Where Minds Come From: The Scaling of Collective Intelligence, AI, and You | Michael Levin Lecture

Salamander kidney tubules scale dynamically to cell size: normally 8-10 cells form a lumen together; when cell size is increased (via polyploidy), fewer cells form the same lumen; when cells are extremely large (5n or 6n polyploid), a single cell bends around itself to create the same structure—this shows cells switch between different molecular mechanisms (cell-cell communication vs. cytoskeletal bending) to achieve the same high-level anatomical goal, which is a valid definition of intelligence.

“normally there's about 8 to 10 of these that eight to 10 of these cells that work together and they form this little this little structure with a space in the middle...when that happens you still get a normal animal and it's normal size but the cells are bigger and when you when you take a cross-section you find out that well the cells are bigger and so now fewer of them created the same kind of lumen”

The caterpillar-to-butterfly metamorphosis involves radical dissolution and restructuring of the brain and body: most neural connections are broken, cells are killed, and the brain is rebuilt from scratch, yet memories learned as a caterpillar persist through this process and are remapped from caterpillar-relevant to butterfly-relevant concepts (e.g., from hunting specific leaf colors to seeking nectar).

“the brain basically in during this metamorphosis process when the caterpillar turns into the butterfly what happens is that U the brain is pretty much dissolved so so most of the connections are broken most of the cells are killed off huge refactoring remodeling new cells new brain is built quite different uh and there's your butterfly”

Planarian flatworms can regenerate entirely new heads after decapitation and still remember information learned before the head was removed, proving that memories can be stored outside the brain and imprinted onto newly developing neural tissue as it grows.

“when you train them to expect food on these little bumpy circles they uh will remember and if you cut off their heads with their brain the tail will sit there doing nothing eventually the tail will regrow a brand new head so they regenerate every every piece will regenerate and eventually when they regenerate you find that they still remember the original information”

Tadpoles with scrambled organ positions ('Picasso frogs') still achieve correct frog face development because cells use error-minimization—organs move in novel paths until reaching the correct anatomical target location, then stop; this is not hardwired movement but goal-directed problem-solving.

“what we did was we created these so-called Picasso frogs because we wanted to test how how much intelligence is actually in that process how much what kind of problems can it solve so we scrambled everything um you can see here the eyes on the top of the head the mouth is off to the side everything is just a complete mess and these guys still become normal frogs because all of these organs will move now in novel paths paths that these guys don't do uh until they get to a correct frog face and then they stop”

Planarian flatworms can be reprogrammed to have two heads by modifying their bioelectric pattern; these two-headed worms remain permanently two-headed through unlimited regeneration cycles in plain water (with no further manipulation), and their DNA remains unchanged—proving that pattern memory for anatomy is stored in bioelectric state, not DNA, and is stable and heritable across regeneration.

“when you modif modify that pattern you make two-headed worms and when you make these two-headed worms uh what happens is that you have Rewritten the memory of what a correct worm is supposed to look like so that when these fragments when these middle fragments go to regenerate they consult the pattern to say what what should a normal Worm look like they see that it says two heads and they go ahead and they build two heads these two-headed worms now in plain water meaning no more manipulation of any kind um are now forever two-headed if you keep cutting them they will just continue to regenerate as two-headed”

Xenobots are constructed from frog embryonic skin cells that self-organize into novel configurations; they exhibit spontaneous movement (propelled by cilia), collective behavior patterns (dancing, patrolling), calcium signaling, and goal-directed navigation, despite being only skin cells and having no evolutionary history of independent life or selection for xenobot traits.

“remember this is just skin this is uh this is these these are just EP epithelial cells and uh the they have they have lots of interesting uh all lots of interesting behaviors”

Xenobots can self-replicate kinematically: they collect loose skin cells into balls, and because cells are agential material, the collected cells self-organize into new xenobots (second and subsequent generations) that repeat the process, creating a novel form of non-genetic reproduction not seen in nature.

“in this novel configuration they figured out a way to make copies of themselves and here's how they do it if you provide them with loose skin cells this loose material what they do is they run around and they com and they sort of collect those loose skin cells into little balls like this and because what they're working with is an agential material meaning these are not passive pellets these are cells for the same reason that we were able to make xenobots they make the second generation of xenobots and these little balls that they make mature and guess what they do they run around and make the next generation of Cabots”

Anthrobots are adult human tracheal cells (from donors in their 70s-80s) that reprogram when liberated from their normal context, developing novel morphologies and behaviors; they can repair neural wounds by knitting together the two sides of a severed nerve tract without genetic modification or immune suppression (being autologous).

