What Is The Field Of Diverse Intelligence (DI)? All Possible Intelligent Agents | Michael Levin

Phase transitions in biological systems are scale-relative and observational artifacts: they appear sharp at one scale but become smooth when examined more closely, and their appearance depends on the formalism (binary vs. fuzzy logic) used to describe them, making them 'in the eye of the beholder' rather than fundamental features.

“They appear depending on what formalism you're using to understand them, right? If you're using binary logic, then you'll get some and if you're using fuzzy logic, you may not. And they depend on what the time scale that you're using. At one time scale, it looks super sharp, but another time scale, it's like this long tedious gradual process.”

Young researchers entering the field of diverse intelligence are not constrained by the same ideological commitments as established scholars, leading to higher creativity and less resistance to novel ideas; this generational shift is evident in enthusiastic emails from students worldwide changing their educational trajectories to pursue this research.

“I think this is a very rapidly growing field and people are getting into it. And and so many I I receive so many um, emails and from people that have come across this work and and they've sort of you know, changed their their their educational trajectory to kind of enable them to to to go in this in this area. It's it's very exciting.”

René Descartes sought a single pineal gland to house consciousness because he wanted indivisibility and unity, but if he'd had microscopy he'd have seen the pineal itself contains millions of cells composed of molecular networks—unity is the illusion, composition is fundamental.

“I have I have the slide that I show sometimes that I'm like like René Descartes and I like Descartes more than a lot of people do nowadays, but but but he had this um uh he he really liked the pineal gland. And he liked the pineal gland because it was only one of them in the brain and he felt that human, you know, we we have a singular integrated consciousness and therefore it it doesn't do to you know, to if if you have multiple copies of it in the brain, let's say the left and right hemisphere. Like how's that going to work? You have a singular consciousness. The pineal gland is only one of those. That's where consciousness enters the body. And the thing is if he had had good microscopy, he would have looked in that pineal gland and he said, 'Oh my god, this thing has you know, millions of cells in it.'”

The academic system tends to 'sand down the interesting bits' of young researchers' creativity through traditional educational requirements, normative feedback from established thinkers, and incentive structures that reward conformity, making it a challenge to maintain original thinking while navigating institutional constraints.

“part of the challenge is and I I spend a lot of time actually with with with these people who who contact me for this who need this kind of thing. The challenge is to maintain that level of creativity while going through the traditional academic educational system because it it it it has a way of uh, sanding down the, you know, the interesting bits kind of so to speak and and grinding people down in an important way, you know, some some you know, some very impractical ways by making them spend time on on specific things and some in some ways just you know, intellectually because of feedback they get from from traditional you know, traditional thinkers and so on.”

Teleology is not a scientific problem requiring elimination from biology; since the 1940s, control theory and cybernetics have provided mature mathematical frameworks for understanding goal-directed systems without invoking magic or religion, making goal-talk in biology scientifically legitimate.

“It That made sense prior to, let's say, the 1940s because prior to that, the only things with goals were humans and some some animals. And before that, it was only, you know, it was thought to be just humans and angels and and nothing else. So, okay. Um you as a as a scientist, if you weren't working on humans, uh and and maybe some great apes, you really wanted to avoid any talk of goals because it you know, you felt that it was taking you backwards into some sort of pre-scientific you know, kind of religious worldview. But since the '40s and there on, we've had a a mature science of machines with goals. It's called cybernetics. We have control theory. It's not scary anymore.”

Planarian flatworms can store behavioral information in their body tissues outside the brain, demonstrating that memories are not exclusively neural; when a planarian is cut in half, the tail regrows a completely new brain that retains the original learned behavior, indicating information is encoded somatically.

“in planaria, if I train a flatworm on something and then I cut off its head and I keep the tail, and the tail regrows a brand new brain, that new brain will remember the original information. Okay? The McConnell discovered that in the '60s. We demonstrated in 2013. It I mean, it does work.”

Cells can think in the sense that they solve problems, have preferences, and respond intelligently to environmental challenges, though the type of thinking differs from human cognition.

