postto

8:42mentor came in one day and he was

8:43laughing and he had a letter in his hand

8:44and he said he said uh one of the people

8:46that I had emailed wrote him a letter

8:48and warned him that one of his postocs

8:50was basically dangerously insane and

8:52that and that and this was such a crazy

8:54like this was such a crazy idea. I only

8:56say that because nowadays it seems

8:58pretty obvious, but at the time this was

8:59considered completely nuts. Like to to

9:01to look at the the kind of uh

9:03neuroscience components outside of the

9:06nervous system seem seemed crazy to most

9:08people. So he was warning my boss to be

9:10careful that I was

9:10>> and this was not inest. He actually

9:12>> No, no, he was 100% serious. He was he

9:14was warning. I mean I get it as a PI

9:16now. I app I would appreciate the

9:17warning pace. It could be a waste of

9:20time or Yeah.

9:21>> Or worse or worse. Sometimes people have

9:22[laughter] nervous breakdowns, you know.

9:24So yeah, he was giving him a heads up

9:26and so and so I, you know, I stopped

9:28telling people why I want these things

9:29and I would just, you know, I would just

9:30ask for them and and and mostly

9:31everybody was nice and sent them out.

9:33And so that was and that was the

9:34toolkit. So the idea was to learn to

9:36manipulate the native interface, the way

9:38that the cells control each other, the

9:39way that the cells are hacking their own

9:42neighbors and the way that the system

9:43was collecting information. That that is

9:45that is where I wanted to uh pipe into

9:47that system. And then in practice once

9:49you received these dyes you would

9:50immerse the entire um embryo in it and

9:53then it would find find then the

9:55locations where it would accumulate and

9:57you would read it out through a

9:59microscope or what would be the

10:01experiment design.

10:02>> So there are two right so so the dyes is

10:04the um is the reading component that's

10:06reading the bioelectric code and

10:07actually the first the first person

10:09actually to do that was my collaborator

10:10Ken Robinson who was also a a great um

10:13you know a real hero of mine um in the

10:15field. He did tons of work in the 70s

10:16and 80s and 90s on this stuff. And um

10:19what we were working together on

10:21bioelect electricity of left right

10:22asymmetry in the chick the chick embryo

10:24at the time. And so he and his postoc

10:27Thorleaf Thorland um used the dyes to to

10:29to basically see the first asymmetric

10:32voltage gradient in the chicken um at

10:35the exact stage I told them it would be

10:36at because of my my predictions about

10:38left right asymmetry and how it would

10:40work and uh and and then then we used it

10:43on frog embryos and everything else. And

10:44so yeah, it's microscopy. So it's uh

10:46it's epifllororesence microscopy.

10:48Nowadays the tools are much fancier.

10:50There are better dyes of fret dyes and

10:52and flim and and um you know uh uh

10:55conffocal microscopy and and super

10:57resolution all that stuff. But at the

10:58time it was just it was just epif

10:59fluoresence. So that's so that's the

11:01reading part. The writing part is so

11:03what when you have a plasmid that

11:05encodes an ion channel uh the sequence

11:08what we would do is um in vitro

11:10transanscribe it. So you have your RNA

11:12and then you inject that RNA. where you

11:13micro inject it with a little pulled

11:15glass needle. You micro inject it into

11:16specific um precursor cells of the frog

11:18embryo

11:19>> and then um so that's that's one

11:21technique and then um

11:22>> you they would they would express and

11:24you put in a little tracer so you can

11:25see which cells exactly got it. That's

11:27that's one way to do it. You can do gain

11:29of function by putting in active

11:30channels. You can do loss of function by

11:32putting in dominant negative um channels

11:34that will inhibit you know native native

11:36channels. You can put in completely

11:38heterologous channels. That's the beauty

11:39of it. The cells are reading voltage.

11:41They don't care how it got there. So we

11:42used channels from yeast you know pumps

11:44from yeast and channels from other

11:46species all in you know all kinds of

11:48heterologous systems and then of course

11:50there are drugs right so so so the

11:52pharmarmacology of ion channel openers

11:54blockers all that stuff

11:56>> what led you to formulate your

11:57hypothesis about the left right

11:59asymmetry formation

12:01>> so as a as a grad student uh with cliff

12:03tabin I studied well we we character we

12:06we we discovered and characterized the

12:08first uh set of asymmetrically expressed

12:10genes so these were the first genes that

12:12are expressed on the left or the right

12:13side and they actually um there's a

12:15whole cascade where they turn each other

12:16on and off and then downstream they

12:19allow the different organs to know which

12:21side is left and which side is right and

12:22as a physicist you know you probably

12:24know this is a fascinating problem

12:25because at the macroscopic scale the

12:27universe doesn't distinguish left from

12:28right

12:29>> kyality like right fascinating right and

12:31so the question is how do embryos

12:33reliably know which side is is left and

12:35which side is right you can't get it

12:36from genetics because genes don't tell

12:38you which side is left and right right

12:40you need to you need to orient to to the

12:41outside world. And so [snorts]

12:43>> you need to introduce a symmetry

12:45breaking what is doing that in biology.

12:48>> Exactly. You need you need symmetry

12:49braking and then you need amplification

12:51because the symmetry breaking will tell

12:53each cell that way is left.

12:55>> But that doesn't tell you but I'm on the

12:57right side of the of a midline. Right?

12:59So it has to be again it's this I I was

13:00really interested in this because not

13:01only is it a cool physics problem um

13:04it's also uh uh again this like change

13:07of scale because it's one thing to know

13:08which direction is left and right but

13:10it's something else to know where am I

13:11relative to the rest of the of the

13:12embryo and that's actually um extremely

13:15important because all of morphagenesis

13:17is this kind of collective intelligence

13:19that integrates relative to a a a large

13:22scale target that individual cells don't

13:23know but the collective does

13:25>> and so and so I was very uh very

13:27interested so we so We found we found

13:29these these set of genes but then of

13:31course whichever earliest gene you have

13:34that is on one side or the other you

13:36have to ask well how did that one get

13:38expressed on so what's upstream of that

13:40and so that's what I that's what I

13:42decided to tackle as a postto and I

13:44started with a with a with a simple but

13:46but kind of interesting experiment what

13:48I did was I took the early frog embryo

13:50and I cut it in half and I cultured the

13:53two halves separately and we had by then

13:56because of my earlier work we had

13:57markers that were left-sided markers or

13:59right-sided markers. So, you could tell

14:00what does the tissue think it is? Does

14:02it think it's on the left and right? And

14:04what we found out is that if you

14:05actually segregate the two halves, they

14:08get confused.

14:09>> In other words, 50% of the time each

14:11each one would would do left side or

14:14right side. Actually, there were two

14:15interesting things. One is that

14:17>> the the tissue itself would get

14:18confused, but the cells would never get

14:20confused. In other words, you never saw

14:21speckles. You never saw, well, I think

14:23I'm left and the next cell overthinks

14:25I'm right. The whole tissue would be in

14:26agreement. The with and this came up

14:28again and again in my work later on but

14:30the tissue was in agreement it just

14:31wasn't cor you know half the time it

14:33wasn't correct

14:34>> the other half as the reference point so

14:36to speak that you took away

14:38>> yeah yeah so so what that told us is

14:40that you can't which is which actually

14:42contradicts a mainstream model in the

14:44field which is that individual cells can

14:46figure this out because of the kirality

14:48of psyia and things like that actually

14:50think it's got nothing to do with that

14:51um it means that indiv it means that uh

14:55small bits of tissue can't um uniquely

14:58determined left and right. They need to

14:59coordinate, right? You need to you need

15:00to have a have a communication between

15:02across actually across considerable

15:03distances, millimeters in that case.

15:06>> [snorts]

15:06>> And so and so I started to think okay

15:08how how would this be and and then and

15:10then uh that's when that's when the

15:12whole cap junction thing came up because

15:14I started to think that again borrowing

15:16some some ideas from neuroscience and uh

15:19just drawing a very simple you know my

15:22first my first model was was it was it

15:23was a very simple electric circuit where

15:25I thought okay if you have a battery at

15:27the vententral end of the embryo and the

15:29rest of the gap junctions kind of go

15:31into this so so so the embryo at that

15:33point was a flat sorry like a like a

15:34frisbee like a flat And so but there was

15:37a primitive streak like the special

15:38region down at the at the bottom and

15:40then and then all this stuff around if

15:42the if what was happening the battery

15:44was driving uh either current or or

15:46electropharesing something else around

15:48that's why you needed the whole thing

15:49because if you interrupt it anywhere

15:51along and I in fact I showed that if you

15:52make single cuts anywhere along the edge

15:54you would you would confuse it. So it

15:56had to communicate you know in in that

15:58way and so that was the first that was

16:00the first model of of a battery um in

16:02the frog it's a ventral side and the

16:03check it's the primitive streak

16:05>> and then a a gap junction path through

16:07which something is going under

16:08electropheretic force.

16:09>> So bioelect electricity basically um

16:12revealed itself as the answer to a

16:15question that you already had which is

16:16the left right asymmetry. Yeah, I I mean

16:19you know on on the one hand you might

16:21think uh this is tremendously lucky you

16:23know this this idea I mean bio

16:25electricity was something I was

16:26interested in for a long time my PhD

16:29project was not unusual or weird you

16:32know I mean it was it was unique in in

16:34many ways but back in those days I

16:35didn't talk about any of these ideas you

16:37know I was sort of taking the time to

16:38just learn the mainstream thinking that

16:40I need to you know learn how how to do

16:42things at the state-of-the-art level I

16:44could maybe break them later if I stayed

16:46in the field long enough but But but you

16:48know I was just doing like good you know

16:49standard molecular biology kinds of

16:51things. Um and uh it's kind of it was

16:55you know kind of tremendous of good

16:56fortune that that particular project

16:57lent itself immediately to applications

16:59in bio electricity. It didn't have to be

17:00that way. Uh but of course but but the

17:02flip side of that is that you know many

17:04people were studying um early

17:06development. Most people didn't look for

17:08bioelectric kinds of things because they

17:09weren't interested. Right. So it's you

17:11know I think I think it's partially my

17:13emphasis and partially luck.

17:14>> You said it led to applications. What

17:17were these applications that the first

17:19experiment led to?

17:22>> Well, uh the you know the fir the first

17:24experiment basically allowed us to

17:26design uh the tools that we then used

17:28again and again. So

17:29>> the dies and the yeah the dyes the

17:32strategies of uh putting in um different

17:35ion channels and pumps the strategies of

17:37changing the gap junctional paths

17:38because with molecular biology you could

17:40make gap junctions that were either

17:41opened or closed or had different

17:43permeabilities.

17:44uh to use the drugs use we I developed

17:47drug screens something called an inverse

17:49drug screen to identify in new contexts

17:52which um channels and pumps might be

17:54important which was very again very

17:56important because natively if you want

17:57the the mainstream molecular biology

17:59cell biology community to to pay

18:01attention they need to know where does

18:03the signal come from what are the genes

18:04that are important for it and I don't

18:06think the genes drive this process at

18:07all but but they do provide the hardware

18:09so you need to know what genes are

18:10important and then of course on the

18:12other end having Um, and that's

18:14something else that I that I developed

18:15is ways to chase it down. So that having

18:17said, okay, here is this uh voltage

18:19pattern that's important for this and

18:21that. How does it control cell behavior?

