A conversation with Andrey Vyshedskiy about the evolution of language and brain development
Andrey Vyshedskiy discusses the evolution of language, underlying brain mechanisms, and implications for developmental disorders such as autism. The conversation also covers his research into developing related therapeutic approaches.
Watch Episode Here
Listen to Episode Here
Show Notes
This is a ~1 hour conversation with Andrey Vyshedskiy (https://www.bu.edu/prsocial/profile/andrey-vyshedskiy/) about the evolution of language, brain mechanisms, implications for autism and developmental disorders, and Andrey's work to develop therapeutic approaches.
CHAPTERS:
(00:00) Syntactic language network development
(05:37) Three levels of comprehension
(28:55) Neurobiology of sensitive periods
(36:46) Recognizing the syntactic phenotype
(42:24) Theory of mind mechanisms
(48:51) Future levels of intelligence
PRODUCED BY:
SOCIAL LINKS:
Podcast Website: https://thoughtforms-life.aipodcast.ing
YouTube: https://www.youtube.com/channel/UC3pVafx6EZqXVI2V_Efu2uw
Apple Podcasts: https://podcasts.apple.com/us/podcast/thoughtforms-life/id1805908099
Spotify: https://open.spotify.com/show/7JCmtoeH53neYyZeOZ6ym5
Twitter: https://x.com/drmichaellevin
Blog: https://thoughtforms.life
The Levin Lab: https://drmichaellevin.org
Transcript
This transcript is automatically generated; we strive for accuracy, but errors in wording or speaker identification may occur. Please verify key details when needed.
Main Episode
[00:00] Andrey Vyshedskiy: Maybe I should start with a little bit of history of how I came to this. I became interested in how imagination works when I was nine years old. Over the years, I approached the question from several points of view. My undergraduate degree was in astrophysics. The master's degree was in engineering. My PhD was in neuroscience. The most productive approaches to understanding imagination were developmental, neurological, and evolutionary. And I can start with the developmental approach and then talk about the evolutionary approach. The brain networks for complex language develop after birth. We are not born with these networks. These networks develop through engagement in conversations and storytelling during an early childhood sensitive period. The magic of stories is that they contain a subject and an object. To understand the story, a child has to imagine what the subject does to the object. And that process of imagination provides the necessary exercise for the brain networks to grow. For neuroscientists, we are talking about the frontotemporal and frontoparietal networks. These are the networks connected by the arcuate fasciculus and superior longitudinal fasciculus. Basically, these are the networks that go from the front to the back. These are some of the longest networks in the brain. Sentences with a subject and an object are called syntactic. Syntactic sentences explain who did what to whom. So syntactic storytelling trains the brain networks that support syntactic comprehension. And this makes sense because it works for everything else. Practicing the piano trains the ability to play the piano. Practicing tennis trains the ability to play tennis. Riding a bicycle trains the ability to ride a bicycle, and so on. Similarly, children need to practice syntactic conversations to develop brain networks that support syntactic conversations.
[02:43] Andrey Vyshedskiy: When syntactic conversations and storytelling are replaced by nonsyntactic conversations or passive video viewing, these networks do not develop. In most autistic children, the sensitive period for development of these networks is shorter than average. Missing opportunities for syntactic conversations and storytelling before age four, and especially before age three, may deprive an autistic child of the opportunity to develop the brain networks underlying syntactic comprehension. Individuals without syntactic comprehension usually remain intellectually disabled and cannot live independently. Maybe I should give you an example of deaf children, children who are congenitally deaf. About ninety percent of deaf children are born to hearing parents. In the United States, these children are identified early. Their parents learn a formal sign language. Formal sign languages are syntactic, so children are immersed in syntactic conversations and storytelling from an early age and develop normally. In contrast, in remote Latin American villages, deaf children are often not diagnosed. When parents recognize that the child is deaf, they start to use simple gestures: "Come to me," "sweep the floors," "wash the dishes." These simple gesture systems are called home sign or kitchen sign. Home sign systems are nonsyntactic. They cannot communicate what a subject did to an object. Home sign systems lack spatial prepositions, possessive pronouns, verb tenses, and other grammatical elements characteristic of a formal sign language. As a result, home signers grow up without syntactic conversations and storytelling. Inevitably, this leads to syntactic network abnormalities and lifelong impairment in syntactic comprehension. When such home signers arrive in the United States, they undergo extensive language therapy. However, after the onset of puberty, the syntactic networks cannot change and children cannot develop syntactic comprehension. Home signers highlight the critical role of syntactic conversations in brain development. Formal sign languages are syntactic. Parents can tell stories explaining who did what to whom and exercise the syntactic networks in children. Home sign is nonsyntactic. Parents cannot communicate who did what to whom. As a result, syntactic networks remain undeveloped.
