Ion channels, electrical synapses, and neurotransmitter pathways predate nervous systems. They ran in bacterial biofilms. Evolution did not invent cognition. It just made it faster.
Before cells navigated 3D space, collectives of them navigated anatomical morphospace — the space of possible body shapes — using the same electrical vocabulary. Planaria store memory in bioelectric networks, not in their DNA: cut the head off a trained worm and it regrows a new brain that still remembers. Rewrite a worm's bioelectric "target morphology" and it grows two heads; cut it again and it rebuilds two heads, genome untouched. Xenobots and anthrobots — frog skin cells, adult human cells — self-organize into motile creatures their evolutionary history never asked them to become.
Levin refuses to draw a hard line between cognition and mechanism. Instead: a cognitive light cone — how far, in space and time, a system can pursue its goals. A bacterium has a tiny one. A salamander cell collective regenerates a limb. A human plans a century ahead.
Friston extends the Free Energy Principle all the way down. Any system that persists — particle, cell, person, institution — has a Markov blanket, and must behave as if it is inferring its world. The FEP is scale-invariant. It applies to existence itself.
And yet Friston draws a line Levin does not. Simple systems minimize free energy reflexively. True cognition requires temporal depth — a generative model that can roll into the future, simulate counterfactuals, ask what would happen if I did that? An E. coli cannot do this. A thermostat cannot. A human can.
Any system that navigates toward a goal, at any scale, is cognitive. The particle following least action is the most primitive version of the same thing.
Reserve "cognition" and "agency" for systems with the architecture to explicitly represent alternative futures. Everything below that is precognitive.
They do not contradict. They use different terminological thresholds for when a thing minimizing action becomes a thinker anticipating the future.
Bayesian belief updating has a thermodynamic cost — Landauer's limit, Jarzynski's equality. The brain runs close to the theoretical minimum. Every counterfactual is paid for in heat.
Nick Lane argues the entire prevailing picture of biology — information, genes, selection — is missing the engine. Life is a continuous chemical reaction driven by charge. Proton gradients across a membrane. Pump protons. Generate an electrical field strong enough to qualify as a bolt of lightning. Use that charge to fix carbon, build ATP, stay alive.
Bacteria were stuck for two billion years because the membrane was also where their genome had to live — at the cell's surface, pressed against the charge. Endosymbiosis broke that bottleneck: one cell swallowed another, the swallowed cell became mitochondria, and suddenly the power grid lived inside the cell. Eukaryotes became possible. Brains became possible.
The cell membrane integrates the whole organism's condition into one electrical state. For Lane, that integration is the biological root of feeling. The proton gradient is both the engine that runs Friston's math and the informational medium that makes awareness possible. You cannot think without paying the heating bill.
The math checks out. The Evidence Lower Bound optimized by an LLM's training loop is mathematically identical to negative variational free energy. Attention heads play a computational role close to the neurotransmitter systems that regulate human attention. In the narrow sense of function approximation, LLMs minimize free energy.
And yet. A true Markov blanket requires bidirectional, reciprocal exchange — sensory states influencing internal states, active states influencing the world. Current LLMs are fed. They cannot select or solicit their own inputs. They are, in Friston's phrase, sessile artifacts.
The missing ingredients are specific: curiosity (epistemic foraging — actions taken to resolve uncertainty), temporal depth (genuine counterfactual simulation), and a self-model (an entity that distinguishes its own actions from the world's). Scaling does not add these. Nor does running faster on more GPUs.
Friston's prerequisite is embodiment — not necessarily biological, but a substrate that must pay its thermodynamic heating bill or dissipate. Until the machine has skin in the game of its own persistence, it is not thinking. It is just calculating in someone else's dream.
Asked about the Free Energy Principle directly, Chomsky: "Frankly, I don't know of any such principle that has any consequences or implications." Asked about basal cognition: human language has no analogs in other organisms. Do not mistake communication systems for the thing.
But he is not a mystic. He is a mysterian, which he treats as a biological truism. We are organisms. Organisms with capacities have limits. The capacity to run is also the incapacity to fly. Our minds are built for certain problems (which we can solve) and against certain mysteries (which are simply outside our cognitive reach). Physics hit this wall in the seventeenth century and gave up on intuitive ultimate meaning. Chomsky says we should do the same with consciousness.
Merge gives humans the counterfactual bandwidth Friston requires — unbounded combinatorial thought. But Chomsky insists this only explains the puppet. About the puppeteer, he says we have nothing to say, and never will.
Half the archive believes the hard problem is an artifact of a false materialist premise. The other half believes qualia are a user illusion. The disagreement is as sharp as physics gets.
The bridge sits in Chris Fields' work: the Markov blanket — the statistical partition between observer and environment — is already, on its own terms, a theory of awareness. Scale it down and you don't hit dead matter. You hit a self-organizing boundary that is perpetually inferring its world.
At the Planck scale, space and time stop having operational meaning — any attempt to measure something that small collapses the region into a black hole. Theoretical physicists are hunting for structures that live outside space and time, from which space-time emerges as a low-resolution projection.
Under the disagreement — quantum vs. classical, continuous vs. discrete, matter vs. mind — a quiet convergence: relationalism. Whatever the substrate is called, its elements have no intrinsic properties in isolation. They are defined entirely by relation.
Friston and Wolfram arrive at the same picture from opposite directions. To be a computationally bounded observer, you must be separated from the rest of the universe — and that separation is a Markov blanket. Internal states insulated by a blanket cannot access the outside directly. To survive, they construct a simplified predictive model of the hidden causes. Space-time and particles are the dashboard that model generates. Render the true infinite complexity and you dissolve back into entropic soup.
The core live disagreement: does consciousness emerge from physical spacetime, or does physical spacetime project out of consciousness? Four tests, each falsifiable, each plausibly near-term.
Every station in this conversation has been quietly about time. Counterfactual depth happens in time. Markov blankets persist through time. Cognition is a light cone. Entropy has a direction. And yet time is the one thing physics has never been able to dissolve cleanly.
Space is the relational network of nodes at a moment. Time is the inexorable, computationally irreducible updating of those nodes. Smolin: space doesn't exist. time exists, and is fundamental. Space is what we aggregate when our observation is slow; time is the engine doing the aggregating.
Rovelli's thermal time: past and future only emerge when you coarse-grain. Barbour: the universe is a timeless collection of Nows. Adlam: laws don't evolve step-by-step; they constrain history globally, like Sudoku. Time doesn't flow. We deduce it from change.
For the active-inference thinkers, time is an internal model. Friston notes that deep planning requires quantizing time into discrete steps — every 250 to 300 milliseconds, mimicking saccades and whisking. Bach calls us vortices riding a traveling wave, our sense of time the rotation of our own updates.
The deepest inversion concerns the arrow of time itself. Does thermodynamic irreversibility create consciousness — leaving footprints and memories, giving our choices the status of real interventions — or does consciousness create thermodynamics, entropy appearing only because a bounded observer has lost information by projecting itself into a headset?
The archive's uncomfortable suggestion is the second. Wolfram: the Second Law is a direct consequence of the observer being computationally bounded and believing itself to be persistent. Drop those cognitive constraints and the arrow disappears. Hoffman: entropy is a mathematical artifact of projection — take a lossy image of the timeless network and the image necessarily grows disordered.