Michael Levin on Minds, Biology and Intelligence

Levin frames the study of minds through first-, second- and third-person perspectives: inner experience, interaction and control, and how observers recognize agency. He argues that behavior and practical applications should be central to understanding.

Using regenerative medicine as an example, Levin describes a spectrum of ways to influence systems, from molecular intervention to high-level prompting. At higher levels of agency, interaction becomes mutual: the researcher may also be changed by the system.

Levin says physics can explain important aspects of systems but may not provide the tools needed to understand or change them in practice. He rejects a sharp line between living and non-living or mind and non-mind, arguing that categories can prevent researchers from trying useful methods across fields.

The conversation considers whether life has a definable starting point. Levin favors studying gradual transformation and scaling rather than searching for a single dividing line, while acknowledging that categories can be useful for communication when their limits are kept in view.

Levin argues that fixed categories can make it harder to recognize unfamiliar forms of cognition, including possible life beyond Earth. He recommends making specific, testable claims about capacities, and the speakers discuss steelmanning opposing views.

Levin clarifies that the engineering framing is intended to make claims useful in regenerative medicine, not to reduce relationships to control. He argues that relying only on low-level molecular tools is an assumption that experiments can challenge.

Levin defines the cognitive light cone as the scale of the largest goal a system can actively pursue. He uses examples ranging from bacteria to humans and describes how cells can coordinate toward collective goals, such as building and regrowing a limb.

The speakers clarify how goals can be separated from current states and discuss the scale of concern and compassion. Levin says researchers should test for agency by putting barriers between a system and a hypothesized goal, rather than relying on intuition or treating anthropomorphism as a reason not to investigate.

Levin discusses trying behavioral and cognitive tools on systems beyond brains, recalling Bose's work on plants and materials. The speakers consider whether methods such as psychedelic experiments could be applied to algorithms, and introduce a framework for studying diverse bodies and minds.

Levin explains the TAME framework as a technological approach to studying minds through interactions and testable protocols. He describes how persuadability, autonomy and the mechanism knowledge needed to influence a system vary across examples such as clocks, thermostats, dogs and people.

Levin explains how abstract goals can ultimately guide bodily processes without conscious micromanagement. He describes each biological level as capable of problem-solving in its own state spaces and argues that life can combine evolved and engineered components.

The speakers consider whether groups or weather systems might display cognition, emphasizing that experiments would be needed. Levin illustrates communication between different ways of representing a problem with a tic-tac-toe and number-puzzle analogy.

Levin describes creating novel biological systems to investigate capabilities that cannot be explained simply by their evolutionary history. He discusses mobile xenobots made from frog cells and anthrobots made from adult human tracheal cells, including reported behaviors and an in-vitro neural wound-healing result.

Levin describes learning in gene-regulatory networks and the potential for AI to help researchers communicate with biological systems. He explains that cells navigate anatomical possibilities and says experiments can reveal how they respond to obstacles and unfamiliar conditions.

Levin contrasts open-loop models with systems that can reach outcomes by different routes. He describes developmental responses to obstacles and says that bioelectric research has enabled his lab to identify and rewrite some biological goal states.

The speakers explore whether ideas and memories might be treated as agents. Levin uses memory persistence through caterpillar metamorphosis to discuss patterns that change in order to persist, then proposes testing patterns for learning, memory and goal-directed behavior.

Levin questions the fixed distinction between software and hardware, offering alternative ways to describe which part of a system acts as agent. He connects these ideas to bioelectric pattern memories, aging and possible research approaches involving either the patterns or the cells that respond to them.

Levin considers whether disease could involve physiological, informational or cognitive states, and says bioelectric imaging may make some such patterns experimentally accessible. The conversation then introduces the Platonic space idea and an interdisciplinary series of talks about it.

Levin explains that the Platonic space name refers to the idea that mathematical truths are discovered rather than invented. He argues that mathematical structures constrain physical outcomes, using cicada cycles and mathematical constants as examples.

Levin suggests that biology can exploit patterns that are not themselves produced by evolutionary history, and questions whether calling surprising capabilities emergent is enough to explain them. He proposes mapping the relationship between physical interfaces and the patterns they make accessible.

Levin distinguishes mathematical patterns from more complex patterns that could be recognized as minds. He uses mathematical examples to explain how physical systems may reveal patterns that are not specified by their components alone.

Levin proposes that brains may be interfaces through which patterns associated with minds are expressed, rather than producers of consciousness. He connects this claim to the role of mathematical truths in classical physics and discusses its implications for physicalism.

