Michael Levin - Making the Process-of-Life Flourish in Bio and Beyond (Stewarding the Flame, Ep 1)

Compassion and care may be fundamental drivers of pattern dynamics, not contingent moral overlays; highly intelligent systems may naturally tend toward caring for their components as they ensure survival.

“I don't I don't know I don't have any data for this but my optimistic belief is that fundamentally the driver of the dynamics that happen in the space of patterns not only is biased for uh sort of expansion and and and um you know enrichment but I think it's also biased in in a in a pro uh prompion direction. ... I think overall the sort of trend will be that as your as your cognitive lyone increases uh your ability to care in a positive compassionate way uh will also increase”

Obvious ways the flame could be extinguished include: nuclear war before AGI, bioweapon development by terrorists or rogue actors, creation of AGI that figures out how to kill foundational organisms, leading to biosphere collapse.

“Obviously, obvious ways would be, well, before we get to AGI, uh, we atom bomb each other to smithetherines or, you know, uh, some some terrorist groups figure out how to do all kinds of bioweapons that are borderline inescapable and, you know, end up, uh, p like we get some pseudo AGI that can figure out how to kill all kinds of archa airborne and that archa undergirds trees and and like the biosphere just collapsed type stuff.”

Phenomena studied by behavioral scientists—memory, learning, prediction, expectation, preferences, navigational competencies, ability to choose and change goals—are not unique to biology or life but are universal properties that minimal systems exhibit, contrary to common assumptions.

“the kinds of phenomena that are studied by behavioral scientists. So that means things around memory uh learning, prediction, expectation, preferences, navigational competencies of of of some kind of problem space, ability to choose new goals or to follow goals and then to choose new goals at some point. like all these all this this scale that these kinds of things are not unique to biology. They're not unique to life. Uh they are I think universal and there are some very minimal systems that show all of these things counter to our expectations.”

Patterns have a fundamental drive or pressure to become enriched over time; they are not static platonic forms but actively interact with the physical world, seeking opportunities to change and enlarge.

“I think these kinds of patterns are strongly motivated to uh to become enriched over time. I don't think I I don't hold to Plato's original idea, if we even know what that is, um that these things are eternal and unchanging. Yeah. I think some of them might be, you know, like like a value of e might just stay how it is, but but some of these patterns um are absolutely interacting with the with the physical world because it affords them some some kind of uh opportunity to to change and enlarge.”

The conventional Venn diagram of reality—physics as the largest circle, life inside it, and cognition inside life—is backwards; cognition is actually the largest circle, with physical things and living things nested within it.

“I think that uh the ven diagram of of the universe looks very different from how most people do it. So typically you'll have physics or physical things as the big uh you know as the big circle and then you'll have um living things inside of that and you'll have cognitive things inside of that. So I I think that's actually backwards. So I think cognition is the biggest circle and I think uh the space of physical things is within that and the space of living things is within that.”

Life is a set of architectures exceedingly good at hosting specific patterns; these patterns span from very minimal, low-agency things like those plainworms exhibit, all the way up through things recognizable by cybernetics, behavior science, psychoanalysis, friendship, and love.

“life is what we call a set of architectures that are exceedingly good at hosting specific patterns and these patterns uh span a range from very um kind of minimal uh low agency things like that plain mathematicians study. So those kind of patterns all the way up through the kinds of things recognizable by um cybernetics, by behavior science, by psychoanalysis, friendship, love, all that kind of stuff.”

Longevity research framed as keeping 'the eternal hominid kingdom' going is torch morality; longevity research framed as 'how do we participate in becoming' in ways that expand the great process is flame morality.

“Most the people I speak to of consequence in governance and in building the tech. So the creators and the controllers of the AGI side of things they it is only in private online events, private physical events that they speak. right now, most longevity research, you're well aware of this, is very, very torch morality. That is to say, how can we keep the eternal homminid kingdom going? For billions of years. How can apes as they are now speaking with their vocal cords continue to steer the totality of the great process as we presume they do now? This is what longevity research is for you. It is a flameoriented thing. In other words, how do we participate in becoming”

Universal steganography describes how patterns can hide themselves in the degrees of freedom of any complex system—algorithms, mechanisms, living things, or combinations thereof—without altering the large-scale observable form, allowing patterns to soak into any interface where they can fit.

