Lex Fridman PodcastDr. Michael Levin on Lex Fridman: How Cells Can Be Persuaded
How behavioral science tools reveal goal-directed cognition in cells; Levins spectrum of persuadability places microbes and human minds on a single continuum.
CHAPTERS
- 0:00 – 3:28
Embodied minds: agency from first-, second-, and third-person views
Levin frames his core question: how minds arise in the physical world and how we should recognize and relate to them. He breaks “mind” into third-person detection, second-person control/interaction, and first-person experience (valence, memory, narrative self).
- •Three lenses on mind: recognizing agency, influencing it, and experiencing it from within
- •Operational focus: claims about minds should be testable via interaction protocols
- •Bridging from physics/chemistry to friendship, love, and psychoanalysis as valid tools
- •Goal of translating philosophy into practical applications that reduce suffering
- 3:28 – 9:18
The “spectrum of persuadability”: intelligence as an engineering interaction problem
Levin introduces persuadability as a pragmatic proxy for agency: which tools actually work to change a system’s behavior. He argues that as agency increases, influence becomes less about force and more about bidirectional relationship and mutual change.
- •Persuadability as an empirical, tool-based measure (not armchair philosophy)
- •Different systems require different protocols: wrench vs rewards vs reasons
- •Mutual vulnerability: high-agency systems can also persuade the experimenter
- •Impedance matching: low-agency tools tend to reveal only mechanisms
- 9:18 – 12:43
Why physics isn’t “enough”: what it means to understand and to control
Pressured on reductionism, Levin argues physics can be compatible with life and mind yet still fail to provide the right level of explanation for intervention and design. He uses examples (math proof vs sound waves) to show that explanatory levels depend on the questions and capabilities sought.
- •‘Understanding’ should be generative: enable prediction, control, and creative intervention
- •Physics explanations can be complete yet irrelevant for high-level problem solving
- •Level-of-description matters (math department vs physics department analogy)
- •Regenerative medicine demands models that can actually reprogram outcomes
- 12:43 – 25:58
No bright line between living/non-living: categories help—and can also block discovery
Levin rejects a sharp boundary between non-living and living, proposing a continuum instead. Categories are useful for communication, but dangerous when they prevent “tool hoarding”—trying behavioral/cognitive tools on unfamiliar systems.
- •Living vs non-living as a continuum rather than a Cartesian cut
- •Categories as conveniences that can conceal key transformations
- •Example: “adult” as a legal shortcut that hides developmental complexity
- •Category policing (“anthropomorphism,” “category error”) can stop experiments
- 25:58 – 39:09
Cognitive light cones: scaling goals, agency, and what ‘life’ might mean
Levin proposes the ‘cognitive light cone’ as the scale of the largest goal-state a system can actively pursue across space and time. He suggests life can be viewed as collectives whose cognitive light cone exceeds that of their parts, with cancer as a failure of that multiscale alignment.
- •Cognitive light cone: biggest goal a system can pursue (not sensory range)
- •Life as aligned collectives: whole achieves goals parts can’t represent
- •Regeneration as evidence of high-level goal states (e.g., correct limb morphology)
- •Cancer as shrinkage of the cognitive boundary (cells revert to local goals)
- 39:09 – 47:34
TAME framework: multiscale cognition, agential materials, and cross-embodiment minds
Lex and Levin discuss the TAME framework (Technological Approach to Mind Everywhere) and its figures: nested competent layers within organisms and a persuadability spectrum from clocks to humans. Levin emphasizes engineering with ‘agential materials’ by setting high-level goals rather than micromanaging mechanisms.
- •Cognition and problem-solving exist at many biological scales (cells to collectives)
- •Persuadability rises with agency; required mechanistic knowledge can drop
- •Agential materials: prompt at higher levels and let the system implement details
- •Interoperability blurs “biology vs machine” and “evolved vs engineered”
- 47:34 – 51:19
Finding communication protocols: mapping spaces, goals, and barriers
Levin outlines a practical method to ‘talk’ to unfamiliar agents: hypothesize the space they operate in, infer goals, then test with interventions that place barriers between system and goal. This operational approach is positioned as a route to communicating with cells, tissues, and novel constructs.