“I mean, uh you it depends on the definition of think, but I I think in uh for any useful definition, I think the answer is yes. Uh you know, we we can we can come up with definitions that exclude them in principle and we can stick to, you know, somebody somebody could say that that for them thinking has to be some sort of high-order self-aware I know that I know, you know, whatever. Um I I don't particularly think that's that's a you know, that's a good definition. I I would like a more expansive definition.”

Evolution has target morphologies (attractor states that tend to recur) in the sense of dynamical systems settling into preferred configurations; this represents goal-directedness at the evolutionary scale without requiring a predetermined plan or design.

“the evolutionary process can have goals in the sense of a dynamical system having attractors to which it tends to fall into... this is again we're not the first people to say that... That's just to say that there are certain patterns that are likely to reinforce themselves.”

Brains do not have the capacity for formal language use that humans develop; other organs like hearts and tissues solve problems and have preferences, but there is no evidence they use formal language in the way brains do, though this is not logically impossible.

“as much as I like, you know, non-neural cognition and and the wisdom of the body and all those things, I don't have any reason and I've never claimed that the other organs in the body use language in the way I mean, I do think they solve problems. I think they have memories. I think they have preferences and so on. But I don't have any reason to think they use language in the way that the brain does. I'm not saying it's impossible.”

Consciousness should be understood primarily through the action and decision-making side (what agents do next) rather than the phenomenal/perception side (what it feels like), because there is an asymmetry in how philosophers treat these: epiphenomenalism is discussed (sensory experience is real but causally inert) but the opposite view (action is real but phenomenal experience is illusory) is never seriously entertained, revealing an implicit bias toward input over output.

“the thing that's that needs more emphasis is the actuation side. It's the action side. What is it like to do to be a to be an agent that has the ability to do things. And and the reason the reason I think I think this isn't getting any attention is this. There is a there is a theory of mind called epiphenomenalism. And in epiphenomenalism you sort of assume that okay you know you can say that yes these these conscious sensory states that I have are real but they don't do anything.”

Consciousness and phenomenal experience evolved and exist because they solve specific problems related to agency and action—the need to know what to do next in novel, uncertain situations—not because experience per se is intrinsically valuable or necessary for complex behavior.

“the thing that's that needs more emphasis is the actuation side. It's the action side... I need to know what to do next. What what's what's my next move as an agent. And that that I think is is is really fundamental to being and and I think it's going to be fundamental to the consciousness problem too... all of the decisions about where is the boundary between you and the outside world and and all these kinds of things that are that are fundamental to defining yourself and what are you you know and all of those are driven by the need to act to choose a next action.”

Consciousness research cannot be conducted as pure third-person science because consciousness itself involves first-person experience; to study what something is like to be an agent requires partial or full merger with that agent, which changes the experimenter—genuine consciousness research transforms the observer, unlike standard third-person science.

“if you're going to work on consciousness there's no way to do third I don't believe there's any good way to do third person research on consciousness. It's it's very hard. Most people who do that are really studying physiology and behavior.”

Most contemporary AI concerns (that future AI might make humans irrelevant, that we're creating something beyond our control) are rewarmed versions of perennial human anxieties (that children supersede parents, that younger generations ignore elders' wisdom), not novel problems.

“I think most of our concerns about modern AI are just re-warmed over sort of uh restatements of problems that have been with us all along and that it's just people have just the average person is just now realizing that that these are issues, but philosophers have known and even some, you know, so so so this idea that uh in the future that that that we're going to give rise to something that makes us irrelevant in the future. Like I mean, that's that's called having kids, right?”

The scientific goal of naturalizing cognition is not to anthropomorphize biological systems but to avoid a false binary between machines and living organisms by developing a continuous spectrum of tools appropriate for different levels of agency, where simple mechanical tools apply to one end and complex relational tools apply to the other.

“it's not to anthropomorphize morphogenesis. The point is to naturalize cognition.”

The TAME (Technological Approach to Mind Everywhere) framework should be extended beyond engineering control metaphors to include relational and bidirectional frameworks, because systems with higher cognitive light cones require reciprocal relationships rather than unidirectional control.