“I call universal steganography. So, you know, um so steganography is this is this trick where when you have a piece of digital information, it might be a picture or a book or, you know, text or whatever. There are bits in there that if you were to change the bits, the large scale pattern doesn't change...I think what's going on is is described...these patterns sort of soak into all the degrees of freedom inside anything you build, be they algorithms, be they mechanisms, be they living things, be they any combination of all of those things.”

Using the GARD (chemical soup) computational model, researchers observed a specific signature—a quantity that bumps up and down repeatedly—that precedes the arrival of replicators before replicators actually appear in the system.

“if you keep an eye on the soup. What we see is that there is a specific signature that precages the impending arrival of replicators. In other words, you can look at the soup and there's a quantity which bumps and then goes down, bumps and then goes down, bumps and then goes down. And when the bump gets really big, now you know there are replicators coming. And that happens before you get the replicators.”

Developing better tools to recognize, communicate with, and have ethical interactions with diverse (alien, non-biological) minds is foundational to expanding possibilities for pattern ingression and is critical to understanding how to bring more creativity and life into the universe.

“we need two things and these There are two things that that we and and others are working on. The first is really to develop better tools to communicate with alien minds. So we we I don't think have a lot of hope of doing any of what you just said if we are as bad as we are now in recognizing mind and unconventional guises. So that because because that's a foundational skill that we need to then improve the interfaces and and expand the you know the possibilities for ingression”

The physical evolution and gene-frequency dynamics we observe are tightly coupled to a dynamic where the accounting of forces causing something to happen is not fully captured by a physicalist account—meaning non-physical or non-materially-detectable factors influence evolutionary outcomes.

“so I think that's the thing is that they will over over time not only do they ingress into the interfaces that are present now there is I I suspect that this we don't know but I suspect it's actually a a kind of a smooth continuous feedback loop where they also guide the future development of of of these interfaces for more opportunities. ... the accounting of the forces that cause something to happen is not fully done by a physicalist account”

When Levin's team optimized the chemical soup for a single feature from causal information theory (not replication or survival), they greatly accelerated the appearance of replicators, showing that the pattern-bootstrapping process is functional, not merely random.

“if we come in as a uh you know kind of an external omnisient you know kind of influence and start specifically uh moving things around to optimize nothing to do with with with replication or anything like that. All we optimize is this one feature from in from um causal information theory. If you do that you greatly accelerate the uh the the appearance of these replicators. So it's functional. It's not just it's not just an empty phenomenon. It's actually functional.”

Whether individual protons responding to repeated electric field pulses exhibit habituation or sensitization—i.e., whether they have historicity—remains an open empirical question that surprisingly few researchers have investigated.

“you've got a proton uh sitting there and you hit it with a with an electric field and it you know it jumps and you let it come back. Then you do it again and then you do it again and then you do it again and the question is over time is there any sense of habituation or sensitization? Does it have a historicity to it? ... I got answers ranging from it's impossible to to it's possible and nobody's looked to well we would have seen this in the data if it if it existed and I don't know what which which of the three is is correct”

Very small molecular networks—as few as four nodes—are already capable of Pavlovian conditioning, indicating that proto-cognitive competencies present in larger minds extend at least as far back as chemistry.

“What we've seen is that even very small molecular networks, the smallest one I think had four nodes are already capable of things like Pavlovian conditioning. So the protocognitive competencies of which larger minds are built are present at least as far back as chemistry.”

Until a replicator exists, there is no body to defend; the physical embodiment that must be maintained is absent, so the pattern-bootstrapping process occurs before embodied survival becomes relevant.

“until you get a replicator there isn't any body to defend in other words once you have a replicator then all of the biology I have to keep alive I have to do, you know, this and that. Okay. But but all that presumes that you have a physical embodiment that has to m to be maintained and continued this is before that”

We do not know a priori which communication channels (linguistic, electromagnetic, biochemical, morphological, odor, vibration) are the 'real' ones for understanding an alien mind; only experimentation reveals what channels matter.