[05:27] Michael Levin: By the way, just to confirm one thing, you're not showing any slides?
[05:31] Andrey Vyshedskiy: Correct? I'm not showing any slides.
[05:34] Michael Levin: Just making sure that I'm not missing.
[05:35] Andrey Vyshedskiy: Okay.
[05:36] Michael Levin: No no problem. Keep going.
[05:37] Andrey Vyshedskiy: Back to autism. Autistic children have low interest in social communications. They usually avoid conversations and storytelling. As a result, about forty percent of autistic individuals do not develop syntactic comprehension. They do not understand who did what to whom. For these children, for their parents, it's a real tragedy. These children don't understand spatial prepositions, verb tenses, possessive pronouns. They don't understand fairy tales. They don't understand complex explanations. They cannot even get a driver's license because they cannot pass the learner's permit test since the test is structured using syntactic sentences. You have to imagine the scenes in the learner's permit test, and this process is syntactic and these children cannot do this. As a result, these individuals cannot live independently and they're often institutionalized away from halfway houses. This is usually a tragedy when the child grows up. About twenty years ago, I posed the question, given that many autistic children have low interest in social communications, could we develop non-conversational exercises to train the same brain networks for understanding who did what to whom? And if such exercises could be developed, would it be possible to deliver them during development, during the peak time for syntactic development between the ages of two and four years? To address this challenge, we developed Mental Imagery Therapy for Autism, MITA, a gamified app. I told you about this app about ten years ago. This app was successful. It climbed up to the top of the app stores, and over the twelve years, it was downloaded by over 3,000,000 families. We conducted several studies of MITA and we have just finished a 2.5-year randomized controlled trial. We used a standard measure of autism severity, the Childhood Autism Rating Scale. The control group improved by 4.8 points. The MITA group improved by eight points. This difference was statistically significant and clinically meaningful. And we are now working with the FDA to allow doctors to prescribe MITA for autistic children. Now MITA collected a lot of data just because parents have to send us reports on their children's development every three or four months. And over twelve years, we collected more than 300,000 evaluations. This is the biggest database of autistic development in the world. We have published more than 20 research papers using this data. We studied how different dietary and cultural factors affect the development of autistic children. But our biggest discovery was something else. The biggest discovery in my mind, that was published in four major papers, showed that syntactic comprehension is organized around three stable attractor states. In other words, the intellect, cognition, reasoning, thinking, all these words are not continuous but have three levels. At the lowest level, the command level, individuals understand single words. At the second level, the modifier level, individuals understand combinations of nouns with adjectives.