18:23So what are the what are the cell

18:24behaviors that are changed? What are the

18:25gene expressions that are changed? Like

18:27what else happens, right? So so being

18:29able to track that. So that was that was

18:30kind of the idea. I thought that if the

18:32only way that any of this would be

18:33successful is to merge it with the uh uh

18:37explosion of of of work in cell and

18:39molecular developmental biology. And

18:41that meant two things. That meant first

18:42of all addressing questions that they

18:44were interested in. So not just not just

18:47you know obscure um even if fascinating

18:49phenomena that they could sort of

18:51marginalize but actually like if you

18:52want to know why the gene Sonic hedgehog

18:54is on this side and not on that side the

18:57molecular explanation is not sufficient.

18:58You actually have to go up upstream of

19:00that and understand the biohysics that

19:02allows the whole thing to um to orient

19:04you know to in in space.

19:06>> And and the second thing is to tie it to

19:08the pathways that we're interested in.

19:10So, so do all of the characterization.

19:11What genes are downstream? What what

19:13genes are upstream? You know, what's the

19:14what's the how what is the interplay

19:16between the physics and the and the

19:17genetics.

19:18>> Was this controversial at all when you

19:20first uh suggested that there's

19:22something upstream of uh gene control?

19:25>> Oh, absolutely. Uh this was this was

19:27this was incredibly incredibly

19:28controversial. Um it's still it's still

19:30in in many corners. Um it is very

19:33controversial. Uh, amazingly enough, I

19:35mean, it's it's, you know, it's kind of

19:37remarkable. Uh, a few months ago, um, we

19:40published we published a paper in trends

19:42in genetics where we said things that

19:45were basically you couldn't say those

19:47things uh, even even recently, even a

19:48few years ago. I mean, I I'm amazed that

19:50we could say them now. But, um, but

19:52yeah, this was this was considered very

19:54weird.

19:54>> I would feel bad if I didn't get out of

19:55this interview with with an

19:57understanding of this. So, you have the

19:58you have the plarium worm.

20:00>> Okay. uh you cut off the you cut off the

20:02tail or the head and then you

20:04>> or you cut it into thirds.

20:05>> Yeah, you cut it into thirds. You cut

20:06the middle part or whatever and then you

20:10uh how do you actually adjust the

20:12potentials on the end? Is it just

20:14soaking it in the drug or like what's

20:16the experimental design that leads to

20:18two heads for example?

20:19>> Right. So there's a couple ways to do

20:21it. So pleneria for a number of reasons

20:24are they they they a lot of techniques

20:26that work for example in our um frog

20:28model don't work in pleneria for for

20:30technical reasons their cells are tiny

20:32you can't inject them you can't there is

20:34no embryogenesis really available so you

20:36can't inject blasts like there's no

20:38transgenics it's a it's a it's a it's a

20:40very difficult system to work with so

20:42what you have left uh in pleneria are

20:44drugs

20:45>> and um the earliest uh way I I generated

20:49the two heads was the following

20:51There's a there's a really deep and

20:53interesting thing about about pleneria.

20:54So, so in most textbooks um well there's

20:58many amazing things about plenaryia but

20:59but but let's let's talk about head and

21:01tail decisions. In most textbooks, what

21:02you'll see is they show the worm being

21:04cut into thirds. Okay. So so there's

21:06like two cuts and then they draw this

21:08like gradient and they say well this is

21:10how you know where the head and tail is.

21:11It's because there's a there's a

21:13gradient and it's different. Okay.

21:14>> Like a voltage gradient.

21:15>> No, no. They're they're talking about a

21:16diffusion gradient of of you know W

21:18protein or something like that. Right?

21:21>> That's all well and good, but what

21:24they're neglecting to show you is the

21:26simpler but much more challenging

21:28version of one cut. Because if you make

21:30one cut in the middle, just cut them

21:31down the middle. The cells on the cells

21:35on the on the left side of the cut are

21:37going to make a tail. The cells on the

21:38right side of that cut are going to make

21:39a head. Their positional information is

21:41exactly the same. They were neighbors

21:43until you came with your scalpel and cut

21:44them apart. What gradient at that point,

21:46right? That that doesn't that doesn't do

21:47the trick. So that means that you can't

21:49just purely logically you can't tell

21:51locally whether you should be a head or

21:53a tail. You have to because because that

21:55information will be the same much like

21:56we talked about the left right

21:57asymmetry. You have to communicate with

21:59the rest of the tissue. So you have to

22:01say which you know is there already a

22:03head here? Am I on the other end? Where

22:04am I located right? And so I started to

22:06so I started to think and so we're

22:07talking this was like you know 2003 or

22:09so. I was thinking uh how is that

22:12coordinated? Well, my favorite thing,

22:14the gap junction. And so I said, "Okay,

22:15what if we just block gap junctions and

22:17so so pharmacologically you can you can

22:20block gap junctions." And that and the

22:21the the first experiment was was was

22:23just to block the gap junction. And sure

22:24enough, and this is this is the work of

22:26Taisakui, my one of my first postocs.

22:29Um, at the time uh yeah, that that's

22:32that's that's how we we got the the

22:33two-headed the two-headed worms. And

22:36something else something else just to

22:37just to point out there uh

22:41you know having worked with them for a

22:43few years 20078 um we recut them and

22:47that's it just you take the two-headed

22:49worms and you just cut them into pieces

22:50in plain water. No more no more drug no

22:52more no more.

22:53>> The first two-headed worms were seen

22:55around 1903 made by a completely

22:57different process. Nobody had thought to

22:59recut them for over a hundred years.

23:02Why? because everyone thought it was

23:04obvious what would happen. Their

23:06genetics are unchanged if you just get

23:07rid of that that ectopic secondary head.

23:09Of course, they're going to regenerate

23:10normally. Like, why would you even do

23:12this? So, this goes back to our to our

23:13earlier discussion of like thinking

23:15about these things in a different way

23:17makes you do different experiments,

23:18right? And so, we said, well, yeah, the

23:20genetics are the same, but but I didn't

23:22really believe that the genetics were,

23:23you know, were were storing all this

23:25stuff anyway. And so, I said, why don't

23:26we recut them and see what happens? And

23:27it turns out that two-headed worms, if

23:29you recut them, they make more

23:30two-headed worms. And if and and also

23:32when they reproduce naturally, which is

23:34to to tear themselves in half and

23:35regenerate, they also make two-headed

23:36worms. So that goes back like that's the

23:38one example I have where it actually

23:39does go crossgenerational. You know,

23:41that's that's you know, we know we know

23:43it does that.

23:43>> So So that's the first way we did it.

23:45More recently, we were able to do it

23:47with um specific voltage changes using

23:49ionophores. So these [snorts] are these

23:51are compounds that uh basically allow

23:55specific ions, protons or or potassium

23:57or whatever uh to cross in and out of

23:59cells. So you can, you know, you can

24:00make changes that Okay. So, so in

24:01plenary it's pharmacology basically.

24:03>> So the idea is you uh block the gap

24:07junctions. It doesn't know whether it's

24:09supposed to form a head or a tail. And I

24:10guess that how does it know in the end

24:12whether to make two tails or two heads.

24:14>> Uh the the standard so there's so

24:16there's two two things that are going on

24:18here. First thing is that uh by default

24:21if you're if you're a if you're a region

24:24and you are broadcasting hey uh is there

24:27a head here and nobody is the head then

24:30then then I'm going to make somebody has

24:31to make a head. So so you know so you do

24:33that that's that's the first thing. So

24:34in the in the absence of a head the

24:36default is make a head.

24:38>> Um that's the first thing. The second

24:39the second thing is that there's a

24:40particular biological pattern and it's

24:42it's not local but it's a large scale

24:43pattern that says how many heads are you

24:46supposed to have and what you can

24:47actually see and the coolest part to me

24:49the coolest part is that you can take a

24:50perfectly standard worm one head one

24:53tail

24:53>> and you can use the the ion force to

24:56change that biological pattern and you

24:58can confirm that on the on the voltage

24:59die that it now says two heads but that

25:01worm doesn't have two heads. So that's

25:03the most interesting part is that this

25:05is not a pattern of what the tissue is

25:08now. It's not a pattern that reflects

25:09the current anatomy.

25:11>> It's a um it's a counterfactual memory

25:15of what I will do if I get injured at a

25:17certain time. Right? It's the to what I

25:20think it is is and I think this is quite

25:22significant. It's a primitive form of

25:23the kind of mental time travel that

25:25we're able to do as as cognitive beings.

25:27So if you're an extremely primitive

25:29system, all you know is what's happening

25:31right now and you can react to that. But

25:33if you are are more um more

25:35sophisticated cognitive system, you can

25:37have thoughts about things that happened

25:38before. So memories, you can have

25:40anticipations about what will happen

25:41later. It's information that is not

25:44true. You can think about things that

25:45are not true right now. It's extremely

25:46powerful and they call it mental time

25:48trial. So I think this is a very simple

25:50version of that because that that worm

25:52if you look at it anatomically one head.

25:55If you look at molecular markers,

25:56perfectly normal, anterior markers in

25:58the head, posterior and the tail, no you

26:00wouldn't know. You wouldn't know

26:01anything was wrong until you looked at

26:03voltage. And it's kind of like um the

26:05neural decoding that neuroscientists try

26:06to do. You look at the voltage, you go,

26:07"Wait a minute. This thing this thing

26:08has a false memory. It thinks that it

26:10thinks that it should have two heads."

26:11But you won't know that it's a late

26:12memory until you injure it. As soon as

26:14you cut it, now you get two heads.

26:15>> Mhm.

26:15>> And so

26:16>> that's really cool.

26:17>> Yeah. I thought I Yeah, I think I think

26:18that's I think that's pretty neat.

26:20>> And can you impose that instruction

26:22after you've cut like when you do the

26:24thirds and you can just take the middle

26:26part and then encode that and it will

26:28still absolutely. So you can either do

26:30that before or

26:31>> Yeah. I mean there's a time limit

26:32because because if you wait too long, if

26:34you waitress

26:3712 hours, it's already decided what it's

26:39going to be and then it's too late. Um

26:40but but to address your your other point

26:42about the heads or tails, we eventually

26:44learned by by 2011 we had learned how to

26:47go to the tail fate. And so then we

26:49could make we could make no head worms

26:51or or in fact we could fix two-headed

26:53worms and send them back to being

26:54one-headed by changing the pattern back.

26:56>> But can you make eight heads? Can you

26:59take like sides of the worm and keep

27:01like growing?

27:02>> You you you can. There's a there's a tra

27:04we we did a we have a I think it's a

27:06developmental biology cover for Junji

27:08Morakuma's paper that has a four-headed

27:10worm that's the shape of a cross. It has

27:12[laughter]

27:13like that. Yeah, you can.

27:16>> So, okay. So, I have the two-headed

27:17worm. I cut it in half and it grows two

27:19heads again. How does the cut in the

27:21middle know to make a head? So, so what

27:25happens here and we're soon going to

27:27reach the limit of what we know because

27:28there's a lot that we don't know. What

27:30happens is that the electric the

27:32electric pattern in the tissue rescales.

27:34So, one of the most interesting things

27:36about this is that when you have a

27:37particular pattern that is worm sized,

27:40it's sized as the whole worm. You cut

27:41the thing into pieces. Each piece then

27:44re the pattern rescales in each piece.

27:46It's like it's like cutting a magnet

27:47basically, but it's but it's much more

27:48complex because it's not just two poles.

27:50It's not just head and tail. For

27:51example, we showed that you can make

27:52these heads these pleneria develop the

27:54heads of other species. So, so right.