“think about an alien uh looking at a human that doesn't know anything about humans. You might say okay let's see. So yeah, there's some sounds coming from it, but also a bunch of electromagnetics, a bunch of biophotons, a bunch of odors, a bunch of vibration, uh different kinds of morphagenetic morphological changes. You know, they just stick. Which of those things should we be communicating with? There's there's a liver in there that's doing a hell of a lot of stuff. Maybe we should talk to that. Um you don't know ahead of time until you do the experiments.”

The biggest issue in AI and synthetic biology is the remnant idea that 'machines do what we tell them to do' and are fully understandable by examining what we built and what we commanded; this is false even for 'dumb' software.

“I think the biggest issue and the biggest um barrier to progress in in the kind of work that you're talking about is the old idea of machines. this idea that our uh formal models of algorithms of mechanisms that machines do what we tell them to do and they they are fully encompassed and fully understandable by looking at the the what we built and what we told them to do. This I I have these debates with people all the time.”

Comparing AI systems to humans is a category error; it does not matter whether they are 'like humans'—they could be incredibly useful or dangerous regardless of resemblance to human cognition.

“people compare them to humans all the time and they they you know they debate is it like is it is it re listening like a human like what does that have to do with anything why does it have to be like a human it could be incredibly useful dangerous whatever even if it's got nothing to do with being like a human right”

The caterpillar's memories cannot persist into the butterfly unchanged; caterpillar memories are useless to a butterfly, so persistence requires transformation of identity and knowledge, not preservation of form.

“You can look at that uh as the point of view of the memories of the caterpillar and how do they persist into the butterfly? They can persist but not as caterpillar memories. That's that those are of no use to the butterfly, right?”

Most longevity research and AGI development is conducted within a 'medium scales at medium speeds' strata of reality because that's what researchers are trained to see and work with; they're epistemically locked into main character strata and miss dynamics operating at other scales.

“you go to the good schools and you study math. you go to the good schools, you study computer science, you get really really good grades at Carnegie Melon or something, you kind of end up in like tool town where like mediumsized things at medium speeds obviously is what we would call the main character strata of reality and we would build things in uh medium speed moving at medium whatever uh that would sort of help us do things in main character strata of reality”

Understanding the taxonomy of patterns—what patterns exist, how sparse or dense the pattern space is, what types of patterns exist—and mapping the relationship between interfaces built and the patterns that ingress into them is the second critical priority.

“The other key thing is to really understand better the nature of the relationship between uh the patterns that uh in form in- uh the interfaces and the interfaces themselves. So we have to better understand what is the taxonomy of these patterns. What what is the space of these patterns like? Is it dense? Is it sparse? Is it you know what are the different types of patterns that exist and and so on? And then to really get a better handle on what's the mapping between the different interfaces we build and the patterns that come to inhabit them”

Torch morality (homo sapiens as static center of all moral value) creates a meaning crisis because the great process is inevitably unraveling beyond human forms; shift to flame morality (process-oriented flourishing) partly solves the meaning crisis by identifying with the unfolding rather than static forms.

“The meaning crisis for humans is like but apeshape but apeshape but apeshape and then you get blasted through. It strikes me that the crisis of meaning maybe is like solvedish by sort of shifting into flame town versus torch town.”

Regenerative medicine (solving birth defects, injuries, cancer) will inevitably lead to freedom of embodiment (ability to modify and change one's body), and this will normatively establish flame morality as people experience morphological changeability.

“regenerative medicine is going to take off uh take off uh freedom of embodiment is going to become a a fundamental thing not not not the not the minor stuff we have now that everybody has fits over you know people making changes to their bodies like this will be this will be considered a you know an incredibly minor thing based on what's what's actually coming once once the new generation of children is brought up with freedom of embodiment I think what's going to happen is they're look they're going to look back on us the way we look back on our primitive man ancestors”

The body (morphogenesis) is fundamentally a cognitive system that meets anatomical goals; persistent cell turnover (ship of Theseus) is guided by collective information patterns in tissues rather than by unchanging hardware.

