[@thegiantsshoulder] Michael Levin & Earl Miller: Bioelectricity, Platonic Spaces, Cell Intelligence & Neurobots

Link: https://youtu.be/6mU5nN5nlLA

Duration: 99 min

Transcript: Download plain text

Short Summary

This episode features two wide-ranging conversations exploring bioelectric fields as a foundational layer of cognition, memory, and morphogenesis. Neuroscientist Earl Miller argues that electric fields—not just spikes—underwrite cortical computation and consciousness, while bioelectricity researcher Michael Levin describes how cellular collectives use pattern memory to drive development, including inducing ectopic eyes on tadpole tails and building xenobots and anthrobots. In a separate segment of The Giant Shoulder podcast, host Evan interviews researcher Jens, who builds minimal deterministic computational systems to test for unprogrammed problem-solving competencies and argues physicalism has been "dead since Pythagoras."

Key Quotes

1. "Physicalism as such has been dead since the time of Pythagoras and probably long before that." (00:00:00)
2. "I think math is a behavioral science of a certain kind of pattern." (00:00:08)
3. "Bioelect electrical waves are the engine of life all the way down." (00:01:06)
4. "Memory, intelligence, goals, and possibly consciousness are no longer just properties of the brain." (00:01:12)
5. "The system knows things that none of the parts know." (00:01:43)

Detailed Summary

Episode Overview

This episode weaves together two wide-ranging conversations united by a common thesis: bioelectric and structural/mathematical patterns—not just spikes or molecules—are the true engine of cognition, memory, and biological form. Both segments challenge conventional reductionist frameworks and reframe emergence, consciousness, and physicalism itself.

Guest Backgrounds

• Earl Miller (neuroscientist) began as a physics major for a year, then a premed who volunteered in a neuroscience lab; his dissertation with Charlie Gross was one of the first multiple-electrode recordings in visual cortex, using just 4 electrodes (vs. hundreds today).
• A PDP-11 took two weeks to run principal components analysis on a two-hour recording in Miller's grad-school era, until George Gerson (University of Pennsylvania) built an all-analog PCA system from capacitors, resistors, and a waveform generator that outperformed the most powerful digital computer of the day.
• Michael Levin (bioelectricity/morphogenesis researcher) traces his interest to age 5–6 when his father opened a vacuum-tube TV set and explained how on-screen content varies while hardware stays constant; around 1986 he encountered Robert Becker's bioelectricity book, whose bibliography pointed to foundational 1940s work by Burr.
• Jens (researcher on The Giant Shoulder podcast) builds minimal, purely classical and deterministic computational systems with fully visible algorithms and no hidden mechanisms, and is co-writing a paper with philosopher of causation Lauren Ross.
• The Giant Shoulder is hosted by Evan and promotes a free 26-book neuroscience download.

Core Thesis: Bioelectric Code as Top-Down Control

• The bioelectric code literally precedes and instructs gene expression, cell behavior, and morphogenesis; Levin and collaborators built the first molecular tools to read and write bioelectric pattern information in tissues.
• Altering bioelectric patterns has produced ectopic eyes on tadpole tails and two-headed flatworms, demonstrating the code is instructive rather than epiphenomenal.
• Levin frames bioelectricity as "cognitive glue" enabling cellular collectives to store goal states and memories inaccessible to any individual cell ("no individual cell knows what an eye is, but the system does").

Neuroscience Framework (Earl Miller)

• The human cortex contains 20 billion neurons and roughly 10^14 synaptic connections; controlling 10^14 synapses top-down under a pure connectionist view is computationally intractable.
• An oscillating electric field with a simple wave equation can impose self-organization across a neural network, offering a tractable top-down mechanism.
• Modern systems neuroscience records from hundreds to thousands of neurons simultaneously; subspace coding on ~1,000 neurons can be explained in just three dimensions.
• Spiking cortex activity resembles "birds flocking" rather than "insects buzzing chaotically"; recurrent electrical systems naturally oscillate, and simple nervous systems began oscillating early in evolution.
• Miller contrasts the dominant "telegraph" view (spikes as Morse code) with the view that spikes sit inside an electrical environment of membrane-potential differences that must matter.