[10:17] Andrey Vyshedskiy: They can correctly follow an instruction: find the large green straw from a set of straws, pencils, and Lego pieces of different colors and sizes. At the third top level, the syntactic level, individuals comprehend spatial prepositions, complex syntactic explanations, and fairy tales. In these four Nature papers, we looked at nearly 100,000 autistic children, and the majority of them get stuck either at the command level or at the modifier level. Some may acquire a higher level later on. However, the probability of upward progression decreases after five years of age and ceases after about 10 years of age. More than half of our participants remained at a lower level. The importance of this discovery is really hard to underestimate in my mind, and it applies to many different scientific fields. For example, mainstream linguistics never characterized the modifier level. Linguistics doesn't even have terminology for the modifier level. For clinicians, the three levels of comprehension are particularly consequential. Many language therapists and ABA therapists are focusing on learning words. But learning words does not develop the brain syntactic networks, these frontoposterior networks, and does not lead to syntactic comprehension. The goal must be combining words. First, we combine nouns and adjectives and reach the modifier level, and then we combine nouns and spatial prepositions and reach the syntactic level. This has to be done while the brain syntactic network is still developing, that is during the sensitive period, which is shorter than average in autistic children. Why do language and ABA therapists often focus on learning words? One reason is that mainstream assessments are evaluating a child's vocabulary. That has to change. Language assessments should identify the level of syntactic comprehension, not just vocabulary or categories or pragmatic understanding. The most exciting part is human evolution, and you'll definitely be interested in that. This is where this discovery has the biggest implication. Human evolution likely transitioned from the command to the modifier to the syntactic level. It wasn't a linear increase from less intelligence to more intelligence; it was this stepwise evolution. Humans diverged from chimpanzees about six million years ago. Our species likely reached the modifier level around two million years ago. The modifier level enabled them to understand combinations of nouns and adjectives. At this point, one should ask: two million years ago, there was no articulate speech. That is totally right. Our ancestors probably used gestures: "Give me this stone," "Give me the bigger stone," "Give me a smaller stone," "Give me a longer stick," "Give me a shorter stick." How do we know that the modifier level was acquired two million years ago? From human manufacturing of stone tools. Our ancestors started to manufacture symmetrical stone tools about two million years ago. I have brought such a tool. This is a Mode 2 tool, symmetrical tool.
[14:56] Andrey Vyshedskiy: It can be retouched. It can be sharpened, and people used that for two million years until really recently. If you dig Indian sites in the United States, you find many of these tools. On the neurological level, manufacturing of these symmetrical stone tools is underpinned by the same network that is required for the modifier-level comprehension. And maybe we should try this. Recall your car. Now imagine your car much bigger, now much smaller. This is the ability our ancestors developed two million years ago. And therefore, they could understand sentences: "Give me a bigger stone, give me a smaller stone," and they could manufacture symmetrical stone tools. This is really the first uniquely human cognitive ability. No animal has ever been trained to comprehend novel combinations of nouns and adjectives. We discussed these uniquely human cognitive abilities thirty years ago, and this is the first cognitive ability that is uniquely human. You can train chimpanzees in understanding words, single words, like adjectives and nouns, but you cannot train chimpanzees or any other animals to combine those words in a novel fashion. It's a big deal. They just don't seem to be able to do it. We now understand how it works on the neurological levels and the reason why chimpanzees cannot acquire those skills. The reason is simple, because their frontotemporal and frontoparietal connections are much shorter. They just don't go as far as human connections. The modifier-level network, this frontoparietal and frontotemporal network, also enabled our ancestors to imagine how the fire gets bigger and smaller. As a result, they developed a model of fire behavior and they could control the fire. So acquisition of the modifier level enabled them to control the fire. Most importantly, the modifier-level cognition slowly drove the development of articulate speech from about two million years ago. We find skeletal changes in *Homo erectus* and then *Homo habilis* and other ancestors of ours that show that articulate speech develops from about two million years ago onward. By about six hundred thousand years ago, this is the time when we split from Neanderthals, our modifier-level ancestors were likely speaking in full sentences in the same articulate speech as we are using now. Isn't that something? Half a million years ago, our ancestors and Neanderthals were using articulate speech to communicate. Despite being articulate, their IQ was likely around 70 using modern units. And of course, cognitively, they were different. They didn't have the syntactic level. They were at the modifier level. And this is the biggest, the most interesting part: the cognitive revolution. Seventy thousand years ago, the behavior of our species completely changed.