“here's this little guy running around and I will ask you uh what this is and you might think that this is something I got off the bottom of a pond somewhere right some sort of primitive organism and then I can tell you that well we've sequenced its genome and you'll say oh what's the what's the genome what does it look like and they say well this is 100% Homo sapiens these are adult human cells there's no embryonics here um this is this is adult uh tracheal cells taken from uh human donors most of them elderly in their 70s and 80s”

We have a long history of wrongly denying moral status to beings based on superficial characteristics (appearance, composition); we must be extremely careful not to repeat this pattern by denying moral worth to novel beings (cyborgs, synthetic life, AIs) based on whether they are 'metallic' or not 'soft and woolly'—the old tests of natural vs. artificial will not work for future beings.

“humans have a very long history of uh trying to make uh separations between ingroup and outgroup based on really pretty ridiculous um uh kind of uh very uh surfacy sorts of um uh characteristics”

Humans harbor historical prejudice against recognizing non-human minds and intelligence; we tend to deny moral status to beings that do not match our template, and this error has been repeated across cultures when applied to human out-groups, making it likely we will repeat it with synthetic and engineered minds if we are not careful.

“we need to be extremely careful about not denying uh uh Moral Moral worth and and and various um kinds of ethical protections to beings because they don't look like us you know humans have a very long history of uh trying to make uh separations between ingroup and outgroup based on really pretty ridiculous um uh kind of uh very uh surfacy sorts of um uh characteristics”

Cancer arises when cells electrically disconnect from the larger network; this causes them to shrink their cognitive horizon from the scale of the organ/organism down to the individual cell, so they behave like amoebas in an environment rather than as cooperating parts of a whole, leading to migration and uncontrolled replication.

“the first thing that happens with this with this anene is that it causes an electrical disconnection uh from cells from their um from their environment and when cells disconnect they immediately lose the ability to remember these large uh grandio goals that they were working on so making you know complex organs and fixing defects and so on uh they're just IM now you know that boundary between self and world has now shrunk that cognitive lyone where before the goal was the size of a of an organ or or maybe the whole organism now it's down to the size of a single cell and to them the rest of the body is just environment now okay so this is metastas this is where you get metastasis”

The space of possible anatomical forms that cells can build is far larger than what has been selected by evolution; a wasp can reprogram an oak leaf to build a gall structure without altering the leaf's DNA, by providing signals that cause leaf cells to adopt a different anatomical set point, proving that morphogenetic competence is not constrained by genetics but by the prompts/language we use to communicate with cells.

“Along Comes A bioengineer and this is a non-human bioengineer this is a little wasp and what these wasps do is they provide some signals that prompt uh uh the the rest of the um the rest of the tissue to build something like this okay they're they're hacking the morphogenetic competencies of this Leaf they're not changing the genome uh and uh what they're doing is providing some instructions that cause the leaf cells to adopt a different uh set point in anatomical space”

Cells are made of agential material—material with its own agendas and problem-solving capacities—unlike passive engineering materials like Legos, wood, or metal that simply stay where you put them.

“we are made of a material with agendas and so this is the kind of thing we're made of this is a single cell okay uh this is this happens to be a free living organism called the lacr Maria uh you can see that it's extremely competent in its local unicellular goals so there's no brain there's no nervous system there are no stem cells but this thing is doing every everything it needs to maintain its physiological its U metabolic and behavioral goals”

The genome specifies proteins and molecular hardware, not anatomical structure; the genome does not directly encode where organs are, their shape, or size—these are outcomes of cellular physiology and behavior, analogous to how termite genomes do not encode termite nests or spider genomes do not encode spider webs.

“of course we can read genomes now and we know that that isn't what the genome specifies at all the genome talks about proteins the genome specifies the the tiny molecular Hardware that every cell gets to have this isn't directly in the genome any more than the structure of a termite Nest or the shape of a spiderweb is in the Genome of those species all of this is the outcome of physiology and behavior”

The electrical activity and bioelectric patterns in frog embryos (detected via voltage-sensitive fluorescent dye imaging) map the collective intelligence's goals before organs are built: the 'electric face' shows where eyes, mouth, and other structures will be located, revealing the intentional structure the cells are working toward.