Memory, Learning, and Continuity

• Single cells exhibit simple Pavlovian conditioning using only chemical gradients—memory and learning predate nervous systems.
• Planarian memory-transfer experiments: a trained worm's head is removed, it regenerates a brand-new centralized brain, and recall of the original training persists; McConnell did similar work in the 1960s on lower organisms.
• Eve Marder at Brandeis found approximately 100,000 different ways to tune the same three lobster ganglion neurons that still produce identical pyloric motion—an example of "multiple realization."
• Miller argues higher forms of memory (personal recall, baked-in knowledge) still depend on patterns of connections between neurons; if substrate turnover preserves those patterns, the brain functions identically even when physically different a week later.
• Nikolai Kokushkin (NYU) argues individual cells may carry their own goal-related memories; case reports describe personality-trait transfers in heart-lung transplant patients, though Miller considers these "one-offs."
• Humans are born with far more neural connections than they keep; infancy winnows them to reflect statistical regularities like corners and edges, so every brain is wired differently in detail but on the same principles.

Xenobots, Anthrobots, and Novel Organisms

• Anthrobots self-assemble from adult human tracheal epithelial biopsies into proto-organisms that move via cilia, have no nervous system and no muscles, and express about 9,000 genes differently than parent tissue (~half the genome).
• Xenobots are built from frog embryo epithelial cells (led by Viktor Kai in Levin's group, with a pre-print already posted and full paper expected); xenobots express several hundred new genes.
• Unlike embryos, xenobots react to a simple sine wave played through a speaker under the dish and express a sound-perception gene cluster.
• Neurobots (first made by Hale Fawat, now at Harvard) are xenobots with a core of neural cells inside.
• Memory in xenobots is read out via behavior, calcium signaling, and gene expression; the xenobots can be distinguished by which of two different experiences they had.
• Levin's group is testing whether anthrobots made from smokers show nicotine affinity despite lacking a nervous system, and plans tests with kappa opioid agonists on tracheal cells from opiate addicts, plus anthrobot transplantation into rats to read out nicotine-seeking behavior.
• Xenobots exhibit kinematic self-replication, raising the question of when they "paid" the computational cost normally associated with designing a frog or human over evolutionary eons.

Anesthesia as a Field-Level Convergence

• Anesthesia was first demonstrated at the Ether Dome at Massachusetts General Hospital in 1849; after 170+ years doctors still do not know exactly why it works.
• Anesthesia shifts brain dynamics from higher frequencies (associated with cognition) to lower frequencies and desynchronizes the waves.
• Three different anesthetics—propofol, ketamine, dexmedetomidine—act on three different receptors in three different brain regions yet converge on the same field-level effect, which reductionism cannot explain.
• Miller's group developed a closed-loop EEG-based anesthesia delivery system; general anesthesia is linked to short- and long-term dementia, especially in older patients.

Consciousness and Executive Function

• The conscious mind lags about half a second behind reality—so people run from a bear first and become scared later, building episodic memory for future avoidance.
• Most memory is unconscious; consciousness is largely "along for the ride" and mainly builds higher-level episodic/autobiographical memories used for future planning.
• The brain's left-hemisphere narrative is compared to "the history books written by the winner"—consciousness is often the after-the-fact story.
• Miller argues consciousness is overrated for control; executive function is what really matters, and consciousness is "really only good for planning future behavior and stopping yourself from doing something stupid you just decided to do."
• Because humans are the only creatures whose consciousness can be verified by asking them, consciousness in other creatures must be inferred.