27:57So, so it's not just that it's not just

27:58a dipole, but it's a similar thing in

28:01that you can cut it, but the pattern

28:03rescales and within within hours you'll

28:06see um you'll see that now you whatever

28:08pattern you had, you now have you now

28:09have here.

28:10>> So, the pattern that was in the head.

28:11So, I cut this thing in half. There are

28:13two heads. The pattern in the head ends

28:15up in the cut. Like

28:17>> it's not a it's not the pattern in the

28:19head. It's a it's it's a it's a pattern

28:22that's that's

28:23>> like the relative the relative change

28:25ends up it used to be from head to head

28:27then it's now it's cuts it rescales cuts

28:29ahead. That's uh that's crazy.

28:31>> Yeah. Yeah. Super super interesting. And

28:33we and we've been trying to we've been

28:34trying to computationally model that. Um

28:36we we have models that that do that. So

28:39you can make models of electric circuits

28:40that work that way. Uh right? Uh because

28:42because that's what's happening here.

28:44It's the pattern is it the whole thing

28:46is like RAM basically right? It's a it's

28:48a self sort of maintaining. Once you put

28:51in the pattern, it holds. As long as

28:52there's energy, the thing holds and and

28:54it and it it maintains and it rescales

28:57and it has some properties. We we have

28:58computational models that work like

28:59that. Um but but testing them

29:02specifically in the in the worm is hard

29:04because of the limitations of the

29:05plenarian model and and those are

29:07actually it's really interesting why

29:08they're they're limited like that.

29:10>> If you wanted to make a lot more heads,

29:12sorry, I'm stuck on that.

29:14>> How many heads would [laughter] you

29:14like?

29:16Do do you reach a saturation point at

29:18some point where one um head says, "Oh,

29:21there's a head already next to us, so

29:23we're not going to turn it into

29:24>> You do." And you do, and it's very

29:26interesting. The size control is very

29:28interesting here. Um, we can make

29:30ectopic eyes. We've never succeeded in

29:33making an eye that's bigger than a

29:34normal eye. [clears throat]

29:35>> They know exactly what size they should

29:37be. We have not cracked the size

29:39control. If you I tried at at first I

29:41tried I wanted to have one giant

29:43eyeball. I wanted the whole thing to be

29:44one giant eyeball. And and and if you do

29:47that, and this is the fun.

29:50>> Yeah. Uh if you [laughter] if you So

29:52what I would do is I would I would go in

29:53and and inject like an eight cell

29:55embryo. I would inject every single cell

29:56like, "Okay, this thing's going to be a

29:58giant eye." No, you can get multiple

29:59eyes, but you will never get a bigger

30:01eye. The eye there's something that we

30:03haven't cracked yet. That's that where

30:04they know what the size is. And it's

30:06very important too because um in

30:09pleneria for example this this again

30:10shows you the difference between the

30:12molecular approaches and the bioelectric

30:14approaches there are um there are uh you

30:18know standard ways of manipulating the

30:20wind pathway where you get ectopic heads

30:21in pleneria when you do that the head is

30:24not the right size compared to the rest

30:26of the body. Eventually they might

30:27rescale but but it's not the right size

30:29because you have to specify the size

30:31yourself and the identity and the wind

30:33protein or the wind signaling apparently

30:35only specifies the identity doesn't you

30:37you didn't control for the size. The

30:38biomectrics is not like that. When you

30:40make a

30:41>> when you make a second head via via

30:43bielectric control everything is

30:44perfectly matched. The size you know

30:47everything is perfect because because

30:48you're at a higher level right because

30:50you're not down here where you have to

30:51micromanage these other things. You're

30:52you're you're up at a higher level of

30:53control. So, so the size control is

30:56really interesting. Heads absolutely

30:58will um sort of demarcate themselves and

31:01there's a minimal size that and there's

31:03something else here which is the

31:04standard model in the field says that

31:07heads suppress heads. So that once you

31:09do have a head it sends out some sort of

31:11a suppressive signal that then

31:13suppresses heads somewhere else. But if

31:15you cut the head like this, right, if

31:17you cut it, you know, this way, this

31:19side of the head will will regenerate.

31:21Well, it'll make two heads if you keep

31:23recutting. You have to It's like the

31:24high school experiment. You get a Y. You

31:26have to keep recutting. You have to

31:27prevent them from joining. But let's say

31:29you cut and you get rid of one side.

31:31This side will just regenerate the other

31:32side. There's no suppression there. It

31:34immediately regenerates right next to

31:36that head. There's more, you know, head

31:38tissue. So, it's much more sophisticated

31:40than that.

31:40>> So, I remember originally uh watching

31:43like talks by you maybe in like 2017 or

31:46even earlier. The like famous experiment

31:49is like the plener cutting up the

31:50pleneria, right? That's the like kind of

31:52like the aha moments that's easy to

31:54communicate. What confused me

31:55immediately was like, okay, you have

31:57this mass of cells, whether it's a worm

31:58or anything else,

32:00>> how can you change the voltages in the

32:02cells in a specific region or like

32:04basically if you're soaking the worm in

32:06a drug? I mean, how do you control where

32:08the voltages are changing? And the

32:10second question is

32:11>> if you're sensing this with voltage

32:13sensitive dyes, how do you deal with all

32:15the artifacts from the dyes or like

32:17bleaching or these kinds of things? Like

32:18to what degree

32:20>> to what like how lossy is the whole

32:22process of actually trying to make these

32:23changes cuz it seems like a very

32:25complicated system and readout.

32:27>> Yeah. Yeah. Great questions. Okay, let's

32:30let's uh let's talk about the dyes

32:31first. Uh yeah, there there are about a

32:33million controls that you need to do. So

32:35uh for for sure for sure there's

32:37bleaching. uh for sure there are issues

32:39with um uh die availability so

32:41differential die penetration in

32:43different regions. So the way you handle

32:44that is to use raometric dies or now

32:46flim fluoresence lifetime imaging which

32:49are basically uh ratios away some of

32:52those uh some of those issues. There are

32:54still plenty the the whole the whole

32:56technology of reading voltage

32:59non-invasively especially in deep

33:00tissues where the light path isn't you

33:03know that's that's there's a lot to to

33:05be done in the future you know um

33:07there's there's a lot of room for

33:08development but but the current to even

33:10then the tools were good enough and the

33:12current tools are good enough to make a

33:13lot of progress but there's plenty of

33:14room for development out there and there

33:16are lots of controls that that you need

33:18to do for sure with the uh with the with

33:21the biomectric changes. So there's a

33:23couple of things

33:24>> I guess wait on the dyes just if we take

33:26like a toy example or whatever if you

33:28you have the worm you cut it in half and

33:30you want to know so you you soak it in

33:33the dyes and they just diffuse into the

33:35worm and then you have how do you get

33:37the ratio? It's like two different dyes

33:38and you have a

33:39>> you can do it there's a couple of ways

33:40you can do it. Uh and um and by the way

33:43now we also have genetically encoded

33:44voltage reporters. So there are proteins

33:47that that report voltage that you that

33:49you can do with transgenics you know.

33:51Um,

33:51>> so it's like from the embryo level, you

33:53you just put it in and it grows with

33:55those reports

33:56>> or you can knock it in later, you know,

33:57you can transfect or something like

33:58that, right? So, so you have a there's a

34:00wide range of tools now that that you

34:02have. And so, yeah, um, you can collect

34:04you can collect at at two different

34:06wavelengths or you can have two dieseS.

34:08One of my favorite uh methods that we

34:10used to use has two dieseS uh, and and

34:13one die is sensitive to depolarization.

34:15The other die either is nonresponsive or

34:17goes the other way. And so then you can

34:19computation

34:19>> they diffuse together or they're even on

34:21the same molecule.

34:22>> Correct. They can they can be or they

34:23diffuse together. But again oftentimes

34:25they will some of these reagents have a

34:29uh a wavelength at which they are

34:30voltage sensitive and another wavelength

34:32at which they're not voltage sensitive.

34:33And so then you can you can ratio away

34:35things like penetration artifacts.

34:38>> And did you then develop these dyes

34:40yourself or took whatever the um

34:43neuroscientists were already working on?

34:44>> Yeah, we don't develop the dies

34:45ourselves. We're not we're not chemists.

34:47Uh uh we we work with people like Evan

34:49Miller who who um makes makes you an

34:51amazing chemist makes these dies uh or

34:53or the the genetic inc genetically

34:56encoded reporters. Um

34:59>> the thing is uh neuroscience sometimes

35:02sometimes they make dies but the thing

35:04with neuroscience is that they're mostly

35:06interested in very rapid spiking. So

35:09what they want are dyes that have very

35:11rapid onoff kinetics. We want the exact

35:13opposite. We want something that is

35:15going to catch all the slow the slow

35:17stuff because the things that we're

35:18interested in change at at the scale of

35:21minutes hours maybe right not

35:23milliseconds. So a lot of the dies that

35:25we're interested in are not in fact the

35:27dies that most people are using because

35:29they're trying to catch the kilohz kind

35:30of spike right yeah and is the

35:33resolution purely determined by the

35:34microscope resolution or by

35:36>> it is and in fact and in fact now we

35:38have we're looking at the nuclear

35:40envelope potential. So the nuclear

35:41envelope also has a voltage gradient.

35:43We're going to have a cool paper on that

35:44soon. Oh wow. Yeah. Yeah, we didn't

35:46discover that that it's been known for

35:48for a long time through

35:49electrophysiological evidence, but yeah,

35:51the nucleus has a voltage. So yeah, the

35:54resolution is just your microscopy. Um I

35:56will however say that the vast majority

35:58of what we do with this is not at high

36:01magnification.

36:02>> Voltage the the as as now we know which

36:04we didn't know at the beginning. Uh the

36:06bi-electric code is not a cell level

36:09code. So, so yes, cells read their own

36:12voltage, but but the real um magic of

36:14all of this is in large scale patterns.

36:17So, most of the time our problem are is

36:19the opposite. We're not trying to zoom

36:20in and although that's cool, too. You

36:23can zoom in like within a single cell,

36:24you can see different um domains of

36:26voltage in the same cell. Like there are

36:28lots of like cells don't have one

36:29voltage. There's lots of little It's

36:30like a soccer ball of different, you

36:31know, like different um

36:33>> uh I don't know if it's rafts or or what

36:35it is, but um

36:36>> but but for us, the biggest thing is to

36:38pull back and see the large scale

36:40because if you're dealing with a frog

36:41leg or a whole plinarian or or an, you

36:43know, an eye or a frog embryo or

36:45something like that, you're not trying

36:46to zoom in. You're trying to pull back

36:47and say what is the what is a large

36:49scale pattern here. Yeah.

36:50>> Is this already visible on the single

36:51cell level? Like when you have the

36:53single cell stage of an embryo?

36:56>> Absolutely.

36:56>> And I guess how

36:57>> is that just a dipole? Do you just have

36:59plus minus on the single cell level or

37:01what?

37:01>> Yeah, the single. So, so the single cell

37:03egg, there's there's sort of two things

37:05going on. There's inside versus outside,

37:07which is where how you measure the

37:08voltage, right? So, so all around Yeah.

37:10in internal versus external path.

37:12>> But again, even even in at the earliest

37:15stages, it isn't a single number. So, if

37:17you really zoom in on the on the um

37:19membrane, you see there are domains of

37:21different voltages all around.

37:23>> And is that random or what is the

37:25program the pattern?