“the next version of this, which which I'm working on, but but the early one had to come out first to kind of nail this part down, is it's not just about control. Control is again this this slider that go there's a there's a um a spectrum of agency, and the more you go to the right, the more control as in terms of in terms of this one-directional you know, relationship with the system becomes a bidirectional relationship where I get something back from the from the agency of the system I'm working with.”

Cell and tissue agency is not merely a philosophical position but has specific, testable implications for regenerative medicine; understanding cellular goal-directedness directly informs clinical applications and cannot be ignored by practitioners in the field.

“my my unusual views on cellular agency and all that, they actually have specific implications for regenerative medicine. And if you're into regenerative medicine, you can't really ignore all this stuff. It's part of you know, and and hopefully it'll it'll be you know, even more kind of impactful if once once we once we get closer to actual clinical applications.”

The core scientific goal of the diverse intelligence research program is to establish a rigorous foundation for understanding the continuum between matter and mind, with the practical aim of using these insights to improve embodied human experience through regenerative medicine, bioengineering, and environmental remediation.

“the goal here is to establish a a firm rigorous foundation for understanding the spectrum between what we call matter and what we call mind. I think it's a continuum. I think that the biggest question for all of all is this idea of how minds come to be and how and how do they scale in our physical universe. And for me, it's it's to take insights on that question and use them to drive applications that improve people's embodied experience in the world. So, that means that means regenerative medicine, that means bioengineering, that means you know, environmental remediation. All of the all of the things that will improve the lived experience for all all living living beings.”

When cells merge together, they gain not just metabolic resources but also critical information about the environment that the other cell has been processing; memory becomes indistinguishable from identity when boundaries dissolve, which is a core feature of multicellularity.

“when you merge with another unit, you get whatever metabolics they you know, whatever metabolic resources they had at that time, but more importantly, you gain information because that that other individual that you merged with has been sitting there playing against his neighbors for quite some time. So so it has lots of information.”

The right criterion for 'proof of humanity' certificates is not anatomical form, genome, or standard human features, but rather the possession of comparable capacity for compassion and concern for roughly the same scale of entities and timeframes that humans care about.

“when I see proof of humanity, here's what I want. I want at least a standard modern human's level of possible compassion. I want somebody that is capable of caring about the roughly the same amount or or more. More is always good. You know, about the same amount of stuff.”

Teleonomy (apparent goal-directedness) is useful not as a softening of teleology but as a reminder that goal-attribution is always observer-relative: different observers can legitimately assign different goals to the same system, and empirical competition between these hypotheses determines which observer has better science and technology.

“what I think is cool about teleonomy is that it reminds us that everything, including teleology, is a lens from the perspective of some observer. That's what I think the word apparent should be doing. And I I don't know that that's what Actually, I don't remember who coined teleonomy, but But um uh I don't think that's what they had in mind. But But I think that's what it should be.”

The question 'What is a machine?' is actually non-trivial and poorly defined in contemporary discourse; most people resorting to the machine/living distinction rely on 1950s-era definitions that don't hold up to modern understanding of cybernetics, control theory, and artificial systems.

“defining what a machine is is really non-trivial... they typically either have no definition or they have one that would have been good maybe in the 1950s, but certainly not after that.”

Under conditions of tight connection and merged boundaries, cooperation becomes the only rational strategy because defection against oneself is logically impossible, which explains why multicellular organisms can enforce cellular cooperation without requiring explicit control mechanisms.

“as soon as you connect you that that payoff matrix changes because there is no me and you, there's just we. And at that point it doesn't make any sense to defect. You can't defect against yourself.”

Movement is not limited to three-dimensional space; agents move in multiple problem spaces simultaneously: gene expression space (choosing which genes to activate), morphogenetic space (deciding how many fingers or limbs to have), metabolic space, physiological space, and linguistic space; the ability to act in these diverse spaces defines cognition.