The speakers discuss whether mathematical facts can be changed by altering physical conditions. Levin argues that once assumptions are chosen, further mathematical truths follow, and says biology may exploit structures that physics is constrained by.

Levin describes his view as radical but related to older dualist ideas, while noting the difficulty of explaining interaction between non-physical minds and physical brains. The discussion turns to perception, constructed experience and the limits of what humans can directly sense.

The speakers consider how sensory extensions and unfamiliar beings could lead people to experience different aspects of the world. Levin says the proposed space would need to be mapped through a research programme that identifies connections among patterns.

Levin describes comparing the outcomes of frog genomes and xenobots to map how physical systems relate to behaviors and forms. He says the space could include mathematical, anatomical and psychological patterns, while distinguishing current findings from hypotheses.

Levin says systems can produce more than researchers explicitly designed, but does not claim to know the extent or structure of the proposed space. He frames the idea as a model whose value will depend on whether it enables successful predictions and research.

Levin expects any underlying structure to have systematic relationships, though he does not expect its laws to resemble physics. He suggests computational systems may also exhibit unanticipated patterns and says researchers should look beyond the behavior they explicitly trained them to produce.

Levin explains why his group studied simple, deterministic sorting algorithms: to test whether unexpected competencies require biological complexity. He describes the basic sorting task and the experiment of making one digit unable to move.

When one digit is immobilized, the sorting algorithm temporarily becomes less ordered before sorting around the obstruction. Levin compares this behavior to delayed gratification and then describes a distributed version in which each number follows an algorithm.

Levin describes combining digits that follow different sorting algorithms. Although the numbers still sort, the digits also cluster by the algorithm they follow, a behavior the code does not explicitly instruct them to perform.

Levin says the clustering appears without extra algorithmic steps and calls it a possible intrinsic motivation. He describes how allowing repeated digits increases clustering and suggests that minimal systems can display behavior researchers did not set out to produce.

The conversation connects the sorting results to the idea that systems may do things that are neither required nor prohibited by their instructions. Levin says researchers should look for these behaviors in algorithms, social systems and AI, rather than assume that machine rules fully predict what matters.

Levin discusses testing other systems for unexpected competencies and says it is not yet known how these behaviors relate to an AI's main task. He also reports that anthrobots healed a neuron-culture wound in the first test his group tried, and describes measuring their biological age.

Levin proposes that anthrobots may appear younger because their new environment changes the cells' expectations about their age, while noting that this is a hypothesis under investigation. He connects this to the challenge of persuading cells to take on new roles in regenerative medicine.

Levin speculates about whether minds could be copied or instantiated through new interfaces, and discusses reports of people with little brain tissue who have normal or above-normal intelligence. He then introduces the idea of an agent 'booting up' and developing ownership of its parts.

Levin describes an organism's early development as the alignment of parts around a shared goal and considers how boundaries between self and environment form. He discusses chemical networks that learn and reports that learning can increase integration in some networks, creating a possible feedback between agency and learning.

The speakers consider how a single cell develops into an organism with complex behavior. Levin describes this as an increase in the capabilities of the physical interface, with new forms of cognition building on existing capacities.

Levin explains two proposed mechanisms for coordination among cells: spreading stress signals that increase plasticity, and gap junctions that let cells share information. He links electrical disconnection in cancer to loss of collective coordination and says reconnecting cells may restore some behavior.

Levin argues that cells already navigate unfamiliar spaces and solve problems that people may overlook. The speakers discuss applying intelligence tests cautiously, the difficulty of studying large-scale systems, and reports that ant colonies respond to visual illusions.

Levin says better tools for recognizing unfamiliar forms of intelligence are important for the search for extraterrestrial life. He notes that scientists do not agree on what defines life, and explains that he shares unconventional ideas publicly as empirical work makes them actionable.

Levin describes questioning assumptions, looking for continuities between categories, walking in nature and taking photographs as parts of his creative process. He records ideas in voice messages and organizes projects and writing through a large mind map.

Levin advises researchers to separate practical decisions about communicating work from a space where ideas can develop without concern for acceptance. The speakers discuss the value of practical feedback and the risk of broader advice constraining unconventional thinking.

Levin describes patterns entering systems as a kind of universal steganography and says he finds it meaningful that such patterns may apply to machines as well as living things. He and Fridman discuss the implications of treating these possibilities as part of one spectrum.

Levin says he would first ask how much people should rely on an advanced AI rather than discover things themselves, and what important question they are missing. Fridman says he would ask how many intelligent alien civilizations exist, before the conversation closes.