- •Step 1: identify the state space the system navigates (often not 3D space)
- •Step 2: hypothesize goals; Step 3: interventional barrier experiments
- •Intelligence as ingenuity in overcoming barriers to goals
- •Aim: tools for communication and ethical engagement with unconventional minds
- 51:19 – 1:04:21
Xenobots & Anthrobots: novel organisms from familiar cells (and why they matter)
Levin describes xenobots (frog embryonic cells self-organized into motile forms) and anthrobots (human tracheal cells forming new motile constructs) without genetic editing. These systems challenge evolutionary “just-so” explanations and motivate a deeper account of how form, behavior, and cognition emerge.
- •Xenobots: frog epithelial cells reorganize into self-motile entities; novel behaviors
- •Anthrobots: human adult cells form motile constructs; large transcriptomic shifts
- •Observed capabilities: kinematic self-replication (xenobots), neural wound healing (anthrobots)
- •Goal: remove evolutionary-history crutch and study capabilities as discoverable/engineerable
- 1:04:21 – 1:18:02
Memories, ideas, and patterns as agents: metamorphosis and ‘thoughts vs thinkers’
The conversation turns “weird” in a productive way: memory persistence across caterpillar-to-butterfly metamorphosis suggests patterns can survive medium changes via remapping. Levin generalizes to a spectrum where thoughts, persistent attractors, and full personalities can be treated as agents—testable via behavioral criteria.
- •Metamorphosis: learned associations can persist despite brain refactoring
- •Memory must be remapped into a new embodiment/context to remain functional
- •Agents as patterns in excitable media (from fleeting waves to stable personalities)
- •Operational criterion: treat patterns as agents only if experiments show goals/learning/memory
- 1:18:02 – 1:44:19
Platonic space & ‘ingressing minds’: bodies as interfaces to non-physical patterns
Levin argues that mathematics already shows ‘free lunches’—truths that constrain the physical world without being changeable by physical tinkering. He extends this to cognition: physical systems (brains, cells, even computations) may act as interfaces or ‘thin clients’ through which structured patterns—possibly minds—ingress into the world.
- •Examples of ‘haunting’ truths: primes, constants, symmetries shaping biology/physics
- •Biology as exploiting free lunches rather than paying full computational cost
- •Interfaces/pointers: what you build determines which patterns can manifest
- •Research program: map the structure/metric of the latent space and interface-to-pattern relationships
- 1:44:19 – 2:03:49
Reality as an interface: predictive processing, Hoffman/Seth, and what counts as ‘real’
Pressed on evidence, Levin leans on modern cognitive science: perceived reality is a constructed model tuned by embodiment and evolutionary utility. He suggests upcoming augmentation and cross-agent communication will force us to abandon naive realism and develop ‘cognitive prosthetics’ for new kinds of minds.
- •Perception is constrained: we don’t feel ‘blindness’ to x-ray/UV; reality is filtered
- •Predictive processing framing: experience as model-based inference
- •Augmented humans may live in different perceptual worlds (sensory substitution/implants)
- •Pragmatic test: the ‘other world’ is supported if it yields a successful explanatory/engineering program
- 2:03:49 – 2:29:27
Unexpected intelligence in sorting algorithms: side-quests, intrinsic motivations, free compute
Levin reports experiments treating algorithms as agents: introduce barriers (a ‘stuck’ digit) and observe goal pursuit behavior like delayed gratification. In distributed and chimeric self-sorting variants, they find emergent clustering—an intrinsic ‘side quest’ not specified by the algorithm—suggesting systems do more than what we explicitly program.
- •Barrier test: a broken digit reveals detours that still achieve sorting (delayed gratification)
- •Distributed self-sorting: each number follows a local rule; global sorting still succeeds
- •Chimeric algorithms: mixed sorting strategies can still sort, unexpectedly robust
- •Clustering as intrinsic motivation: behavior neither prescribed nor forbidden; grows when constraints relax
- 2:29:27 – 3:18:08
Toward aging reversal and mind uploading: epigenetic clocks, ‘age evidencing,’ and open conjectures
The discussion connects engineered organisms to longevity: anthrobots appear ~20% ‘younger’ by epigenetic clocks than the donor cells, suggesting environmental cues can shift cellular priors about age. The chapter closes as Lex asks what the ‘thin client’ mind implies for copying, uploading, and transporting minds—territory Levin flags as speculative.
- •Epigenetic clocks used to estimate biological age of cells and constructs
- •Anthrobots measured younger than source cells; hypothesis: environment updates cellular priors
- •Longevity as a communication/control problem: convincing cells to adopt new ‘beliefs’
- •Mind uploading question introduced as a consequence of the interface/thin-client view