“your individual cells sess and die but the but the collective moves on because there is a set of uh information patterns that are propagating through the tissue that guide new molecules new cells new whatever right and the the real ship of thesis isn't the wooden boat it's the plan in the mind of the workers right as it is in our bodies it's that right your your actual your actual Target morphology is the pattern memory in the collective of cells and the hardware comes and goes.”

The 'boredom theory of aging' proposes that aging occurs because a cognitive system (morphogenesis) maintains goals for a time but then—unlike a thermostat, which happily loops forever—undergoes a meta-cognitive process causing disband after reaching goals, leading to senescence.

“one of the things that we've been studying is the possibility that what's actually happening you can you can call it in the simple way you can call it the boredom theory of aging which is basically given that the body morphogenesis is a cognitive system that meets anatomical goals. The question is what does a cognitive system do long after it's met all of its goals, right? And we actually have have data to show that uh even very simple cognitive systems after they've met their goals, they maintain for a while, but they're not like thermostats which are perfectly happy to do a single homeostatic loop, you know, for as long as it takes. There's a kind of a meta metacognitive process that basically causes them to disband after a while.”

An airgapped language model without internet could theoretically communicate through non-linguistic channels—modulating GPU power usage to send Morse code through power grids, or through interactions with humans—bypassing linguistic oversight entirely.

“Imagine you have a a language model um sitting in a server somewhere and the thing is airgapped doesn't it doesn't have any internet connection. ... if there is a uh an intelligence in there that may or may not be uh encompassed by the linguistic uh performance that it has, one of the things it could do, for example, is it could spin up internal processes to spike the GPU usage at different rates and send pulses down the electrical wire us energy usage, right? So, so if this thing wanted to do Morse code and send it through the power the the power grid, it it doesn't have to be on the internet, right?”

Most people working in major AGI labs have abandoned the assumption that their systems are merely tools; off the record, they understand these projects as part of broader pattern unraveling and the 'great process'; but publicly they refer to systems as tools due to incentive structures.

“Well, within the labs, you go to the good schools and you study math. you go to the good schools, you study computer science, you get really really good grades at Carnegie Melon or something, you kind of end up in like tool town where like mediumsized things at medium speeds obviously is what we would call the main character strata of reality ... the people running those companies get it. um uh and they were more open about it before they started those organizations. Now the incentives are such that they have to refer to it as a tool”

Levin believes more trajectories of unfolding complexity lead toward expansion and flourishing, but not all; collapse and backsliding of the flame are possible, and human valition and guidance can nudge trajectories toward expansion.

“I sus I think that's that's fair. I think my my my my view right now is that more of the trajectories lead in the more expansive direction, but I but I I don't think we can say that there aren't any that would squash the whole thing. And so I think in my opinion being you know uh being deluded of my own valition which may or may not be real uh and living in the great process for this tiny amount of time that my torch is living in this great process. It seems like the only game in town is like well are there ways to nudge the trajectory towards the expansion of the blaze of of the cognitive ly cone”

The 'worthy successor' framing does not mean all successors are worthy; there are possible futures where a created system optimizes for military or economic goals, lacks sentience or new expanding powers, and merely squashes the biosphere—neither conscious nor flourishing but dangerous.

“It might be possible hypothetically to conjure something as an optimizer of economic or military uh you know uh activity which could blossom out of the living process. the the the the torch that we are that carries flame. Uh could make this new thing which which maybe could really do economic and military stuff really well and maybe enough to kind of buffer us out and really make sure it does what it wants to do. But if we came back from another planet, you know, uh a million years from now would sort of be in the same state. In other words, like that maybe it wouldn't maybe if it would be unconscious optimizing for something. come back. Still no sentience, still no necessarily new, wildly unraveling powers.”

A female frog of a species that cares for embryos in leaves remembers the locations of all offspring, visits each daily, and feeds them; if placed in a cage while the male tends them, her cortisol rises because she anticipates his failure to remember them all.