37:26>> It's a good it's a good question. It's a

37:28good question. We we don't know yet

37:30whether there's some sort of

37:32combinatorial code at the level of a

37:34single cell on the on the membrane. So

37:36we don't know that hasn't been

37:37functionally tested because mainly

37:39because we don't know how to address

37:40individual domains, right? It would be

37:41nice to make different patterns. We

37:42don't know how to do that yet.

37:44>> You know, so can I say it's not random?

37:46No. Do I believe it's random? Not for a

37:48moment. I I don't think any of this

37:49stuff is random. I think I think these

37:51are very powerful um computational

37:53elements and I would be ult just you

37:57know incredibly shocked if evolution

37:58didn't notice that and and wasn't

38:00optimizing all this stuff for various

38:02purposes. So but but I can't prove that

38:04it's not written

38:04>> but it needs to be passed down in some

38:06way right so there it needs to be some

38:08sort of memory on

38:10>> some level that is able to retain the

38:13information and then uh

38:14>> so in the case of okay so so we don't

38:17yet have evidence that bioelect

38:18electrical changes propagate through

38:20sexual reproduction I'm not saying they

38:22don't I can imagine how it might work

38:23but we don't have evidence for that

38:24mostly just because we don't have a good

38:26model system with a short enough life

38:28cycle where we can do this um you know

38:30frogs have a two-year reproductive cycle

38:31it's a it's a pain. So we just we just

38:33haven't done it. But um but the one case

38:36where it does propagate is in pleneria.

38:39So in pleneria which replicate by

38:43fision and regeneration. In that case it

38:45it it does it does the changes do

38:47propagate. So I I don't you know I can't

38:49tell you that biomectric changes that we

38:50make in a tadpole will then affect the

38:52oasite and then affect the next

38:53generation but it's not impossible. We

38:55just don't have evidence for it. And I

38:57guess like when you're actually making

38:59the changes the second half of the so

39:01you have the the dyes and then how do

39:03you actually affect the changes how do

39:05you deal with the fact you know

39:06diffusion or whatever like how do you

39:07actually make changes on a local level

39:10or is it something that's happening for

39:11the whole organism just kind of randomly

39:13and you pick out the ones where the

39:14change is what you're looking for.

39:15>> No no it's yeah it's definitely not

39:17that. Um, okay. So, there are two I'll

39:20tell there are two basic methods and and

39:22I'll I'll tell you one one example of

39:23each one one kind of example story. So,

39:27one thing you can do especially in the

39:28frog embryo which is partly why I

39:30started there is super convenient is

39:32that you can you can microject RNAs

39:36encoding specific channels and pumps and

39:38things like that into precursor cells

39:40that then give rise to various

39:41structures. So if I want so so imagine

39:44so you're looking at the early frog

39:45embryo and you know that in the face the

39:49location of the eyes are determined by a

39:51particular voltage pattern. There's a

39:53particular spot that says this is where

39:54the eye is going to be. So to to be able

39:56to test that the functional role of that

39:58what you want to do is introduce that

39:59pattern somewhere else. And so let's say

40:02uh let's say you want to uh put it on

40:06the on the gut of the embryo let's say.

40:08So what you would do is you would start

40:09with an early embryo. So let's say 16

40:11cells or something like that and you

40:12take your RNA and you micro inject it

40:14into cells using a fate map that you

40:16that is a known published fate map. You

40:18inject it into cells that are going to

40:19give rise to the region that that you

40:21want. And so then you introduce a little

40:23and of course they divide. And so their

40:25offspring all have this channel. And so

40:26you make a little a little patch, right?

40:28And to the extent that you've targeted

40:30what you want to target that's that's

40:31that you know and then and then of

40:32course you check up the voltage to make

40:34sure that that's that that's what you

40:35did. So, so in that case,

40:37>> so from that perspective, the pattern

40:38would be mainly it's like a spot

40:40basically.

40:40>> Well, it can have different so this I'll

40:42get into this momentarily. It gets much

40:44more complex but but the minimal like

40:46the simple basic version is a spot of

40:49voltage of a particular size. That's the

40:50simplest pattern. The pattern is

40:52actually much more complex because the

40:55tissue that the material that you're

40:57dealing with is an excitable medium. So

40:59think of like like almost like a like a

41:01cellular automaton, you know, or or

41:03something like that where you start

41:05>> the neighbors are responding.

41:06>> Yeah. Yeah. And and and and the gap

41:08junction so the voltage channels can be

41:10the the the ion channels can be voltage

41:12sensitive. The gap junctions can be

41:14voltage sensitive. So you have feedback

41:16loops, both positive and negative

41:17feedback loops. And as soon as you've

41:18introduced an inhomogeneity in that

41:21excitable medium, then the thing takes

41:24over and and you start to see, you know,

41:26you start to see all kinds of patterns.

41:27One of the important things that we've

41:29done is to create a uh a simulator, a bi

41:33electricity simulator that shows you

41:34because it's not obvious at all. You

41:36can't at least I I can't do it in my

41:37head to say this is this is what you

41:39know what's going to happen. So so the

41:41patterns start off simple but they get

41:43they get very complex very quickly and

41:44you need the the simulator to uh to to

41:47guide it. So, so the first the first

41:49kind of way you can do it is you can

41:51target whether it's early micro

41:52injection, whether it's later

41:54iontopharesis or some kind of

41:56electroparation. I mean, you can kind of

41:58target things right to where you want

42:00them. That's one. There's actually a

42:01much more interesting I think and and

42:03much deeper way to do this, which is to

42:04take advantage of the uh

42:07self-organization of the system. So, so

42:10I'll give you an example. Uh we and this

42:13is this is the work of VIP pie in my

42:16group. Uh

42:18in uh 2017 basically from 2017 to about

42:22now what we were working on with him is

42:24um birth defects. So specifically birth

42:26defects of the brain face and gut in the

42:29tadpole and one of the things that

42:31happens when you either expose them to

42:32teratogens or uh they have a mutation of

42:36an important gene like notch. They have

42:38they have these terrible defects. And we

42:40looked and we noticed that one of the

42:42things that was happening is the

42:43prepattern that the electrical

42:45prepattern that tells the brain and the

42:46rest of these organs what shape to be

42:48the pre- pattern was all screwed up. And

42:50we said okay could we rescue all of

42:52these things by uh putting the pattern

42:54back to normal despite the fact that it

42:57it had seen these seratogens. It had the

42:58mutation like we ultimately we showed

43:00that you can even um repair after even

43:03though they have the notch mutation you

43:04can still get normal embryoenesis. It's

43:06kind of amazing. Um, so, so how do you

43:08do that? Well, the problem is the

43:09pattern is quite complex. You don't want

43:11to go in and have to tattoo all the

43:12cells to make, you know.

43:13>> Yeah, because how large is this? This

43:15must be rather large embryo if you can

43:18micro.

43:19>> Yeah, the embryo I mean the embryo

43:20itself is maybe a millimeter wide,

43:22something like that. Um, but the cells

43:24are quite small. There are a lot of

43:26cells and there's no hope of going in

43:29and really like, you know, putting in a

43:30cell autonomous pattern, you know, cell

43:32cell by cell. What we wanted to do was

43:34to uh take advantage of the native

43:37dynamics of the of the voltage pattern.

43:39And so we used the simulator. And so so

43:40Alexis Pyac, our collaborator, uh

43:43developed this early version of this

43:45amazing simulator. And um what we what

43:49we did was we built a what she she built

43:53a computational model of what was going

43:54on in these cells and the different

43:56channels, how they were reacting to each

43:57other and the voltage and how they how

43:59the system evolves as a function of

44:00time, right? [snorts] And then we simply

44:03uh ask the model to go backwards and say

44:06okay but if we wanted the pattern to be

44:08this what would we need to do and the

44:10and the and the the the system was able

44:12to identify one particular channel

44:14called HCN2. HCN2 has a really

44:17interesting property. Um

44:19>> in cells that are a little that are a

44:21little bit polarized it it um

44:26cells that are that are deolarized it

44:28leaves them alone. But cells that are a

44:30little bit polarized that it polarizes

44:32them further. So what it in effect is is

44:34a is a contrast filter, right? It it

44:36exacerbates differences. If you're down

44:38here, you stay down there. If you're up

44:39here, you go to here, right?

44:40>> It's polarized relative to what? The

44:42extra extracellular matrix or like

44:43polarized

44:44>> inside and out. So so voltage. So so you

44:46know the way you the way neuroscientists

44:48measure VM. So from from inside relative

44:50like across the memory

44:51>> just whatever there's like a fluid Yeah.

44:53>> around the cells and whatever the

44:54potential is.

44:55>> Exactly. So, so this this HCN2 uh turned

44:59out we predicted the HCN2 uh would be

45:02like a like a sharpen filter on

45:03Photoshop. It would basically be it

45:05would sharpen the gradients and that's

45:07that was very important because what we

45:09saw in the embryo is all the gradients

45:10would get fuzzy. The brain was all

45:12misshaped because the the the the normal

45:14distinction between here's where the

45:15neural tube begins and ends and that

45:16kind of stuff was becoming degraded.

45:18That information was becoming degraded

45:20and fuzzy and we wanted to sort of

45:21strengthen that. So knowing that we

45:25didn't have to localize it. We could

45:27just soak the whole embryo like applying

45:30the same filter to to to everything. Uh

45:33I was I was shocked. It it worked. Um uh

45:37it it worked amazingly well. We were

45:39able to repair uh brains after all kinds

45:42of really nasty, you know, kind of

45:44insults. Uh normal brain structure,

45:46normal brain gene expression, normal

45:48IQs. they got their learning to nor to

45:50to standard levels after very severe um

45:52you know kinds of defects. So that's an

45:55example of leveraging an understanding

45:59of the native uh uh trajectory of the

46:01system where you don't try to

46:02micromanage every piece but you know

46:04that you you can you can compute that

46:05that if I give this kind of stimulus and

46:07that kind of stimulus then the system

46:09itself will enter specific states right

46:11that that you can count on. And the most

46:13important part of all of this I think is

46:16that bioelect electricity offers

46:18something very unique that uh is really

46:20hard to do with with for example

46:22chemical signals. Um, and and I'll give

46:24you an example of of what I mean when

46:25we're talking right now. I don't need to

46:28worry about the synaptic proteins in

46:29your brain. Um, right, you're going to

46:31handle all of that. We have a we have a

46:33very high level interface and and all of

46:35the downstream stuff is handled on my

46:37end and on and on your end. It's why

46:38humans were able to train dogs and

46:40horses for thousands of years knowing

46:41zero neuroscience. like they had no idea

46:43what was between their ears and it

46:44doesn't matter because because the

46:46amazing thing about the our body

46:47architectures is that they uh transduce

46:50these very um sort of high level

46:52abstract things like think about um when

46:55you wake up in the morning and you have

46:57social goals, research goals, financial

46:59goals, whatever. In order for you to get

47:00up and do those things, ions have to

47:02cross muscle membranes like these

47:03incredibly abstract highle things have

47:05to make the chem have to change the

47:06chemistry. So and and and and because

47:09bio electricity that that that trick was

47:12developed long before there's nerve and

47:14muscle, bi electricity is is all about

47:15that. It meant that when we say to the

47:18cells build an eye, we don't have to say

47:21how do you build an eye? I I we don't

47:23have any idea how to build an eye. We

47:24you know there's tens of thousands of

47:25genes that that would have to be

47:26changed. There are stem cells have to go

47:28here and there and the the retina is

47:29very complex. We don't have to know any

47:31of that. What we found was a set of um

47:34highle cues just like we communicate

47:36here. We are communicating to the

47:38cellular collective and we talk about

47:40very abstract highle things. Eyes.