[19:36] Andrey Vyshedskiy: The best explanation for why it changed is acquisition of the syntactic level. This acquisition of the syntactic level was quickly followed by a technological revolution: the appearance of the bow and arrow, the construction of sophisticated dwellings, religious thinking reflected in adorned burials, sewing needles with an eye, musical instruments, figurative art. I brought some of these toys. This is the Lion-man from Germany, head of a lion and a body of a man. This was built 39,000 years ago. And, of course, everybody is familiar with these Venuses, and they're approximately from the same time. Importantly, the tribe that acquired the syntactic level 70,000 years ago experienced explosive growth and conquered the world. I have recently published a paper, "How One Small Tribe Conquered the World 70,000 Years Ago." In that article, I summarized five lines of convergent genetic evidence for one small tribe conquering the world 70,000 years ago. The total domination of a single tribe was hard to understand outside of the framework of the modifier-to-syntactic level transition. What happened 70,000 years ago deserves special attention. Before the shift from the modifier to the syntactic level, all adults possessed only the modifier phenotype. They were unable to engage children in syntactic conversations. As we know from home signers, even children who are genetically ready for the syntactic level do not acquire the syntactic level without engaging in syntactic conversations and stories. And because acquisition of the syntactic level is limited to the sensitive period, they could not acquire it later in life. Consequently, they were also unable to provide the next generation with syntactic language input. This cycle could only have been broken by several children who independently developed syntactic language. I named these children Romulus and Remus, the legendary founders of Rome. Romulus and Remus must have introduced spatial prepositions into an existing modifier communication system and engaged one another in syntactic conversations and storytelling during their sensitive period. There are actually several historical examples of the invention of syntactic languages. The best studied invention of syntactic language occurred in Nicaragua. In the 1970s, deaf children were assembled in one place—actually in two places—and left on their own. They spontaneously developed a fully syntactic language capable of explaining who did what to whom. It is now known as Nicaraguan Sign Language. Everybody is familiar with this Chomskyan idea of mutation. Was it really a mutation that triggered acquisition of the syntactic level 70,000 years ago? And the related question is, why didn't Romulus and Remus invent syntactic language earlier? I think the answer to this is that the sensitive period was too short until 70,000 years ago. As we know from modern autistic children, when the sensitive period is too short, children cannot acquire the syntactic level at all. Forty percent of autistic individuals do not acquire the syntactic level despite intensive language therapy. We recently published a study in which we measured the syntactic learning rate in autistic and typically developing children. In typically developing children, the learning rate stays constant and high until at least seven years. In autistic children, the onset of decline occurs at approximately three and a half years in mild, two years in moderate, and 1.4 years in severe autism. When the onset of decline is two years or less, it is probably impossible to reach the syntactic level despite any amount of language therapy.
[24:15] Andrey Vyshedskiy: When the onset of decline is three years or less, it is probably possible to reach the syntactic level if the language therapy starts very early. Before 70,000 years ago, the onset of decline in our ancestors was probably around two years of age, and there was no way they could invent syntactic language. Romulus and Remus were likely born with very late onset of decline, maybe around four years. Just like Nicaraguan children, they invented some special prepositions and told stories to each other and thus were able to self-develop their syntactic networks before the end of their ascendancy period. Saltationists are probably right: there was a mutation that delayed the onset of decline in two or more children. The important thing, though, is that once Romulus and Remus's tribe had acquired syntactic language, they must have acted as a magnet for multiple mutations that delay the onset of decline. In fact, the whole evolution of man is the story of accumulating mutations that delay the onset of decline and prolong childhood. This is called neoteny. What do we do? What's the future for this? The biggest goal is, of course, to enable more children to reach the syntactic level. Reaching the syntactic level opens the door for independent living and employment that protects individuals from being sent to institutions and prisons. The low-hanging fruit is introducing syntactic language exercises as early as possible. We have an epidemic in our society of addictive video content like YouTube. Imagine a hypothetical case of a child watching videos all day long. Videos are passive. Videos completely displace syntactic conversations and oral storytelling. A child does not exercise his voluntary imagination. Without training, the syntactic network does not develop, and the child ends up with intellectual disability and no syntactic level. In most cases children do not spend all day watching videos, but for children with an early onset of decline, it is essential to experience as much syntactic exercise as possible as early as possible. This is the reason we developed MITA syntactic language exercises. Parents can download and start MITA exercises as soon as they see the first signs of trouble, language delay. They can evaluate a child's comprehension at home. Let me give you a few examples of how you can evaluate your child at home: - By age two, your child should follow instructions such as "Give the cup to the monkey," "Give the pencil to the dog." - By age 2.5, your child should identify a green pencil, a red Lego, and a blue straw from a set of objects of different colors. - By age three, your child should be able to combine size adjectives with nouns. He should identify a small pencil and a large Lego from a set of objects of different sizes. - By 3.5, your child should be able to combine the size and the color with nouns. He should be able to identify a small red pencil or a large blue Lego from a set of objects with different sizes and colors. If your child has missed any of these fundamental developmental milestones, eliminate passive video watching completely, read simple fairy tales to your child, and seek professional language therapy. It's very important. You can save your child's future. In the future, researchers should be able to find pharmaceuticals to prolong the sensitive period that will enable exercises to act on the developing brain, but it will always be exercises. Only this mental gymnastics, these exercises, the voluntary imagination, listening to fairy tales—this is what is training the syntactic language and high intelligence. If you can come up with some drugs that change bioelectricity and enhance the sensitive period, that would be great.
