This episode features a deep conversation between Curt Jaimungal, biologist Michael Levin, and neuroscientist Karl Friston about the nature of information, meaning, self-organization, and the free energy principle. The discussion weaves together physics, information theory, and biology to explore how living systems create meaning, maintain identity, and navigate an uncertain world.
Information vs. Meaning
Information in the Shannon or information-theoretic sense is about quantifying the improbability of events (self-information) and the average uncertainty (entropy). But meaning requires an observer.
Under the free energy principle, meaning arises when an observer holds probabilistic (Bayesian) beliefs about the observed. The meaning of neuronal activity, for example, is what it represents about the sensory input that caused it.
Meaning is inherently relational: it exists in the separation between observer and observed, and in what the observer’s beliefs are about the causes of its sensations.
The Bowtie Architecture and Creativity
Michael Levin proposes a “bowtie” architecture for how agents process information across time. The left side compresses past experience into a generative model at the center (the present). The right side involves interpreting those compressed traces to guide future behavior.
The right side of the bowtie is fundamentally creative because information is lost during compression; you cannot algorithmically reconstruct the past. Instead, you must constructively fill in gaps and make the information useful for novel future scenarios.
Karl Friston connects this to the information bottleneck idea (from Tishby and others): the present is the simplest Markov blanket, a bottleneck through which all predictive information about the future must pass. Compression forces the system to find simple explanations for its sensory input.
Creativity, from the physicist’s perspective, is acting to gather the right kind of information that allows simpler explanations of one’s world—a form of generalized homeostasis.
In-Painting vs. Out-Painting in Biology
Levin distinguishes between “in-painting” (filling in minor gaps within an existing structure, where old tricks still work) and “out-painting” (going beyond past experience into genuinely novel territory).
Example: When human tracheal cells are placed in a new environment, they self-assemble into xenobots or “anthropods” and radically alter gene expression—about half the human genome is differentially expressed—without any genetic modification. They find novel transcriptional motifs and adaptive behaviors that were not pre-programmed.
This out-painting is not random trial-and-error; it is coherent, adaptive problem-solving that goes beyond what the past has prepared the organism for.
The Genome as a Bag of Tricks, Not a Blueprint
Levin argues the genome is not a memory of past environments but a collection of information-processing tricks that together create a problem-solving agent. This is “selection for learnability”—a step beyond Stuart Kauffman’s “selection for selectability.”
When the layer between genome and phenotype is treated as intelligent (a problem-solving system), evolution dynamics change: the system’s ability to repair and adapt hides information from selection about what the genotype actually encodes. Pressure shifts from the structural genome to the competency of the problem-solving layer.
Planaria exemplify this extreme: because their somatic cells are genetically unstable and lack a strict Weissman barrier, they have evolved an algorithm that builds a correct worm regardless of hardware mutations. This explains why transgenic Planaria and mutant strains essentially don’t exist—the system doesn’t take the genome literally.
Other organisms sit on a continuum: C. elegans takes the genome very seriously (hardwired cell lineages), salamanders less so, and Planaria hardly at all.
The Present as a Markov Blanket
Karl Friston explains that the present moment is the simplest Markov blanket because everything needed to predict the future that can be determined from the past is written on the present. This is definitional of Markovian processes.
This makes the present a natural information bottleneck and a useful mathematical device for understanding how systems compress past information to predict the future.
Noise, Prediction Error, and Creativity
There are two kinds of noise: (1) foundational random fluctuations that give the universe its probabilistic structure (underwriting quantum mechanics and inference), and (2) prediction error—the surprise or mismatch between what the model expects and what the world provides.
Prediction error is not something to be merely minimized; it is the driving force of belief updating and self-organization. Without it, a system would be at thermodynamic equilibrium—effectively dead.
Prediction error is technically the variational free energy gradient; all self-organization that looks like inference is driven by it.
The Self as a Story and a Pattern
Levin and Friston discuss the self as an ongoing constructive process of storytelling—interpreting traces left by past selves to guide future action. The self is a “self-telling story.”