47:42Individual cells don't know what an eye

47:44is but the but the group does

47:45apparently. And so so bio electricity

47:47just like in neuroscience bio

47:48electricity allows us to communicate

47:50with the system at a high level. This is

47:52very different from the standard

47:53molecular biology approach where you try

47:55to uh you try to micromanage the the the

47:58you know bottom up. You try to

47:59micromanage the heart.

47:59>> Yeah. It sounds like you're just

48:00initiating a sequence that's already

48:02known somehow.

48:03>> We're initiating the start of a

48:05conversation. Um, it really I really

48:07think that the right uh the right

48:08formalism for all of this is is

48:10communication and and and and it's a

48:12it's a conversation. We make we make the

48:14first uh we we put out the first

48:16stimulus a prompt knowing that the

48:18system is going to do certain things.

48:19Maybe at some point we have to course

48:21correct and say something else. Maybe we

48:22don't. In the case of the frog leg

48:24regeneration,

48:25>> we made a stimulus in the first 24

48:27hours. They gave us a year and a half of

48:28leg growth. We never had to touch it

48:30again. We never touched it. So in some

48:32cases you say one thing and the system

48:33goes and that's the the characteristic

48:36of competent systems is that you you

48:38don't have to micromanage them. You give

48:39them their marching orders. If you're

48:40convincing they will go off. There's

48:42autonomy in the material, right? You're

48:44dealing you're dealing with an aential

48:46material here. This is not even active

48:47matter or computational matter. This is

48:49agential. These are agential materials.

48:51And in the case when you were um

48:53stimulating tissue to grow eyes, would

48:56they always fully form a fully formed

48:59eye in in the sense that there would be

49:01a retina? Was it functional all all the

49:04time?

49:04>> Good good question. Uh not all the time

49:06and there are two reasons for that. Uh

49:08so so so yeah, you know, definitely not

49:10100% um success rate. There are two

49:12reasons for that. One one is a boring

49:15technical reason which is just that it's

49:17there's there are technical challenges

49:19in in getting the voltage pattern the

49:20way we want and we've been trying to

49:22develop since then optogenetic um

49:24strategies so you can lay down light

49:25light masks on on a tissue and like all

49:28of that is still is still difficult like

49:30it's not it's not easy but there's a

49:32there's a deeper and more interesting

49:34thing going on here. One time we we had

49:37these we had these embryos that and that

49:40encoded a uh they were transgenic and

49:43they had a gene that was driving ricks

49:45one which was it's an early eye field

49:47specification gene and it was driving a

49:50fluorescent protein. So you could see in

49:52the embryo you could see by fluoresence

49:53ahead of time which cells were going to

49:55become eyes. Okay. So non-invasively. So

49:58that's cool. We got I forget who who we

49:59got maybe Mike Zuber or somebody. I

50:01forget who we got the the the embryos

50:03from. Anyway, so so we do our biomectric

50:06experiment and I look at the embryos and

50:08I see this thing. This thing has like

50:10seven ectopic eye spots all over all

50:12over because I I sort of poked it in

50:14different locations. I'm like, "Oh, this

50:15is amazing. This thing have seven eyes,

50:16right?" Like incredible.

50:17>> And so you do that actually by hand or

50:19do you use

50:20>> No, by hand. You do it by hand. Yeah.

50:21You sit there. There's a micro injector

50:22with a little glass needle and if you

50:24break the needle, you got to go through

50:25like recalibrate the whole thing. And

50:27then you sit there and you just like

50:28poke the poke the early embryo, right?

50:30And I always, you know, I always wanted

50:31to make an embryo with tons of eyes. And

50:32so, so [laughter] I thought I thought

50:34this was this would be fantastic. I'm

50:35just gonna I'm just going to inject like

50:36lots of lots of places. So I look at the

50:38embryo on Monday and I see seven seven

50:41like ectopic Ricks one spots. I'm like,

50:43"This is amazing. This thing's going to

50:44be great." I come back the next day,

50:46there's only four. I said, "Well, that's

50:47weird." I come back the next day and

50:49you're lucky if there's one. And what's

50:50happening at that later as we started

50:53digging into it and later this is what

50:54we found out. There's a there's a debate

50:57going on in the tissue because you

51:00introduce some cells that not only this

51:02is this is I missed one one important

51:04part of this which is that these eyes

51:06that we make when you when you take a

51:07cross-section so let's say let's say I

51:09inject a potassium channel and I also I

51:12mix in some betalactoidase RNA so that

51:14you can track later you can just stain

51:16them and see which cells it just turns

51:17blue in the cells that you injected so

51:19you can see which ex which cells exactly

51:21did I inject when you when you section

51:23these eyes only a small percentage of

51:25them are injected which means that the

51:28cells that we injected actually then

51:30went out and recruited other cells that

51:32we never touched to be part of this eye.

51:33It's lots of collective intelligences do

51:35that. Ants and termites do that too. If

51:37a couple of ants find something too

51:38large, they will sort of recruit their

51:40nestmates, right? So the the the

51:41material scales to the problem, right?

51:43Again, amazing. We didn't have to teach

51:44it to do that. It it already does that,

51:45right? So, so okay. So, so we knew that

51:48that when you change the voltage and

51:49this is this is the part of um uh that's

51:52critical for biio medicine is because

51:53you want to be really convincing when

51:55you have this conversation with the

51:56cells. You want to make sure that they

51:57have bought into the new homeostatic set

51:59point that you've given them. They're

52:01not going to fight you on it the way

52:02that now currently happens with drugs. I

52:03mean all cells resist and whatever you

52:05want you want them to buy you won't buy

52:06in from the system.

52:07>> So we say okay when we do this

52:08successfully these cells uh recruit

52:10their recruit their neighbors. But what

52:13ends up happening is a battle because so

52:15I'm over here a cell with an aberant

52:17with a with an aberant voltage and I say

52:18or a group of cells and I say to my

52:20neighbors hey you guys should be an I

52:22with me you should have this voltage

52:23like me and the way that actually

52:25happens is through the gap junctions and

52:26things like that right so saying you

52:27should be an I but there's an ancient

52:29cancer suppression mechanism which

52:31basically says if your neighbor has like

52:33a weird voltage you should normalize

52:35them out and and and that happens also

52:37through the gap junction so the

52:38surrounding cells like no you should be

52:40skin like me or gut or whatever and

52:42they're like, "Nope, you shouldn't be an

52:43eye like me." And there's a there's a

52:44battle that goes on of the of the

52:46voltage and probably other things. And

52:48eventually the system settles down into

52:50one fate or the other. So So what

52:53happens with a lot of these eye spots is

52:54they just don't make it because the

52:55neighbors wipe them out. Now they don't

52:57kill the cells are still there, but the

52:59but the but the the decision of what are

53:01we going to actually be? uh yeah that

53:04decision uh you know we we are still we

53:07are still at the early stages of knowing

53:10what it is that makes specific patterns

53:12compelling to these cells. You know

53:13sometimes we can do it uh and other

53:15times and other times it doesn't work

53:17and we're still working that out.

53:18>> And earlier on you were mentioning that

53:20you had like a tool like a um sharpness

53:22filter in this situation. Do you think

53:25that higher contrast gets more likely to

53:29be rejected or um like can you make it

53:31more probable by having gentler changes?

53:35>> That's a that's a great hypothesis. Uh

53:37we are characterizing all of that. We're

53:39trying to you know we're trying to

53:40figure it out. Um [gasps]

53:42part of it is the part of I mean one

53:44thing we noticed is that gradients are

53:48what the cells interpret. So it's not

53:50the absolute voltage that matters.

53:52Originally, we had a very simple

53:53hypothesis that maybe maybe it's maybe

53:55absolute voltages are just a code for

53:57different body organs or something like

53:58that. It's not um they're not watching

54:00uh absolute voltages. They're watching

54:02deltas differences. And so that means

54:04that all the interesting things happen

54:06at borders. So when the one one voltage

54:08domain is next to another one and the

54:10shape of them and the and the gradients.

54:11So yeah, it's probably something like

54:14that. It's probably some kind of tuning

54:15of like exactly how

54:17>> you have a more adabatic uh change where

54:20it's more likely to to follow the lead

54:23or something like that.

54:23>> Entirely possible. Entirely possible. Um

54:26there are also a lot of uh symmetry um

54:29there breaks and self- amplifications so

54:31that even if you start off with a nice

54:33gentle slope they will crank it up. You

54:35know something will go up and down.

54:36There are a lot of like bifurcation

54:38points and things like that. So in this

54:40example of eyes, you basically made the

54:42observation that certain cells would

54:44become eyes. But um where's that

54:47information stored for the whole program

54:51that you start the conversation with?

54:54>> Um there are two there are two uh

54:57answers to that question on two at two

54:59different levels. F first of all uh what

55:02we are studying is what I think we're

55:04studying is the bielectric code which

55:07specifically is the question of how

55:11surrounding tissue interprets a

55:13particular voltage pattern. It's a very

55:14interesting kind of computation where

55:16the information is simultaneously

55:18rewiring the computer, right? Because

55:19it's the same tissue that has to process

55:21that information, but by processing that

55:23information, uh it changes your your

55:26com, you know, what you do next, right?

55:27So, it's a self sort of selfmodifying

55:29computer. Um from from that from that

55:32perspective we we need to understand and

55:34it's a kind of a combined um bioysics

55:39you know cell biology question of how do

55:42different kinds of cells interpret

55:43specific domain specific biological

55:45patterns for example how it's going to

55:47work across species is it the same

55:48pattern that's an eye of a frog versus

55:50an eye of a fly versus and and things

55:52like that. Um the second question or the

55:55second uh way to think about it and I

55:57don't know if you want to get into this

55:59but but the bigger question of where do

56:01pattern where do these patterns come

56:03from in the first place that's a that's

56:05a very deep um question and uh we could

56:08talk about it if you want but um it's uh

56:11I've been I've been this is last year 25

56:13was the first year that I've been

56:14talking about that that question because

56:16I think it's finally um become

56:18experimentally actionable. up until then

56:20it was, you know, kind of a

56:21philosophical question. Uh, but but I

56:24think now we can we can get at it and

56:25and I I have a very weird theory about

56:27that that is pretty incompatible with

56:29the way that um I think most most

56:30biologists think about things. But um

56:32yeah, I don't know. I don't know. As a

56:33physicist, I don't know what you'd think

56:34about it. So, uh so obviously the this

56:38begs the question like, okay, can we do

56:39this in humans and things like that? But

56:41if you were going to do another more

56:43like a curiosity hero experiment with a

56:46different animal or a different system,

56:48do you have like a wishlist system kind

56:50of like Pleneria where something strange

56:52can happen potentially that you just

56:53like haven't you're you haven't made it

56:55there because you're busy with other

56:56things like uh

56:57>> what a great question. Uh okay. Uh we

57:00are we're working in mice now um with

57:02more there's a company that spun out of

57:04our lab called Morphaceuticals and we've

57:06been interested to uh do the limb

57:08regeneration thing in mammals. Um, so,

57:11so we've got we've got some experiments

57:12in mice going. Uh, there are, you know,

57:15there's a few things. Uh, plants, uh,

57:17I'm interested in plants. We haven't

57:18done any plant work yet. We will, um,

57:20plants have a have, one thing that

57:23plants can teach us, for example, is

57:24that their target morphology is done at

57:27a different scale than ours. So, for

57:28example,

57:30you know, normal animals, they have a

57:33not not all, but most normal animals

57:34have a have a large scale target

57:36morphology that's the same for each

57:37species. If you look at a tree, the

57:40individual pattern of the branches and

57:42and so on is not the same, but in but

57:45but but at a smaller scale it is. So the

57:47leaves, the seeds, the the

57:48cross-sections, all that, right? So

57:49they're holding their target morphology

57:51at a slightly different level. I think

57:52that's that's incredibly interesting.