[28:55] Michael Levin: Amazing. Very, very interesting, and really important work on all this. A couple of things to talk about. So enhancing those sensitive periods, do we know what is it about the neural networks that render them sensitive? Is it an issue of more neurogenesis, more connectomic plasticity? In bioelectricity, the one thing you can do is you can prolong cell cycle competency. So you can take mature, terminally differentiated neurons, depolarize them for a couple of days, and they'll reenter the cell cycle, and we can do that. And I'm not arguing that bioelectricity is necessarily the way to do this. Since you mentioned it, there are ways. But I'm just curious: what is it that we need to target? Is that known?
[29:52] Andrey Vyshedskiy: This is current thinking. Maybe there will be other ways of approaching this. But what we know is that myelination of axons blocks synaptogenesis, blocks the formation of new synapses. The idea is that these frontoposterior networks, temporal and frontoparietal, need to be fine-tuned. They need to be myelinated so that you end up with synchronous networks. But before you build the synchronous networks, actually, while you're building them, you're also branching the axons. These branches can be very long. A single axon coming from the thalamus, for example, can branch 10 centimeters away. The same axon can go and make synapses at multiple areas of the cortex. Myelination stops that branching. Maybe that plays a role, and this is as much as we know. There was a really great paper by Philipp Khaitovich. Philipp looked at synaptogenesis in the prefrontal cortex. The prefrontal cortex is behind the forehead, and he compared a whole group of RNA molecules responsible for synaptogenesis. The time course in humans is delayed compared to chimpanzees by about five years. The difference is really amazing. These are the same RNA molecules that are making the same proteins, but they delay by five years from chimpanzees to humans. And then Philipp Khaitovich looked at autistic brains. Of course, these are frozen brains from people who died. The peak in autistic brains is shifted back to chimpanzees by as much as three to four years. Different fields of science use different terms. The most popular is neoteny. In humans, childhood is longer. We are neotenic. We keep childhood features. I'm not sure how the word neoteny applies here. It's really that there is a delay in the development of the prefrontal cortex, and more specifically a delay in the development of networks connecting the lateral prefrontal cortex to the temporal and the parietal cortices. This is the delay that we are after. And this is a million-dollar question: how can we reverse that delay to the average human level, which is around five years of age?
[33:44] Michael Levin: One of the implications of what you're saying is that generic regenerative therapies, which we and many others are trying to get to, where we can basically reinstate a plastic condition that will induce the regrowth and repair of tissue, including the brain. What it sounds like is that that ought to then be a corrective as much as it's going to be, I think, a corrective for all kinds of somatic birth defects and injuries and so on. It sounds like it might, by reopening that window again at whatever age, actually have an impact on aspects of these conditions that you're talking about. That's pretty cool.