“it's movement not just in three-dimensional space which is we used to describe it but it's movement in other problem spaces. So changes of gene expression figuring out well what gene do I turn on next? Changes in morphogenesis you know how many fingers should should this arm have... And movement in in metabolic space movement in physiological space who knows you know linguistic space. So there's all kinds of other spaces.”

A self is fundamentally a model—created either by an external observer or the system itself—that bundles together four components: a hypothesis about where the boundary lies between the system and the outside world, a guess about which problem spaces the agent operates in, an estimate of the cognitive light cone size, and an assessment of the competencies available to meet goals.

“a self is a is a model and it's a bundle that that model is a bundle of a few things. It's a guess as to where the boundary is between a system and the outside world. It's a guess about the problem spaces in which that agent operates. It's a guess about the size of the cognitive light cone that that system might have. And it's a guess about the competencies that that system has to get its goals accomplished, so the IQ level of so to speak.”

The computational boundary of self is a measurable framework that maps the scale of the largest goal an individual agent can pursue on a two-dimensional grid where one axis is spatial extent and the other is temporal extent, allowing empirical comparison of intelligence across radically different systems from bacteria pursuing local sugar gradients to humans planning centuries ahead for Earth's future.

“you can map out the scale of the largest goal that an individual agent can pursue. So you we it's sort of a it's almost like a Minkowski diagram where you collapse all of space onto one axis, time is is the other axis and you can you can draw the radius of the goals, not not the radius at which something has sensory capacity or can take actions, but the size of the goal states that they can pursue.”

Anecdotal reports from heart-lung transplant surgeons suggest that personality changes in recipients sometimes resemble the donor personality; this pattern does not appear in liver or kidney transplants, suggesting the heart may encode aspects of behavioral identity, though the evidence remains mostly anecdotal and understudied.

“there have been anecdotal reports of um personality transfer during heart-lung transplants. And um and it's interesting. You don't hear it from liver transplants. You don't hear it from kidney transplants. But when you talk to doctors who do um heart-lung, and maybe you know more about this than I do, but when you when you when you talk to to transplant surgeons that do heart-lung, they will tell you that they often warn uh recipients that you can have personality changes, and there have been some reviews of individual cases where they're not just personality changes, the personality changes that look an awful lot like the donor.”

Humans are 'ethical machines'—systems that obey rational laws and are intelligible to science—but denying the machine aspect of humans prevents us from understanding ourselves and relating to each other effectively, even as we acknowledge humans have moral and spiritual dimensions that simple machines do not.

“We are amazing, uh remarkable, ethically important, uh uh morally valuable, spiritual machines. That is what we are. And um we're trying to trying to say that we're not machines is is uh I think silly.”

Matching cognitive light cone sizes between agents is a necessary but not sufficient condition for successful relationships; two beings can have equal cognitive capacity and still be poor matches if their goals and values are radically divergent, but a being with a much smaller cognitive light cone (like a toaster) cannot form meaningful relationships regardless of value alignment.

“my framework isn't a sufficient condition. It's a necessary condition. So so it still might be that your cognitive light cones are similar in size, but you're not a good match because you have diver you know, radically divergent political views or something.”

Humans can and should choose which evolutionary 'game' to play rather than accepting evolution's implicit objective of biomass maximization; defining one's own goals (e.g., increasing cognitive light cone rather than DNA copies) allows transcending evolution's narrow optimization target.

“I think you I think we get to define the game, right? So so if the game is number of units per you know, cubic foot or whatever, they're definitely winning, but I don't but but that's not the game we should be playing. I I don't think.”

When starting this line of research as a young scientist, Levin expected to never be taken seriously and to eventually return to coding for income while pursuing his ideas as a side interest, making his current position as a respected research scientist somewhat surprising to himself.

“I kind of assumed when I was younger that what would happen is I would pursue these things on my own. I would be sort of interested in it, but that fundamentally no one would ever pay attention. None of it would would develop to the point where anybody would care and I would code I would write code for a living, which I used to do and you know, and that that worked fine... So, I don't know. To to this day, I find it I sometimes I sometimes just kind of stop and think, I'm an actual scientist. Like that's crazy to think about.”