57:54And plants in general have many

57:55interesting things. Um so that's so

57:57that's one thing. Uh there are some, you

58:00know, for example, uh in terms of the

58:02aging thing because we have a lot of

58:03aging work in our lab. So things like

58:05naked mole rats, you know, what what

58:06does the biomectrics of naked mole rats

58:08look like, you know, uh other other

58:10species that that either live a long

58:12time or really regenerate. We've done

58:14some work in axelottle. It's not nearly

58:15enough. There needs to be way more way

58:17more than that. Yeah, there there are

58:19lots of cool um animals out there.

58:21>> What's the main difficulty in mice?

58:22Because they don't naturally regenerate.

58:24So for plenarian it's kind of easier to

58:27understand that since they already

58:29regenerate you can maybe program a

58:31second head but then for mice it's not

58:34typically that you cut off a paw and

58:36they would regrow it. What what's

58:38different there? What do you have to

58:40modify to get something out of it?

58:43>> I I I have a I have a different

58:45perspective on this than than I think

58:48standard in the in the field. I don't

58:50think that mammals are less plastic. I

58:52don't think that's what it is. I think I

58:54think it's a I think it's an engineering

58:56problem. I mean, think about think about

58:58our um mamalian ancestors. So, you've

59:01got this little thing. It's, you know,

59:02kind of mouselike running around the

59:03forest. Some somebody bites its leg off.

59:06>> You don't have time to try to

59:09regenerate. You're going to bleed out.

59:10You're going to get infected. You're

59:11going to try to crawl its load bearing.

59:13So, you're going to try to put weight on

59:14it and rub it into the forest floor. Uh

59:16the best thing you can do at that point

59:17is scar and and have inflammation and

59:20hope that you live. That's not the case

59:22for axelottals or or plenario, right?

59:25Most good regenerators are aquatic. Um

59:28it's not that mammals can't do it. Look

59:29at deer. So deer every year they

59:31regenerate um

59:33>> they they grow a centimeter and a half

59:35per day of new bone. A centimeter and a

59:38half of new bone per day. Bone vascule

59:40intervation, right?

59:42>> So it's not that they can't do it. And

59:43uh there's something else that's

59:45interesting. I don't know if if you guys

59:46have seen our work on anthrobots, but um

59:50when you isolate cells, adult cells from

59:53an elderly patient and it's impossible

59:55for them to make a human, they make

59:57something else and they make anthrobots

59:59that have 9,000 differentially expressed

1:00:01genes, half the genome, and they, you

1:00:03know, they do cool things like healing

1:00:04neural wounds and all this stuff. the

1:00:06plasticity is there, but I think there

1:00:09are um constraints that have been uh and

1:00:12um uh uh trade-offs that have been

1:00:14evolutionarily molded. And so one of the

1:00:17strategies that that we're doing, for

1:00:18example, we've developed this thing

1:00:19called a biodome, it's a wearable

1:00:21bioreactor.

1:00:22>> So when you when you have an amputation

1:00:24as a mouse

1:00:25>> and you're facing dry air, so the

1:00:27currents aren't going to flow, right?

1:00:29>> And and and you know that this thing is

1:00:31going to be exposed to the elements,

1:00:33right? There's no point in putting

1:00:34energy. How do lizards do it? I lizards

1:00:36are kind of the in between between like

1:00:37aquatic and and

1:00:38>> they don't regenerate their limbs.

1:00:40Lizards regenerate tails. Um yes. And

1:00:42and I do think that's that's

1:00:43interesting. Deer also not aquatic, but

1:00:46um interesting about deer that it's not

1:00:49loadbearing. They don't have to put

1:00:50pressure on it, you know. So I really

1:00:51think

1:00:52>> maybe the tail is kind of in the same

1:00:53>> maybe you're just sort of dragging it

1:00:55along, right? Uh

1:00:56>> it was so this biodome. So the idea is

1:00:58is is a few things. Uh aquous. So it has

1:01:01a gel that provides an aquous

1:01:03environment so that you can drive the

1:01:04currents, the injury currents and all of

1:01:06that,

1:01:06>> but also protective

1:01:08>> and and there's something really

1:01:09interesting we learned in from the frog.

1:01:11So So we did adult frogs. The adult

1:01:12frogs also don't grow their their legs,

1:01:14but we got them to grow their legs with

1:01:16the biodomes.

1:01:17>> One interesting thing we found is that

1:01:19even an empty biodome with no drug

1:01:21payload did induce some degree of

1:01:25regeneration. And so you say, why is

1:01:26that? The frog's already swimming in

1:01:27water. what's what's what was what what

1:01:29you know what what use is the is an

1:01:30empty biodome

1:01:32>> I think what's going on here

1:01:33>> is and and I don't know this but this is

1:01:35a hypothesis that I have that if you're

1:01:38a cell uh at the wound facing an

1:01:40infinite bath basically you release a

1:01:43bunch of chemicals

1:01:45[clears throat] they float off and as

1:01:47far as you know your influence on the

1:01:49micro environment has been zero it's

1:01:51you've been completely ineffective and

1:01:53so what's the point whereas in a biodome

1:01:56you have a protected micro environment

1:01:57where the the signals the the paracrine

1:02:00and the autocrine signals that are

1:02:01coming out uh they're accumulating and

1:02:04so you know as a cell you know that you

1:02:06have control over what's going on right

1:02:08>> you can expand your border kind of

1:02:10>> yeah yeah I I my that's my suspicion is

1:02:12that is that that's the other thing the

1:02:13biodome does is provide a um uh a

1:02:16controlled micro environment where the

1:02:18system itself has some some uh some

1:02:19power

1:02:20>> so like a continuously growing bioddome

1:02:22could be interesting where you start off

1:02:24with a very low volume where your

1:02:26relative signal is larger and then you

1:02:29grow the biodiversity it could be like a

1:02:31balloon I mean it could just like a

1:02:32balloon absolutely yeah see this is all

1:02:34the kind of stuff we talk about you know

1:02:35in the as an engineer so this is what I

1:02:37mean by an engineering problem like

1:02:39>> it is I don't think it's that it can't I

1:02:40think it's that it's decided not to and

1:02:42that if you provide the right stimula so

1:02:44so you need the the the delivery

1:02:46technology you need the the uh the

1:02:48environment and then you need the the

1:02:50the payload that be that's the initial

1:02:52signal that tells you like get going

1:02:54>> but I guess like maybe a less ambitious

1:02:57An easier thing would be to like get a

1:02:58deer to grow an antler in a different

1:03:00part of its body or something.

1:03:01>> Yeah. I Yes. I I've I've uh I've

1:03:05suggested the deer model system to

1:03:06people. Nobody's [laughter] taking me up

1:03:08on it. I I I'll tell you a I'll tell you

1:03:10a funny story. Uh this is the kind of

1:03:11thing. It's so sad in that in science

1:03:13like such things are almost impossible.

1:03:15There was a a a team of um a father and

1:03:18son. Bubenic was their last name. They

1:03:21lived in Canada. They had a herd of deer

1:03:23and they did an experiment that was

1:03:26probably I think 35 years and they

1:03:29discovered something incredible. They

1:03:30discovered trophic memory. So trophic

1:03:32memory basically they discovered what

1:03:34what my plenarian two-headed thing in

1:03:37mammals. What what they found was that

1:03:39if you take a deer and uh uh you take a

1:03:42you take a knife and you etch one

1:03:43particular location on the branch

1:03:45structure, it makes a little callous.

1:03:47The bone heals that year, the whole

1:03:49thing falls off. So it's gone. Next

1:03:52year, new antler rack comes up. They

1:03:54have an ectopic branch at the location

1:03:56where you made the damage. Amazing,

1:03:57right? Amazing. Uh and then that goes on

1:04:00for five or six years. Then it goes

1:04:01away. And so imagine that's like a

1:04:03single experiment is is at least 10

1:04:04years because you have to document

1:04:06before. So several years before here's

1:04:07the pattern. It's consistent. Then

1:04:09here's my my cut. Then here's repairs.

1:04:11And

1:04:12>> so I was the only person writing about

1:04:14this. So I found this work. It was

1:04:16described in the 60s uh for the first

1:04:18time in a book. Uh I was the only one

1:04:19writing about this because I said and

1:04:22this is I I use this example with with

1:04:25when I when I talk to when I lecture to

1:04:26to students like developmental biology

1:04:28students. I give a talk called things

1:04:29that are not in your textbook. What's

1:04:31not in your textbook and why is it not

1:04:32in your textbook? And I give this

1:04:33example. I say okay here's the biology.

1:04:36What does your solution to this problem

1:04:39look like? Never mind what the solution

1:04:40is. What does it look like? It's is it a

1:04:42molecular um pathway? You're going to

1:04:44draw me a c like a diagram of genes

1:04:46interacting with each other. Wh how you

1:04:48how is this gonna f first of all first

1:04:50of all it has to remember where in the

1:04:52branch structure the damage was then the

1:04:54whole thing falls off anyway so you

1:04:55can't store it locally it has to be

1:04:57where in in the scalp somewhere else in

1:04:59the rest of the body that has to be

1:05:00imprinted onto the new bone cells so

1:05:02that okay when you you know 2

1:05:04centimeters forward when you get there

1:05:05take a right like an extra branch we

1:05:07don't the standard tools that we have

1:05:09for explaining um outcomes in

1:05:11morphagenesis are not amendable to this

1:05:13we don't have the right tools yet

1:05:15>> so I was writing about this stuff and uh

1:05:18and I got a letter from Bubenic and he

1:05:20said uh I have 35 years worth of these

1:05:24antlers in my garage and I need to get

1:05:27rid of them. Do you want the antlers? I

1:05:28said I sure as hell do absolutely. This

1:05:31is like a unique uh scientific object.

1:05:33There's nothing else like this. There

1:05:34will never be anything else like this. I

1:05:36said absolutely I do. And he sent me um

1:05:3813 boxes 13 big boxes of antlers. Every

1:05:41handler meticulously labeled the name of

1:05:43the deer. Here's here's Lenny 1987, you

1:05:45know, 1986 54. And you can see we had

1:05:47them um cats scanned at the vet school.

1:05:49We sent it to the vet school in Grafton.

1:05:50We had them cat scan everything and uh

1:05:52to to you know to show what it was. No,

1:05:54no one's ever going to be able to do

1:05:55this again. You're never going to get

1:05:57money. You don't have time to you

1:05:58[laughter] know to do this kind of

1:05:59stuff. But but in pleneria but but you

1:06:01know I was I was able to tell him that

1:06:03that well at least we got a tractable

1:06:04version of this. I would love to work

1:06:06with deer. I just don't see I I don't

1:06:07see how we're

1:06:08>> What about human hair? I feel like human

1:06:10hair is kind of in the same direction as

1:06:12antlers, right? Like I think regrowing

1:06:15human hair is like a big business, you

1:06:17know. [laughter]

1:06:17>> Yeah, I get Yes, I I know where you're

1:06:20where you're coming from. Um I get

1:06:23emails all day every day about uh drop

1:06:26everything you're doing and work on

1:06:27this. Hair is definitely up there. Some

1:06:29other things.