[34:28] Andrey Vyshedskiy: A lot of parents will be so grateful. I have a good friend of mine. His child is now 22 or 23. I know him from being a child, and he was verbal. He was always verbal, normal child diagnosed with autism about three years of age, four years of age, and he stopped at the modifier level. So he understands combinations of nouns and adjectives, but not the syntactic phenotype, not the syntactic level. A lot of people don't realize the importance of the syntactic level. This child is really a good example of why the syntactic level is so important. Without the syntactic phenotype, you cannot explain the negative consequences of something. This child became violent around puberty. At around 13, he loaded a skillet on his mother's head, knocked her unconscious. At around 18, he was institutionalized. He still talks to his parents by phone. He's verbal. He speaks in sentences. But again, this is the modifier phenotype. You cannot explain who did what to whom to him. I spoke to a number of parents like that. Verbal children, modifier phenotype, but children are institutionalized or in halfway houses. And this is a tragedy. A lot of people would pay any money, inject any type of therapy into the brain to get back the sensitive period and teach syntactic phenotype.
[36:46] Michael Levin: Well, that's very important. I hadn't really thought of it before, but that's a whole other new set of applications to add to the consequences of these regenerative therapies. We know they will do all kinds of great things for other areas, but I didn't really appreciate the potential for reversing these kinds of things. So that's interesting. I have a couple of other questions. Are there any tools or approaches to recognize the levels, in particular, the syntactic phenotype? Let's imagine aliens. You're communicating with something that is not one-to-one mappable to our experience or our anatomy or anything like that. Are there generic tools to recognize—let's say we have a large corpus of back-and-forth. Maybe there's even a joint environment that we all live in. How do you recognize it when it doesn't look like what we're used to?
[37:53] Andrey Vyshedskiy: Excellent question. This is almost the task of archaeologists. How do archaeologists recognize that at some point a syntactic phenotype was acquired by humans?
[38:09] Michael Levin: And you know where I'm going with this, because we do the deep archaeology down to the precellular level. We're building tools here to communicate with cells and tissues, and you're going to talk to your liver about the things that the liver cares about and things like this. So we're very interested in communicating with all sorts of unconventional beings: biological, nonbiological, embodied, nonembodied, all sorts of weird things, and trying to understand—it's a two-way IQ test. When you're not understanding, maybe it's not very smart or maybe we're not very smart. So I think this is very interesting is how you identify those kinds of levels when you have communication going.
[38:53] Andrey Vyshedskiy: No. You brought up IQ tests, and this is exactly how syntactic phenotype is measured today. IQ tests, for the scores above 90 approximately, are all about combining different objects together. A circle, imagine another circle. Now move these two circles together. What mathematical sign this figure looks like. And of course, it's an infinity sign. These are the type of IQ questions that we are using to measure IQ. You can probably come up with a two-way IQ test where you facilitate this interaction and give harder and harder IQ questions, just like the modern IQ tests are working. We start with simple questions that don't require a syntactic network. And then there are modifier-level questions. Pictorially, testees have to combine color and object. And then the very complex IQ tests go into combining multiple objects together. Slowly, they increase the number of objects. So by the end, by an IQ of 130, you get three or four or five different objects moving in different directions.
[40:50] Michael Levin: The who did what to whom is very interesting because it's fundamentally applying an agentic model to the world. You're trying to identify agents and represent perspectives and interests and agency outside yourself. And this is something we're very interested in is basal kinds of life forms: to what extent do they coarse-grain their experience? Because the one thing about being a life form in a realistic environment is that you don't have time to be a Laplacian demon and track microstates. You'll be dead and eaten in no time. So, as a matter of efficiency in a world of limited metabolic resources and time, you have to coarse-grain reality and say, "Okay. All these details, well, that's an enemy or that's a mate or that's some food or whatever." This kind of thing then has the potential to turn that same modeling on yourself and say, "Wait a minute. I'm an agent that does things," and have some kind of reflexive— In those cases where the people don't reach that level, has anyone studied what is the sense of self-agency that they have? In other words, are they missing it entirely? Are they missing it about the outside world, but they know they're an agent? It's just that they're not catching it in the outside world, or is it completely gone? Is that known?