“this is a time-lapse of a frog putting its face together you can get a map of all the electrical States and thus you can learn you can try to learn to decode what the system is trying to build what what are the uh the computations that they're doing this this we call this the electric face this is a a bioelectric um map of of of the uh of the face ectoderm long before any of these organs actually appear”

Injecting an ion channel modulator RNA that restores electrical connectivity prevents cancer progression, even when the genetic mutation (enogenic transformation) is still expressed; this shows that restoring large-scale bioelectric coherence can override genetic defects and restore normal developmental behavior.

“we co-injected an ion channel that modulates uh it it's it's an RNA that modulates the electrical property of the cell to force them to remain in connection despite what the enene is saying and after that it doesn't matter that they that they have this genetic defect uh they behave appropriately and they build nice organs and so on”

Injecting potassium ion channel RNA into the gut region of a tadpole changes the bioelectric pattern at that location to match the eye pattern, causing the cells to build an ectopic eye (with lens, retina, and optic nerve) in the gut—demonstrating that bioelectric signals alone, without genetic modification, can reprogram cell behavior to build novel structures.

“we injected some potassium ion Channel RNA into a location that normally would make a gut so here's this this is a tatp here's the mouth here's the eye here's the gut um and and if you tell these cells using an appropriate voltage signal that hey you should be making an eye boom that's what they do and these eyes have the normal lens Reda optic Nerf all the same stuff”

The notion that species are discrete natural kinds—each separate, numbered, and categorizable—must be abandoned because evolution and developmental biology show continuous gradients from single cells to complex organisms across both evolutionary and developmental timescales, with no bright line marking where 'just physics' becomes mind.

“this is not a sharp category in other words if you just track backwards in time you uh the these modern humans are at the uh at a at a particular point in the evolution of a lineage that goes all the way back to single cell organisms”

The ethics of AI is not about today's language models (which are not like human minds) but about future beings that will be cyborgs, synthetically engineered life, hybrid material-software systems, and truly alien intelligences that do not fit into traditional categories of 'natural' or 'artificial,' and we must develop ethical frameworks that do not rely on material composition or origin story.

“it isn't uh the question of AI isn't about today's language models you know today it's pretty easy to say well gp4 isn't really like like a human mind and and so on no no it isn't and it doesn't matter for two reasons first because there are many interesting Minds that are not human minds and some of them are are quite quite alien and second because uh the whole point is that you you can't just assume these things you have to do experiments”

Humans are very poor at detecting intelligence outside medium-sized objects moving at medium speeds in 3D space; complex organisms navigate and solve problems in spaces we find difficult to perceive (gene expression space, physiological state space, anatomical morphospace) and we lack the intuition to recognize this as intelligence.

“we humans are pretty terrible at detecting uh intelligence I mean we do okay with middles sized objects moving at medium speeds in three-dimensional space you know crows and birds and and and and maybe a dolphin or something like that you know we can we can kind of um uh we can kind of recognize those as intelligence but there are other spaces in which uh complex creatures navigate and uh and uh spend effort and try to achieve goals and and and so on uh there are gene expression spaces and physiological spaces”

A fly exhibits an ant-morphogenetic pattern on its wings to deter predators, simulating ants moving across its surface; this pattern is carried out by the same morphogenetic hardware as other patterns and demonstrates the flexibility and plasticity of how cells use their genetic toolkit.

“this fly is running a strip down uh ant morphogenetic program on its wings it's running a virtual ant on its on its wings and and the idea is that it it Wiggles uh it Wiggles its um its wings to simulate these ants scurrying about and that keeps the um Predators away”

We need to avoid two opposite extremes in relating to diverse minds: objectophilia (falling in love with inanimate objects like bridges) at one end, and the opposite extreme of denying consciousness and moral worth to any being not matching a narrow template, which quickly becomes 'love only your own kind' and is likely more harmful.