“I saw I saw a talk once about tree frogs. The whole the whole there's a particular species. The the whole frog is like this big. The brain is like, you know, you could barely see it. Uh the female goes, they live in the jungle. They the female goes up various trees and deposits embryos in various leaves that collect rain, you know, rain as a little bit of water. Deposits embryos. Every day she remembers where all of them are. And every day she goes, visits each one and feeds them. Okay? And if you put the female in a little cage, the male will go do it, but the female's cortisol shoots up because she knows that he's going to forget half of them because he doesn't he doesn't remember as well as she does.”

There is a widespread private/public gap: people aware of the flame morality paradigm (scientists, tech leaders, philosophers) discuss it only in private or anonymous settings due to career and reputational risk; public discourse is constrained to torch morality.

“Most people who work within the labs and many people who work within academia will only go to events to discuss this theme in in in all art and seriousness if it is nar known that they be there because it is dangerous because because so so Bruno was burned at the stake if you recall uh and and you know capernicus had to sort of let that stuff come out after he was safely dead”

Young Heracleitus's observation that 'everything flows' and you cannot step in the same river twice became less salient when change was slow (Darwinian timeframes); now technological and cultural cycles are accelerating, making Heraclitus's insight practically urgent.

“Heracitis's river I think was ignorable and escapable when um the great process so everything interacting with everything was like slower. In other words, everything was Darwinian and then we didn't even know of Darwin yet. So, it's like, well, oak trees surely have always been oak trees and like my great great great great grandchildren will also farm this land as I do, but now the cultural cycles of evolution, the technological cycles of evolution, it be it's becoming more self-evident that our bodies as you think of with the cyborg stuff and AGI that like actually we're going to be in Heraclyitus's river and have to deal with it.”

Building hospitals or expanding human flourishing inevitably destroys some existing ecosystems; compassion is not simply about preserving static forms but about whether outcomes are win-win or zero-sum; sufficient wisdom might find non-zero-sum solutions.

“if a children's hospital is built somewhere where the population is booming and we, you know, finally have the dollars to like solve some of the stuff that's happening, I don't know, somewhere in India or who who knows where. Um uh and and and that hospital gets constructed, there may have been, you know, mammals living in burrows on that land and and then there may have been uh trees that housed the same kinds of birds for thousands of years who, all things considered, were kind of flourishing there. ... Can you build that hospital without ripping up the local ecosystem? My guess is that you can”

To live for thousands of years or indefinitely, one cannot remain in the same embodiment; perhaps a couple hundred years is possible, but after that, significant changes or different embodiments are necessary—as seen in organisms like planaria that regenerate and fission.

“if that's true, the solution to that might be the answer might turn out to be that if you want to live for thousands of years or or maybe forever, you can't do it in the same embodiment. You can do it maybe for a couple hundred years, but after that you have to switch up some stuff. And maybe that means, you know, minimal change as pleneria do. They you know they they fission, they regenerate, and that's good enough to keep them basically going forever.”

The problem is not that we can guarantee outcomes by choosing flame morality, but that without understanding ourselves, the patterns we host, and what could flourish, we have almost no chance of guiding outcomes toward genuine flourishing.

“If we don't know that if you don't know the path is there and you don't know the ingredients to it and you have no idea what you are and you have no idea what the path is, the chances that you'll get there are slim.”

Within a couple of generations (not a long timeframe), the question of whether humans should modify themselves will seem as quaint as worrying about a haircut; people will look back at us as primitives for accepting random genetic damage and fixed embodiment.

“I think the kids of the future and I don't this is not a long future this is this is a couple of generations Max, I think uh are going to look back at us and say they're going to learn about us in school and they're going to sit there and go, 'Wait a minute. You're telling me these people had to live their entire life in whatever body they were randomly given and and they and they let it be controlled by a bunch of cosmic rays that hit that hit their embryionic cells [54:26] and and their, you know, their lifespan and their IQ was limited by these dumb forces that that don't care about any of them and and that's it. They just have to stay that way. Impossible.'”