[42:24] Andrey Vyshedskiy: This is known as theory of mind, and it has been studied in autistic children for a long time. Autistic children have real difficulty with theory of mind. Theory of mind is usually thought of as stepwise: I know what you're thinking about; I know what you're thinking about what I'm thinking about, and so on and so forth. Individuals with autism usually have trouble with every step of the way; they have a problem understanding what you are thinking. But that is dissociable from the language level. If you, for example, compare individuals with autism and individuals with Down syndrome, both being at the modifier level, the autistic individuals would have very little theory of mind, but individuals with Down syndrome would have nearly normal theory of mind. In my mind, this can be explained by differential development of the medial prefrontal cortex. The prefrontal cortex consists of two parts. They are very different. The lateral prefrontal cortex is logical reasoning, and the medial part is about social and rewards and so on. This model of oneself and model of other people is encoded by the medial prefrontal cortex, this part in the middle between hemispheres. And the ability to understand who did what to whom is controlled by the lateral prefrontal cortex. So autism is clinically defined as a disease of the medial prefrontal cortex. And then as a result of the medial prefrontal cortex, you may not acquire the syntactic level, and therefore, the problem becomes a problem with the lateral prefrontal cortex.
[45:08] Michael Levin: That's also interesting. Having trouble knowing what other people are thinking, do they have trouble knowing what they are thinking? Because, of course, there's confabulation, which is often studied, but then there's the weirder aspects that we study, which is having to reconstruct content from engrams. So whatever the biophysical structures you have as a memory, it's an ongoing process to interpret them after all the particulars have been squeezed out of this generative representation to then reinflate it and know, what do my own memories mean? So on that level, having to keep a picture of yourself as a persistent agent, as most of us model ourselves, with consistent thoughts and preferences and goal states and commitments and so on—I don't know how easy or difficult it is to study that. Is there any work on how they model themselves? Do they have a better understanding of their own agency than they do of agency in the outside world?
[46:20] Andrey Vyshedskiy: It's hard to say. It is known that people with autism prefer constancy of the environment. Again, we're talking of the forty percent of autistic children or individuals who don't acquire the syntactic phenotype. There are plenty of autistic individuals who acquire the syntactic level who are very smart, high intelligence, high achievers. We are not touching that part of the spectrum. We are talking about the low part of the spectrum, sometimes called low-functioning individuals. They usually have IQ 70 or less. These are the people we are talking about. It's hard to study self-agency, but they do know what they want. The problem that they experience is that they cannot model the results of their actions. They cannot model the future. Your syntactic level allows you to imagine what will happen if I jump out of the window. They cannot do it. They cannot voluntarily imagine something that they have never seen. This is the basic problem. I call it prefrontal synthesis. The prefrontal cortex can model these scenes, these events inside the posterior cortex. Posterior cortex is the sensory place of the brain. That's the brain that perceives, and the only way to perceive anything is for that to be modeled into the back of the cortex. And the only organ that can do that is the lateral prefrontal cortex through this frontotemporal, frontoparietal network. They're really huge. Until recently, I haven't realized how huge they are. Corpus callosum is about 200,000,000 axons. The frontotemporal, frontoparietal tracts are about 100 to 200,000,000 axons. It's similar to corpus callosum in size. It's really huge.
[48:51] Michael Levin: Going back to your three levels, can you imagine what a next level would look like? Whether for us through augmentation, or maybe—I forget who wrote it, maybe Simak, but there's an old science fiction story about somebody cracking longevity and finding out that actually standard humans are just a neotenous version of some other thing. We've never lived long enough to get there. And when people started living 300, 400 years, suddenly you find out, wait, there's a next set of developmental steps. So whether that kind of stuff, or whether it's through augmentation, or just even another lineage somewhere in the universe, what's the next jump beyond the who did what to whom? What's next?