Minimal Systems and Xenobot Competency (Jens)

• Jens's program constructs minimal classical/deterministic systems with fully visible algorithms and places them in novel circumstances to detect mismatches between predicted and actual competencies.
• The team deliberately avoids the three standard sources of competence: engineer-written algorithms, evolution across large variants, and prior learning—none of which occurred in their systems.
• The "delta" studied is problem-solving recognizable to any behavior scientist that appears in substrates the standard paradigm would not predict—distinct from complexity or "perverse instantiation."
• Known examples of systems producing more competency than programmed include shells, sunflowers, and Jens's newly-built minimal systems.
• One result is already published, with a couple more expected in summer, all showing systems with more competency than specified.

Frog Development, Bioelectric Networks, and Mathematical Structure

• A simulator tracking frog development from the one-cell fertilized state models ion channels, voltages, and electric fields; homogeneous ion-channel distributions produce symmetry breaking, amplification, and positive feedback yielding bilateral or four-fold symmetry.
• Jens argues the real "why" behind these patterns is mathematical—"the solutions to certain equations have specific forms" (citing an example value of 2.7 versus an alternative of 9).
• Bioelectric networks in the brain or body are particularly good at hosting patterns from a mathematical space, partially tuned by evolution, hosting both static patterns and active inference, Bayesian computation, and memory.

Non-Physical Causation and Physicalism

• Jens rejects the framing that anomalous competencies imply non-physical causes as novel, arguing we have already been in that conceptual land for a long time.
• He argues physicalism "has been dead since the time of Pythagoras and probably long before" because the best explanations of physical phenomena are often mathematical/structural rather than event-based.
• He conjectures math may itself be a behavioral science describing patterns in a latent space, possibly "more real" than physics, and notes the billiard-ball definition of causation has been undermined by quantum theory.

Emergence Debate

• Evan argues emergence is a real property—behavior not explainable from components alone.
• Jens agrees technically but calls conventional uses of emergence "defeatist," contending people use emergence as an explanation to avoid asking "why this pattern versus that pattern," and denying the existence of a latent space of possible patterns.

Disagreements and Pushback

• A co-host challenges the framing, asking why assume a hidden structural layer rather than that we just haven't figured things out in the known realm.
• Applying goal-state, navigation, memory, and learning frameworks from neuroscience to cellular/developmental biology is controversial because many in those fields consider good models to be mechanical, low-level chemistry with no goals.
• About 25–30 years ago, studying electric fields in the brain was treated as "rubbing crystals together" or "magic"; the field is now far more accepted—framed by Miller, Levin, and Evan as a Thomas Kuhn-style paradigm shift, with Harold Saxton Burr (late 1930s–early 1950s) and Donald Hebb and Walter Freeman (1930s–40s) being rediscovered.
• Jens argues standard stories about xenobots are incomplete and not easily quantified, since biology's complexity prevents definitive proof (e.g., unknown quantum microtubule events may always be present).

Communication with Non-Linguistic Systems

• Levin's lab (with Yambu Jang and others) is building language interfaces for non-linguistic biological structures—"talking to things that normally don't talk."
• They are applying anesthetics and hallucinogens to xenobots and anthrobots (no neurons) while recording calcium activity; they do not yet have a DMT permit.
• In simulations of non-neural cells, disrupting gap junctional connections often prevents the system from returning to the same state afterward.

Mathematics, Platonism, and Evolutionary Constraints

• Mathematical truths—the four color theorem, halting problem, exact value of e, Feigenbaum's constant, sphere-packing rules—are independent of physical constants and cannot be altered by changing Big Bang parameters.
• Cicadas emerge at 13- and 17-year cycles specifically because those are prime numbers, helping them avoid predators that time populations to non-prime cycles.
• The brain-body system is not isolated; metabolism heavily affects brain function, so memory could plausibly live in the body, not the brain alone, and selection pressure from the birth canal constrains human brain size.

Clinical and Practical Implications

• Documented clinical cases show normal or above-normal intelligence despite radically reduced brain function (e.g., hydrocephalus, half the brain missing on MRI); Miller attributes this to early developmental plasticity, since equivalent adult damage would not yield normal function.
• Anesthesiologists currently monitor heart rate, respiration, and blood pressure rather than directly measuring the brain's electrical signatures of unconsciousness, which Miller's closed-loop EEG system aims to fix.