Levin proposes dissolving the distinction between the thinker (the physical machine) and the thought (the pattern in the medium). From the perspective of “mortal computation” (a concept from Jeff Hinton and Alex Wissner-Gross), it is the patterns that matter, not the substrate. The pattern is the data, the dynamics, and the self-organization all at once.
The self is not deflated by being called a story; it is a self-referential loop that locks itself into existence. The pattern is the real thing; the physical substrate is what implements it.
Friston agrees: in mortal computation, there is no meaningful separation between pattern and substrate. The interesting questions are in the patterns themselves.
This has practical implications: for example, in bioelectricity, it matters whether the body operates on bioelectric patterns or whether the patterns drive the biochemistry.
Organic vs. Psychological Disease
Under the free energy principle, the distinction between organic and psychological disease dissolves. All self-organization can be read as belief updating, which is inherently psychological.
Psychopathology can be cast as false inference: hallucinations, delusions, and dysmorphophobia are all failures to attend to the right things or to ignore the right things—errors in the constructive process of sense-making.
Cancer, in Levin’s framing, can be understood as a cell having wrong beliefs about the boundaries of itself—a failure of self-modeling analogous to ego boundary issues.
Aging as a Psychological Phenomenon
Standard theories of aging treat it as noise-based: accumulated damage or errors over time. Levin’s computational models of goal-directed morphogenetic systems suggest a different possibility.
In these models, once a system accomplishes its morphogenetic goal (building the correct body), it has no further goal. Without a new goal, the system begins to degrade—not from external damage, but from what Levin describes as an “existential boredom” of the somatic intelligence.
This suggests aging might be fundamentally a psychological phenomenon of goal-directed systems, not merely accumulated hardware errors.
Death, Immortality, and Solenoidal Dynamics
Friston argues that even in a hypothetical heaven with no organic damage, a human could not stay sane indefinitely. Mathematically, reaching a steady state and staying there is “oscillator death”—equilibrium physics death.
Life requires solenoidal dynamics (also called Red Queen dynamics or detailed balance violation): the system must keep moving through its characteristic state space. If you attain detailed balance, you are dead in the dynamical sense.
Death is part of the natural cycle of life, not a failure. Immortalizing a system in a fixed state would be a form of death because it loses the itinerant, far-from-equilibrium dynamics that define living systems.
However, a form of continuity is possible across scales: reproduction is itself a solenoidal dynamic. At the species or transgenerational scale, the self can persist in a changing form. Cyclical motions at every scale—from planetary orbits to gamma oscillations in dendrites—contextualize each other, and the recurring theme is revisiting characteristic states.
Scale-Free Consciousness and Counterfactuals
Consciousness may be scale-free: processes associated with conscious processing (curiosity, prediction, exploration, counterfactual reasoning) can be found at multiple scales, from cells to societies to celestial mechanics.
A counterfactual (as used here) is a special kind of hypothetical: it is a hypothesis about what would happen in the future if you acted in a particular way. It is future-pointing and action-dependent.
Counterfactual reasoning requires a certain statistical architecture: the system must be large enough that it does not have direct access to its own actions and must infer what it is doing from sensory consequences. This is “planning as inference.”
Once a system can infer its own actions, it can also infer the consequences of alternative actions—and then evaluate which course of action is better. This opens the door to metacognition (inferring about one’s own inferences) and higher-order thought.
The Constructive Nature of Science
Levin expresses concern that scientific accounts of the self as a collective intelligence or self-constructing pattern can be misinterpreted as deflationary—leading some people to feel they don’t exist or that something magical has been lost. He emphasizes that these accounts should be constructive and useful for living, not destabilizing.
Friston responds that the scientific process itself is a creative, constructive act that complies with the same principles (inference, compression, free energy minimization) that it describes. The goal is understanding the mathematical architecture, not diminishing human experience.
Both agree that the stories scientists tell should be the best picture of the world and also practically useful for people’s lives.