[49:49] Andrey Vyshedskiy: The jump is actually easier to understand, because what improves is the number of objects that you can combine together. At the syntactic level, you're able to combine two objects, the subject and an object in the sentence: "The cow carries the horse," "The horse carries the cow." And this is an IQ around 90. At an IQ of 110, you can mix three objects. At an IQ of 120, you can combine four objects. So that's the step. What is most interesting here is actually what you pay for it. Because as you go along this intelligence scale, you're paying by prolonging childhood. Imagine your children have doubled their childhood duration. So now you have to take them by hand to go to school until seven years of age. When you double the childhood, you now have to hold them by hand until 14. If you double that, you hold them by hand until 28. And maybe some of that is going on. How long are we paying for children now? We are paying for their college, and then we are paying for their after-college experience and medical school, and so on and so forth. So maybe some of that is going on, but we're all paying. The longer the childhood, the smarter they are usually, but the more childcare we have to invest. If you look at the evolution, childhood is very short in chimpanzees, maybe two, three years, but in humans, it is at least doubled. This is what we are paying. Again, this is a combination of our cultural changes and genetics that was driving the evolution of humans. The culture enabled prolonged childhood, and genetics responded with slow development of the prefrontal cortex.
[52:17] Michael Levin: It seems to me that advances in this area, and I think there are going to be advances as far as extending it and all of that, are gonna have interesting implications for the legal system. Because right now, we know what diminished capacity is. Somebody shows up in court, and they say, well, he couldn't have foreseen the consequences of his actions. It's not the same standard. Eventually, somebody's gonna show up with enhanced capacity, and they're gonna say, well, the rest of us couldn't have seen this coming, but you should've, with your third hemisphere and your whatever. You could have totally seen this coming down, whereas the rest of us would have been act of God, whatever. No. You should have seen this. So we're gonna have to, I think, get beyond this binary idea that you're either the standard human that features in all of the philosophy of mind books and responsibility. And we're gonna get to the spectrum where everybody's just still at different places of how much responsibility you have. I think that's gonna be really wild for our social structures to deal with.
[53:31] Andrey Vyshedskiy: Currently, the legal system is looking for the presence of the syntactic phenotype. The IQ around 70 to 80, maybe 90. Under that level, one doesn't have syntactic phenotype, doesn't have prefrontal synthesis, cannot predict the outcomes of the action. They do not know what will happen when they put a skillet on the head of their relative, that they can knock their relative unconscious. And people who reach the syntactic phenotype can predict the consequences of their action. So that division is very clear to me. But that's an excellent point. Some CEO of some big company who is very smart, with an IQ of 140, they should have predicted the outcome, and they might be held to a much higher standard.
[54:42] Michael Levin: Aside from the cost of prolonged childhood, do you think there's any limit to this per IQ process? Does it flatten out at some point, or is there a top of the curve? What does it look like?
[55:01] Andrey Vyshedskiy: Good question. I'm not sure. I think childhood can be prolonged again and again. I don't think there is any upper limit. The upper limit is puberty, but puberty can also be delayed. So childhood can be delayed at least until puberty. Maybe these are two different processes. At least I'm thinking about them as two different processes.
[55:36] Michael Levin: Axolotls figured it out. They are a newt-like thing that isn't like that anymore. But natively, they live their whole life that way, and they reproduce and everything. So I don't know. Maybe there's hope that we can keep both.
[56:08] Andrey Vyshedskiy: Maybe. But at least, evolutionarily, how it worked is that those processes were probably influenced by different factors. Puberty developed driven by a set of circumstances separate from the duration of childhood. This is not a single gene that regulates the duration of childhood. These are tens or maybe hundreds of different genes that together regulate the development of the prefrontal cortex, and a few of those we know for sure. In that paper, "How One Tribe Conquered the World 70,000 Years Ago," I have a graph of probably 20 different genes: genes that regulate neurogenesis, formation of new neurons; genes regulating many other things like FOXP2, for example; and genes that directly regulate the time course of brain development. Usually in animals, the goal is to develop as fast as possible, like a horse. A horse is born, it can walk a minute after birth. Humans are born and they cannot crawl for another year. In most animals, the goal was to reduce childhood as much as possible. In primates and especially in humans, the goal reversed. The goal was to prolong childhood to enable more training of the frontoposterior and frontoparietal networks. We see the effect of this very clearly. You take a child: the child who grows in Japan would get tuned to Japanese phonemes, and the same child who grows in the United States would get tuned to English phonemes. English people have R and L very clearly distinguished, and Japanese cannot distinguish English R and L easily because they have different fine-tuning. These are all the effect of prolonged childhood. These are the effect of the brain networks being able to modify by experience.