“we're going to have to figure out how you uh how you detect where things are on this Continuum so that we find appropriate balance points between objectophilia objectophilia is the thing where people fall in love with Bridges and you know they think the Eiffel Tower is is you know married to them and so on uh right so so things down here you're not going to have that deep relationship with things up here but other people are really trapped in the other end of the Continuum where they see things that are not like today's standard humans and they say uh you know those are not those are not real Minds worthy of of of of care and respect and so on and that very quickly devolves to a kind of idea that's basically love only your own kind and and I think that's that's that's probably Lots much worse than this”

A rat learning to associate lever-pressing with food reward shows that ownership of memory is collective, not individual: no single cell experiences both the lever press and the food reward, yet the rat as a whole 'knows' these are connected; the nervous system is cognitive glue that integrates information across cells that do not individually share experience.

“there are cells in this rat that interact with the lever there are cells in this gut that get the sugar but there is no single cell that has both experiences so who is it that owns this in this um associative memory the that the pellet is associated with pressing the lever who owns that well the owner of that memory is not any individual cell it's the entire rat it has to be because no cell has that experience”

Alan Turing published a paper on chemical morphogenesis (self-organization of embryos from chemicals) in addition to his work on computation and machine intelligence, suggesting he recognized a profound connection between body self-assembly and mind origin, although he did not explicitly write about this connection.

“Alan toring who is basically the the father of modern computer science um he uh was was very interested in in machine intelligence and in uh computation and uh and and programmability things like this but what and so so everybody knows knows that part but one weird thing is that he also published this paper called the chemical basis of morphogenesis which was basically an attempt to understand how the chemicals of an embryo organize into order from the from their initial disorder”

Individual humans are not true unified intelligences but rather collective intelligences made of parts with their own agendas; the pineal gland is not a unified seat of consciousness, and the brain contains trillions of cells, each with its own competencies; there is no single cell in a brain that experiences the unified self we feel.

“at least we are a true unified intelligence right so so maybe you can call an ant colony some kind of collective intelligence but but we have a we have a brain where you know we we feel like a single unified being words so at least we are a true unified intelligence and in fact um uh Renee dekart uh really liked the pineal gland because there's only one of them in the brain and he thought that that's where uh the the The Human Experience was was centralized”

Cells possess competency at multiple scales—molecular networks, organs, tissues, and groups—and at each scale they demonstrate the ability to solve problems in their respective problem spaces (gene expression space, physiological state space, anatomical morphospace); these competencies are not merely the output of genetic instructions but genuine problem-solving capacities.

“we have this amazing multiscale competency architecture at every uh scale in your body from from the molecular networks uh to the um organs and tissues and and even groups these different layers are not just uh biological scales but but actually they're they're they have the ability to solve problems in other words they have competencies”

Xenobots have no evolutionary history of being independent organisms and no selection pressure to be xenobots; their behaviors, shapes, and kinematic self-replication are not explained by Darwinian selection—suggesting evolution creates general-purpose problem-solving agents whose capabilities extend far beyond the specific niches they were selected for.

“there's never been any xenobots there's never been any selection to be a good zenbot this is not in the lineage of uh of frogs um where does this come from their shape their behaviors the kinematic self-replication all the other things they can do where where does that come from uh it's very hard to say that that that this was the the the subject of selection”

Embryonic fields can fragment or recombine to form different numbers of coherent individuals; the number of 'selves' in an embryo is not fixed by genetics but emerges in real time as a process of physiology and can range from zero to more than half a dozen depending on developmental context.

“the question of how many beings how many selves are actually in an embryo is tricky it's not fixed by the genetics it's anywhere from zero to probably half a dozen or more depending on what happens the the selves arise in this medium this the cellular blastoderm uh in real time as a as a process of physiology they are not it's not hardcoded”

Morphogenesis—the creation of complex anatomical shapes—is behavior of a collective intelligence operating in anatomical space (a multi-dimensional space of possible forms), analogous to how a brain operates in behavioral space; cells are solving problems in this anatomical space using goal-directed, problem-solving logic.

“if morphogenesis is the behavior of a collective intelligence operating in anatomical space that's what I've been um getting at for you know for for for the last few slides is the idea that that the the the the act of creating complex anatomical shapes is really Behavior it's behavior in a in a an anatomical space of possibilities and it's done by a collective intelligence uh we're just like us we're all Collective intelligences but that collective intelligence thinks about shape as opposed to moving around in 3D space”