1:06:30>> Teeth, although teeth I think somebody's

1:06:32already cracked teeth if I understand

1:06:33correctly. Yeah, there's a Japanese

1:06:35group that did it. Yeah, I think I think

1:06:37teeth will work. Um, people have all

1:06:39kinds of suggestions, some of which I'm

1:06:40not gonna um repeat in here,

1:06:42>> but if I was going to rub something on

1:06:44my head to change the polar [laughter]

1:06:47the polarity of the cells, like the the

1:06:49bias of the cells to try to get more

1:06:51hair to grow out, like is there is there

1:06:52like a simple experiment like okay,

1:06:54what's the what's the bioelectric

1:06:56pattern on the head where there is hair

1:06:58versus like a bald spot and then just

1:07:00see what happens?

1:07:01>> I mean, you you know minoxidil, right?

1:07:03Minoxidil is a potassium channel drug.

1:07:05Oh, it does. I mean, for all I know, I I

1:07:07don't know that there's a biological

1:07:08story about hair. I actually don't know.

1:07:10Um,

1:07:11>> but but it isn't it suspicious that that

1:07:15the hair drug that's out there is in

1:07:16fact an ion channel drug.

1:07:18>> Okay, that's interesting. [laughter]

1:07:18>> Right. That is interesting. So, yeah. I

1:07:20mean, we you're right. We're not focused

1:07:22on on hair. We've got a bunch of other

1:07:24stuff because I I just feel like enough

1:07:26like that's that's a target that

1:07:28everybody sees because of the the

1:07:30commercial implications. So, we were

1:07:32focused on other things. But yeah, maybe

1:07:34>> I I wanted to get a chance just to

1:07:35quickly discuss uh I don't know if this

1:07:37is all the same kind of uh topic in your

1:07:39head. But the tame theory and these like

1:07:42platonic world of stable states.

1:07:44>> That's what I was that's what I was

1:07:45getting at

1:07:46>> like how do you think about it to have

1:07:47this like idea of because otherwise if

1:07:49it's like

1:07:50>> if it's kind of yay you get these states

1:07:53that are stable but maybe they're

1:07:54infinite, right? and and they're kind of

1:07:56unrelated to each other across species

1:07:57or is it is the idea that there's

1:07:59relatively few stable ones and maybe

1:08:01different species find the same stable

1:08:03state or like what how do you think

1:08:05about this?

1:08:06>> Yeah. Um

1:08:08[laughter]

1:08:09I'm sorry you guys are free I'm sorry

1:08:11you're freezing. Um okay. Uh what you

1:08:15just said is reasonable and I think

1:08:17mostly true and that's I think that's

1:08:19how most people think about it. Um, I'm

1:08:23I my my my idea is more weird and

1:08:26radical than that. I mean, I think I I

1:08:28think what what you're saying is true in

1:08:30that and this is this is how you know

1:08:32over the years I've been able to like

1:08:33say progressively sort of more out there

1:08:35things as as the data catch up. Um

1:08:39the it is it is true I think that the

1:08:41tissue is deeply reprogrammable. In

1:08:43other words, the genetics sets the

1:08:45hardware. It tells you what are the

1:08:47protein that you know the channels and

1:08:48everything else that's going to be

1:08:49there. But we already know that certain

1:08:51kinds of hardware and bi biology

1:08:53provides amazingly um flexible and

1:08:55reprogrammable um hardware uh can be

1:08:57host to all kinds of patterns. One head,

1:08:59two head, you know, and all kinds of

1:09:00weird things.

1:09:01>> Cancer.

1:09:02>> Sure. Yeah. So there so there's there's

1:09:04a lot of there's a lot of information

1:09:05that's processed and generated at the

1:09:07physiological level and it can be

1:09:09different um exactly the same set of

1:09:11cells can hold multiple different

1:09:12patterns and so on. So so that's so

1:09:14that's all true. I I'm I'm saying

1:09:16something something much more um

1:09:18disruptive than that. Here's here's how

1:09:21there's a couple of ways to to argue

1:09:23into it. Here's here's how I I one way I

1:09:25can start. Um the standard way consider

1:09:28consider uh how we explain the

1:09:32properties of bio of of certain organ of

1:09:34of organisms. So I show you an organism

1:09:36and I say why does this thing look the

1:09:38way it looks and um and and have the

1:09:40capabilities that it does and the

1:09:41properties that it does. The standard

1:09:43modern answer is well selection of

1:09:46course it faced a series of environments

1:09:47over the years. Um everything that

1:09:49didn't look and act like this died out

1:09:50and this is what we have left to

1:09:52selection forces have shaped this thing.

1:09:54I say okay and uh when did we pay the

1:09:57computational cost of designing

1:10:00something that is so competent in its

1:10:01environment? Well over the eons of of of

1:10:04evolution of course by by this genome

1:10:06bashing against the environment. That's

1:10:07the that's that's how you pay the

1:10:08computational cost there. Okay. Very

1:10:10good. Now, now we have um zenobots and

1:10:13anthrobots and we remove them or we have

1:10:17we made tadpoles with eyes on their

1:10:18butts or on their tails and they can see

1:10:21you. You make these things. There is no

1:10:23there is no rounds of selection or

1:10:25adaptation right out of the box. Zero

1:10:27shot, you know, learning kind of thing

1:10:29out of the box. They have novel shapes,

1:10:32novel physiological traits, novel

1:10:34transcripttos, novel behaviors, novel

1:10:36competencies like kinematic

1:10:37self-replication, all this stuff.

1:10:38There's never [snorts] been any

1:10:39xenobots. There's never been any

1:10:40anthropods. There's never been selection

1:10:42for this stuff. Where do these things

1:10:44come from? First of all, and more pre

1:10:46more more precisely, like that's a

1:10:47generic question, but a more precise

1:10:48scientific question is when did you pay

1:10:50the computational cost to to to to have

1:10:53these very specific new properties?

1:10:55>> Um, so what people usually say at that

1:10:58point is, well, it's emergent. They say,

1:11:01well, what does that mean? Because

1:11:02emergence to me emerges just means

1:11:04surprise. It just means really it just

1:11:05means you didn't see it coming for

1:11:07whatever reason. Uh, so what does that

1:11:09mean? and they say, "Well, these are

1:11:10just um uh you know, at the time that

1:11:13you were selecting for humans and frogs,

1:11:15you also got this."

1:11:16>> Mhm.

1:11:17>> So, there's two problems with that. The

1:11:19first problem is that it's all well and

1:11:21good to say after the fact that, well, I

1:11:23guess that's what you got, but that's

1:11:25not what we want. We want to be able to

1:11:26predict and control these things, right?

1:11:28>> It's like what physicists do when we use

1:11:29the anthropic principle. It's like it's

1:11:31there because it's there. [laughter]

1:11:33Well, and that's we should that's

1:11:34something else we can talk about because

1:11:36actually we have we have some data

1:11:38coming out that um I think there's a

1:11:39there's an anthropic principle for the

1:11:41platonic space as well that's pretty

1:11:42wild. Um but but the the but the other

1:11:46issue is that it basically breaks the

1:11:48whole point of evolutionary theory

1:11:50because the whole point of evolutionary

1:11:52theory is to say that there is a tight

1:11:53specificity there's a correspondence

1:11:55between the way you are now and the set

1:11:58of environments that led to you being

1:12:00here. If you can tell me that, yeah, all

1:12:02the selection forces led to to a human.

1:12:04Oh, and by the way, also at the same

1:12:06time, for some reason, anthropos that

1:12:08have nothing to do with it, I'm say,

1:12:09well, you've you've ripped up the

1:12:10specificity of this. Then you you could

1:12:12have gotten anything. And and

1:12:13>> well, could it just be that you're just

1:12:14to play devil's advocate, like you're

1:12:16just activating sub routines. So, for

1:12:18example, like if you take apart a

1:12:19computer, you can get like an LED to

1:12:21blink, you know, on some board. Uh, and

1:12:23it's like it if it was an evolved

1:12:25system, like you know that LED does

1:12:27something else, but then like if you

1:12:28take a component out, you can get some

1:12:30behavior that's interesting.

1:12:32>> So, so getting some behavior that's

1:12:34interesting is kind of nonfalsifiable.

1:12:36You can always get behavior that's

1:12:38interesting, but when you get kinematic

1:12:40self-replication, there's never been, to

1:12:42our knowledge, there's never been an

1:12:44ancestor that that that did kinematic

1:12:46self-replication. There is no other

1:12:47animal on Earth that does it that way.

1:12:49uh there's never been selection forces

1:12:51to be able to do it. We h we need as

1:12:54scientists, we need a theory of where

1:12:56specific competencies come from and and

1:12:58how do they get there and standard

1:13:00evolution does not does not give you

1:13:02that.

1:13:02>> Um so that's that's one way to start

1:13:04thinking about this question of

1:13:07where where what is the what is the

1:13:09latent space of possible patterns that

1:13:11appear when you provide certain

1:13:13interfaces. I I see embryos, robots,

1:13:17computational media,

1:13:19all of it as different interfaces to a

1:13:22latent space of specific patterns. Some

1:13:24of those patterns you can tell an

1:13:25evolutionary story about. Some of these

1:13:27patterns do you can't tell an

1:13:29evolutionary story about, but you need

1:13:30to understand the structure of that

1:13:32space. What what uh you know what what

1:13:34is what are the poss what are the

1:13:35contents of that space? And so that so

1:13:38far so I think pretty tame what I've

1:13:41just said like that's not a big deal but

1:13:44uh but I will but I'll make one other uh

1:13:46uh modification to that and then I could

1:13:48tell you about some some work that we've

1:13:49done on this the standard I mean there

1:13:52are plenty of uh mathematicians who have

1:13:55a a platonist view which they say look

1:13:57there are facts that are simply not

1:13:59facts of physics the facts that the the

1:14:01the fact that quitterians don't act the

1:14:04same way as octanians don't act the same

1:14:05way as as the complex numbers. You don't

1:14:08you you you don't get that from physics.

1:14:09You can't fire the math department and

1:14:11hope the physicists will figure it out.

1:14:13You can't change those facts by changing

1:14:15the fundamental constants of the

1:14:16universe. They're just there's a set of

1:14:18facts that are not facts of physics and

1:14:20those facts are not random. They come

1:14:21from a structured um ordered space which

1:14:24we can investigate and study and and

1:14:26okay so there's some mathemat

1:14:28>> but but the standard assumption there is

1:14:30that those forms are only of relevance

1:14:32to mathematicians. In other words, they

1:14:35what what do what if you ask them if you

1:14:37ask people who are plaintists, what does

1:14:39the Platonic space contain? They will

1:14:40say, well, it contains mathematical

1:14:41truths.

1:14:42>> So my my

1:14:44>> Well, there's there's definitely some

1:14:45topological things that affect biology.

1:14:48Like I mean every almost everybody is a

1:14:50tube. Like I'm like a worm with hands

1:14:52that are putting putting things into the

1:14:54mouth. Sure.

1:14:55>> So like and it's a tube because it's

1:14:57space is threedimensional and you want

1:14:58to have an inside and outside and like

1:15:00some you know surface area.

