Levin, Michael
▶ Watch on YouTubeCore ideas
Levin frames embryogenesis as proof that the transition from "mere physics" to mind is gradual and continuous rather than marked by any discrete threshold, arguing DNA encodes only the cellular "hardware" (proteins, ion channels, signaling machinery) while the resulting organism's form and behavior emerge from that hardware's interaction with generic physical and computational laws — much as evolving one angle of a triangle gives you the third "for free" from geometry. He illustrates this with xenobots: frog skin cells, when removed from the rest of the embryo (with no genetic modification), abandon their default passive role and self-organize into small mobile proto-organisms capable of navigating mazes and even kinematically self-replicating by gathering loose cells into piles that mature into new xenobots — behavior with no evolutionary history, arising purely from cells' latent plasticity once freed from the signals that normally "bully" them into being skin. He resists a sharp hardware/robot vs. biology/organism distinction, arguing the meaningful engineering question is simply how much a given material or system can be depended on to act autonomously (comparing building with Legos, which stay where placed, versus training dogs, which have their own agenda but can be motivated) — a framework he extends to regenerative medicine, arguing that top-down, goal-directed communication with tissues (as opposed to micromanaging gene expression) is the more tractable path forward, illustrated by salamander limb regeneration, which reliably stops exactly when a correctly-shaped limb is complete.
A central recurring theme is that biological intelligence is inherently collective and hierarchical: there is no "central dictator" cell, only a multi-scale competency architecture in which molecules, cells, tissues, organs, and whole organisms each pursue their own local goals, with higher levels "bending the option space" for lower levels so that local, ignorant behavior nonetheless produces coherent large-scale outcomes (illustrated by "Picasso tadpoles" with scrambled facial features, whose tissues nonetheless self-organize into normal frog faces via novel migration paths). He connects this to bioelectricity — networks of ion channels and gap junctions that predate neurons and brains by hundreds of millions of years, allowing cell collectives to store and recall "memories" of target anatomical shapes (demonstrated by inducing two-headed planaria via bioelectric pattern changes alone, with no genomic edits, an alteration that persists across the worms' asexual reproduction) — as a plausible origin point for the sense of self, agency, and eventually free will, which he suggests arises in any energy-constrained, self-constructing agent that must coarse-grain its environment into approximate, actionable models rather than track every microstate. He extends this same bioelectric framework to cancer, describing tumor cells as reverting to a "smaller self" once disconnected (via closed gap junctions) from the tissue-level electrical network that normally keeps their goals aligned with the organism, and describes his lab's success reducing tumorigenesis from potent oncogenes by artificially forcing cells to remain electrically networked — alongside a company (Morphoceuticals) already using similar bioelectric signals to trigger frog limb regeneration in a 24-hour treatment window, with trials now underway in mammals.
On classification and terminology, Levin argues that categories like "robot vs. organism," "living vs. non-living," and worries about "anthropomorphizing" cellular or synthetic systems are outdated holdovers that will not survive advancing science, and insists that any claim about a system's cognition — that it "wants," "remembers," or "anticipates" something — should be treated as an engineering hypothesis to be tested by whether it usefully predicts and controls behavior, not settled by philosophical fiat. He introduces several of his own coined or adopted concepts, including the "cognitive light cone" (a way of comparing how large a spatial/temporal goal any agent — from a bacterium to a human pursuing world peace — is capable of pursuing, independent of substrate), and states his ultimate applied goal is an "anatomical compiler": a future tool where a desired body shape or organ could be specified and converted into the precise cellular stimuli needed to grow it, which he argues would eliminate birth defects, traumatic injury, cancer, and degenerative aging as medical problems by enabling continuous regeneration rather than late-stage intervention. Asked about consciousness directly, he is candid that he does not have a theory of the "hard problem" he considers useful yet, distinguishing it from cognition (which he considers empirically tractable) and suggesting that any correct theory of subjective experience might only ever be expressible as something closer to poetry than as falsifiable predictions.
The conversation closes with reflections on death (noting planaria are effectively immortal and biologically continuous with individuals 400 million years old, and discussing puzzling cases — like rats that "give up" and drown despite physical capacity to keep swimming — where organisms appear to deprioritize survival itself), on the ethics of expanding moral consideration to non-traditional minds (while noting that anatomical similarity has historically been a poor predictor of ethical treatment among humans themselves), and advice to young scientists to sharply distinguish specific, substantive critique (valuable) from senior figures' general "meta" guidance about what to study or how to think (often miscalibrated to a field's past rather than to genuinely new ideas, and best filtered through one's own gradually built intuition).
Excerpts
On what xenobots reveal about the "hidden" plasticity of ordinary cells:
“None of this are things that you would have expected from the frog genome... We've made it impossible for them to replicate their normal way, and within a couple days these guys find a new way of doing it that's not done anywhere else in the biosphere.”
On why he rejects sharp categories like "robot" versus "organism," reframing the real engineering question:
“As an engineer, what you really want to know is: what can you depend on this thing to do? ... How much agency does it have, which then tells you what techniques to use — do you micromanage it, do you train it, do you give it signals, do you try to convince it to do things?”
On his ultimate long-term goal for regenerative medicine:
“The end game is something you would call an anatomical compiler... you would sit down in front of the computer and draw the body or the organ that you wanted... and it would convert that anatomical description into a set of stimuli that would have to be given to cells to convince them to build exactly that thing.”