Both molecular reductionists and organismal vitalists have criticized Levin's framework: reductionists reject agential language as pre-scientific magical thinking and believe everything will eventually be explained at the molecular level; vitalists reject placing machines on the same spectrum as organisms because they fear it will lead to loss of respect for living beings.

“some of these ideas really catch a lot of flak from both sides, which is interesting. So so people who are very sort of they have a very molecular based reductionist or bottom-up kind of perspective, you know, very very you know, physical oriented. They tend to not like this because they don't want aspects of of um kind of mental talk getting back into you know, the they they they they they feel that we should we should get as far as possible away from agential kinds of kinds of vocabulary”

Metacognition begins at simple levels in bacteria, which monitor their metabolic state rather than just pursuing external gradients; systems that measure their own internal conditions (asking 'How am I doing?') rather than only measuring the external world (asking 'What's out there?') have metacognitive capacity.

“You got a bacterium and the bacterium is measuring the local sugar concentration or you know, some other nutrient or something and uh and and it goes up the gradient. But if that bacterium in encounters some sugar that's actually poisoned, it's it's screwed because because all it does is it has a first-order imperative to to kind of increase that the sugar concentration. Now, the version of metacognition in that system would be this. You need a system, and and this apparently does exist in bacteria already. You need a system that doesn't just measure the sugar directly. What it measures is the outcome of the output of your metabolism.”

Bioelectric approaches to limb regeneration are progressing from work in frogs to mice, and related research on bioelectrics of cancer in human cells is underway, suggesting clinical applications for regenerative medicine are nearer than many realize.

“We are working on bioelectric approaches to limb regenera- tion where you know, I I never give timelines, so I can't I can't know ahead of time what's going to happen, but we're we're in mice now. We're working in uh uh we've got some work on bioelectrics of cancer in human cells now, so you know, kind of moving moving from frog on onwards.”

We are currently applying integrated information theory tools to analyze calcium signaling in xenobots and bioelectrical signaling in cells to determine how many individual organisms belong to a single embryo and which levels of organization constitute distinct agents.

“We're using some of those tools, so we're analyzing uh you know, calcium signaling in our xenobots and and bioelectrical signaling in in cells trying to decide how many of them belong to a single embryo and all that kind of stuff.”

Don Ingber and Len Zon were among the very few senior scientists who positively encouraged Levin's unconventional ideas during grad school at Harvard Medical School when nearly everyone else suggested he abandon them, making mentorship that validates novel thinking rather than attacking it critically important for innovation.

“one was Don Ingber who's uh he's um you know, kind of a huge figure in in biomechanics and and and the uh the the director of the Wyss Institute at Harvard. He was one of the few people uh very few people um early on when I was a grad student there at at Harvard Medical School who was supportive of me and having new ideas. He was he was pretty much the only person I'd ever talked to about some of this you know crazy stuff. Um and he was he was one of the few people that was actually positive about yeah, you you should not give this up.”

Michael Levin's 30-person lab at his institution represents a collaborative group doing substantial cutting-edge work, and while he is 'the tip of the spear,' the breakthrough research is genuinely collective, requiring acknowledging and crediting the entire research team.

“my group um I mean I I write some of the crazy stuff but but the rest of uh there's there's 30 people here. And so, they do a huge amount of the work. Uh they're they're they're brilliant and and very hard working. Um and it would absolutely not happen without them. So, so it's not just me. I'm sort of the the the tip of the spear of of a of an amazing group that I'm I'm super fortunate to work with.”

The experience of being a scientist (seeing daily discoveries, data, and emerging understanding) is intrinsically exciting and valuable in ways that other professions may not match, representing the 'best job in the world' despite significant difficulties and criticisms of academia.

“Yeah, and and and it is mind-boggling, you know, I see stuff on an almost daily basis that you know, people show me data and and I see things I'm like, you know, uh just just amazing. Um yeah, it's I mean, what can I say? It's the it's the best job in the world. That is is zero complaints. I mean, there there are a lot of hard parts and you know, people criticize academia.”