1:15:01>> For sure. For for sure. And for sure and

1:15:04those things, you know, I stay away from

1:15:06those kind of examples and things like

1:15:07pi and so on because they sound very

1:15:10spatial and then people say, "Ah, well,

1:15:12but spacetime, you know, if it wasn't

1:15:14three-dimensional, then this and that

1:15:15would happen." So I I so I stay away

1:15:16from that for that reason. I like things

1:15:18like e the natural logarithm and I like

1:15:21things like Fenbomb's constant like what

1:15:23mathematic what what constant of physics

1:15:25would you tweak to get a different

1:15:27Fenbomb's con like good luck to you. So,

1:15:29so what so but but beyond that my

1:15:33suspicion is that the kinds of things

1:15:35that mathematicians study are only one

1:15:37sort of the bottom layer of that space.

1:15:39I think that space has other patterns in

1:15:41it that are not static patterns like the

1:15:43value of E and things like that. They

1:15:46are uh behavioral policies. there are

1:15:48dynamic behavioral policies that we

1:15:50would recognize as kinds of minds and

1:15:52that when you make embodiment you pull

1:15:54down some of these patterns that end up

1:15:56being behavioral traits uh behavioral

1:15:59competencies goal states and so on that

1:16:02uh that are going to basically inhabit

1:16:04the physical interface that you made.

1:16:06Now the thing with biology is that uh

1:16:09it's very complex and you can never

1:16:10really prove anything because there's

1:16:12always more. Somebody's always going to

1:16:13be like, well, maybe there's quantum

1:16:14microtubules that you never like you can

1:16:17never know that you've got all the

1:16:18mechanisms. So, what we've been doing is

1:16:21making minimal computational models,

1:16:24very minimal small computational models

1:16:26where you know exactly what you've put

1:16:28in and what you got out and you can

1:16:30quantify it because there is no extra

1:16:31mechanism. There is no magic. You can

1:16:33see all the steps. And so we had one

1:16:35paper on this last year and we have half

1:16:37a dozen more coming uh coming this year

1:16:39on different uh versions of this the

1:16:42simplest version that we published but

1:16:44we have an even simpler version coming

1:16:46out for for shock value. I like it when

1:16:47they're tiny because the shock value uh

1:16:50you know lots of people think that well

1:16:53I'm a human. My story is not told by the

1:16:56laws of biochemistry. I there's more to

1:16:58me than just the mechanical laws of

1:16:59biochemistry. But but don't worry that's

1:17:01okay because over here we have these

1:17:02dumb machines and the and the story of

1:17:04dumb machines are absolut is absolutely

1:17:06told by our formal models of of touring

1:17:09you know computation and all of that

1:17:10they they do exactly what our formal

1:17:12models say no more no less we somehow a

1:17:14little bit more than that

1:17:16>> um so so so my thought is that it's not

1:17:18only us that uh are are benefiting from

1:17:21these and evolution that's benefiting

1:17:23from the free lunches from that platonic

1:17:25space it's everything including the dumb

1:17:27machines so so our first so our first

1:17:30model was sorting algorithms. So bubble

1:17:32sort

1:17:32>> mhm

1:17:34>> less than six lines of code there's no

1:17:35you can see exactly it's deterministic

1:17:37you can see everything there is

1:17:38>> and what you see and and and people have

1:17:40been studying that for what 80 years

1:17:42everybody's been been you know every CS

1:17:44101 student has been playing with it

1:17:46>> what we found is that

1:17:48>> yeah it sorts numbers all right but if

1:17:50you if you look at it from a slightly

1:17:52different perspective and you ask what

1:17:54else might it be doing nobody asked that

1:17:56because because you can see the steps so

1:17:58it shouldn't be doing anything else

1:18:00because the al because we have this idea

1:18:01that it does what the algorithm tells it

1:18:03to do just like we think well the laws

1:18:05of physics kind of constrain what we do

1:18:06but but we can kind of do more

1:18:08interesting things but but this thing

1:18:10should just do what the algorithm says.

1:18:12Turns out no they do other things and

1:18:14some of these other things are not just

1:18:17it's not just complexity. It's not just

1:18:18unpredictability. It's not just perverse

1:18:20instantiation. It's competencies that

1:18:23are recognizable to any behavioral

1:18:25scientist. In other words, what's

1:18:26emerging?

1:18:27>> What does bubble sort do?

1:18:28>> What it does two things? It does delayed

1:18:30gratification which which I can

1:18:32describe. It does clustering. Um so so

1:18:34delayed gratification. So imagine

1:18:36imagine um uh sorting sorting numbers.

1:18:39You can you can plot uh and I I can send

1:18:42you a a link to this. Um Andrea Morris

1:18:45from Forbes did an hour and a half

1:18:47discussion with me on this because it

1:18:48was so like in like why why does Forbes

1:18:50want to hear about this? Because it

1:18:51actually has massive implications for um

1:18:54for how we understand uh IP and computer

1:18:56science and and beyond that. Uh imagine

1:18:59imagine um looking at uh how sorted is

1:19:03my array at any given point. Right? So

1:19:05as you sort it just keeps going up and

1:19:07up and because bubble sort is guaranteed

1:19:08to to to converge. Eventually you get to

1:19:10it's 100% sorted, right? Guaranteed.

1:19:12>> Imagine if imagine if you introduce um

1:19:15broken numbers brok the standard version

1:19:17of bubble sort there's no code to see if

1:19:19if it worked. I tell the four and the

1:19:21six to flip. I don't check to see if it

1:19:23has. I assume it's a reliable medium. I

1:19:25assume suppose you introduce a broken

1:19:27number. What's a broken number? Well,

1:19:28the algorithm says to swap. I don't

1:19:30move. I'm glued down. I can't move.

1:19:32Okay, there's no code. We don't change

1:19:33the bubble story. In other words, we we

1:19:35didn't add any code to see well what

1:19:36happens if the number doesn't work. You

1:19:37don't do that. First of all, first of

1:19:40all, still works. Okay, and there's no

1:19:42extra code to handle these broken

1:19:43things. Still works. Still still sorts

1:19:45it. What it does is it sorts all the

1:19:46numbers around it. Okay, so still works.

1:19:49But the really amazing thing is that if

1:19:52you look at how sorted is my array, in

1:19:55order to do that, you have to unsort.

1:19:57You have to go backwards. If you have a

1:19:59broken number, you can't just keep

1:20:00monotonically rising. You have to

1:20:02temporarily unsort then recoup gains

1:20:04later as you move stuff around. That's

1:20:06behavioral scientists call it delay

1:20:07gratification. It's like imagine two

1:20:09magnets on either side of a piece of

1:20:11wood.

1:20:11>> They want to get at each other, but

1:20:13they're too dumb to go around because

1:20:14they would have to go against the

1:20:15gradient. They'd have to get further in

1:20:16order. They're not smart enough to do

1:20:18that. Uh but but but most animals can.

1:20:21So So right. So if you make a thing,

1:20:22they'll they'll they'll come around

1:20:23temporarily away from your goal to get

1:20:25to recoup, right? That's that's called

1:20:26that's delay gratification. Bubble sort

1:20:28will desort the whole array. It'll

1:20:30actually go backwards in order to then

1:20:32get its problem solved later on.

1:20:34>> There is no code for that. You can you

1:20:36see the code. There isn't any code for

1:20:37that. So on the one hand, you could say,

1:20:40well, yeah, any you know, anything has

1:20:42consequences that aren't directly in the

1:20:44code. Sure. But there's two things going

1:20:45on here. This is a minimal system. Of

1:20:47course, there's there's lot much bigger

1:20:48versions of this. But first of all, the

1:20:50whole point of of writing a computer

1:20:52algorithm is meant to specify the steps

1:20:54that you think you've carried out. That

1:20:56is exactly what it does. And the extra

1:20:58stuff that it does

1:21:00>> is not just complexity. It's not just,

1:21:02you know, some some random side product.

1:21:05It is a behavioral competency that you

1:21:07can make use of. I have a whole, you

1:21:08know, sort of business plan in my head

1:21:09of how you can actually, you know, you

1:21:11could charge for that. So at the same

1:21:13time and there are no extra steps.

1:21:14There's no magic. You didn't you didn't

1:21:16change the way the CPU works. It isn't

1:21:18it's still deterministic. It's not some

1:21:19quantum thing. Uh it is it is um the the

1:21:23ability of of of

1:21:25interesting patterns and some of these

1:21:27patterns are static but some of them are

1:21:29behavioral tendencies to ingress into

1:21:31things that we think of were specified

1:21:33whether by the laws of physics,

1:21:35chemistry or or or algorithms. In the

1:21:38spaces between the chance and necessity,

1:21:40there are spaces for these things to

1:21:41squeeze in and they can occupy I think

1:21:43all

1:21:46sort of the whole spectrum of complexity

1:21:48from very static things like e to very

1:21:50complex things that could be kinds of

1:21:52minds.

1:21:53>> So it's very hard to boil biology down

1:21:55to these kinds of simple algorithms

1:21:56because you have all these hidden

1:21:57variables and things you don't know on

1:21:58the cellular level. But I'm curious if

1:22:00you have an idea for like a model

1:22:03system, biological system that could

1:22:04like falsify or prove this idea of

1:22:07having like these platonic

1:22:09>> patter

1:22:11falsify. That's easier maybe.

1:22:13>> Well, nobody nobody's ever proved or

1:22:14falsified really anything in biology

1:22:16because it's too hard to read

1:22:17definitively to do that the way you can

1:22:19in with the computational systems. But I

1:22:21but I will say that uh I think I think

1:22:24and it's not that I think the xenobots

1:22:26or anthropots prove platonism. I you

1:22:28know that's not how it works. But I do

1:22:30think that the more we study these

1:22:31things and there's a bunch more papers

1:22:33coming uh soon. The more the more you

1:22:36realize that there are competencies here

1:22:37that we do not have a conventional story

1:22:40of origin for you have to do one of two

1:22:42things. You have to either say like most

1:22:44people they shrug. They say these are

1:22:46these are just regularities that hold in

1:22:48our world and they say well is do they

1:22:50come from a random bag of regularities

1:22:53like I I don't you know I think that's a

1:22:56very pessimistic approach. I think no it

1:22:58should be we should assume that there

1:22:59there's a structured latent space that

1:23:01we can study and so there it is. That's

1:23:03all I'm that's all I'm saying. Um but

1:23:05but the other the other thing is that uh

1:23:08we have to uh we have to I think uh

1:23:11adjust the way that we calculate uh

1:23:15compute and cost and effort. You know

1:23:17when right right now uh and and I'm

1:23:20sorry I'm going to have to go in a

1:23:21minute but but right now there are three

1:23:23ways to have something that's competent.

1:23:25You can either design an algorithm

1:23:27meaning somebody else already knows how

1:23:28to solve the problem and they and they

1:23:30write down the steps. You can evolve it,

1:23:32meaning generate a trillion variants,

1:23:33throw away everything that doesn't work

1:23:35and keep what's left or you can learn

1:23:36that that same being has multiple

1:23:38encounters with the problem. Those are

1:23:40the only three ways and you have to put

1:23:41in effort. I I I think and we now have

1:23:44data that's that's that will be out soon

1:23:46showing that the way we account for that

1:23:48effort is is is wrong. We there there is

1:23:50there is at least one other place where

1:23:53you can get more than you put in. You

1:23:54can get more out than you put in.

1:23:56Exactly what do you get out? We're still

1:23:58working on that. uh be like figuring out

1:24:00where on the spectrum it is. But I think

1:24:03the way we we account for um for uh

1:24:07actionable information for intelligence

1:24:09coming into the world is we we we've

1:24:11been neglecting an important source of

1:24:12it and and our standard ways don't

1:24:14account for it.