Planaria Memory & Xenobots: Detailed Breakdown
PART 1: PLANARIA AND MORPHOLOGICAL MEMORY
Background: Why Planaria?
Planaria (Dugesia japonica and related species) are freshwater flatworms with an almost absurd capacity for regeneration. Cut one into 200 pieces, and you get 200 new worms. Every fragment — even a tiny tail segment with no brain tissue whatsoever — will regenerate a complete, properly oriented animal.
This alone is remarkable. But Levin's lab discovered something far stranger.
The Classic Setup: Polarity and Bioelectrics
Every planarian has a defined anterior-posterior axis — a head end and a tail end. This polarity is maintained by a gradient of bioelectric signals across the body, particularly involving:
- Gap junctions — protein channels that directly connect neighboring cells, allowing ions and small molecules to flow between them
- Ion pumps (especially V-ATPase, an H⁺ pump) that create voltage differentials across tissues
- Neurotransmitters like serotonin that act as patterning signals even outside the nervous system
The key insight: this bioelectric pattern is not just a readout of genetic instructions. It is itself instructive — it tells cells where they are and what they should become. It functions like a map or a blueprint held in electrical form across the tissue.
The Experiment: Erasing and Overwriting the Memory
Here is what Levin's lab did, step by step:
Step 1 — Cut the worm A planarian is cut transversely, producing a head fragment and a tail fragment. Normally:
- The head fragment regenerates a new tail at the wound site
- The tail fragment regenerates a new head at the wound site
This is normal polarity — the body "knows" which end is which.
Step 2 — Pharmacological intervention during regeneration While the tail fragment is regenerating (the critical window is the first ~24 hours after cutting), researchers bathe it in octanol — a drug that blocks gap junctions, severing the bioelectric communication between cells.
What happens? The tail fragment, unable to read its normal bioelectric positional information, sometimes regenerates a head at both ends — a two-headed worm. The chemical disruption confused the tissue's sense of axis.
Step 3 — The crucial discovery: the memory persists Here is where it gets extraordinary. After the two-headed worm is fully regenerated, researchers:
- Remove the drug (restore normal chemical environment)
- Allow the worm to live normally for weeks
- Then cut it again — no drugs, no intervention
A normal worm, cut in this state, would regenerate with correct polarity. But these worms regenerated two-headed worms again — even though no drug was present during the second regeneration.
The altered bioelectric state had become self-sustaining. The tissue had "learned" a new body plan and perpetuated it across generations of regeneration without any genetic change and without ongoing chemical intervention.
What Is Actually Being "Remembered"?
This is not memory in the neural sense — there are no neurons storing an engram. What's happening is:
Bioelectric state as memory: The pattern of ion flows, membrane voltages, and gap junction connectivity across the tissue constitutes a stable attractor state — a self-reinforcing electrical pattern that cells maintain and propagate when they divide and communicate.
Think of it like a standing wave in a medium. The wave pattern persists not because any single molecule holds it, but because the dynamics of the system continuously recreate it. When you cut the worm, the new cells inherit the electrical environment of their neighbors and reproduce the same pattern.
The "Body Plan Morphogenetic Field": Levin frames this in terms of a morphogenetic field — a distributed representation of target anatomy encoded not in DNA sequence but in the bioelectric state of the tissue. DNA provides the parts list (what proteins are available); the bioelectric field provides the blueprint (how those parts should be assembled).
The planaria experiments show this blueprint can be:
- Altered (by disrupting gap junctions)
- Stabilized in its new form (self-reinforcing dynamics)
- Transmitted across regenerative events (new cells adopt the electrical state of existing cells)
- Maintained without ongoing external signals
The Deeper Implication
This means the worm's body plan is not simply computed fresh from the genome each time. There is a second layer of information — above the genome, encoded in bioelectric dynamics — that stores and propagates the target morphology.
Levin uses the analogy of computer hardware vs. software: the genome is the hardware (fixed, inherited), but the bioelectric pattern is the software running on top — and that software can be edited, corrupted, and rewritten independently of the hardware.
This is teleological in a deep sense: the tissue is not just following chemical gradients passively. It is actively comparing its current state to a stored target state and making corrections. When that target state is altered, the corrections drive toward the new target.
PART 2: XENOBOTS — LIVING MACHINES AND KINEMATIC SELF-REPLICATION
What Are Xenobots?
Xenobots were first reported in 2020, created collaboratively by Levin's lab at Tufts and Josh Bongard's computational lab at the University of Vermont.
They are made from Xenopus laevis — the African clawed frog — specifically from ectodermal cells (cells that would normally become skin) harvested from early embryos.
Crucially: these cells are removed from their normal developmental context. They are no longer receiving the chemical signals that would tell them "you are part of a frog embryo, become skin." Freed from that context, they exhibit novel, emergent behaviors that no frog has ever exhibited — and that were never programmed.
Stage 1: Self-Assembly and Spontaneous Organization
When ectodermal cells are harvested and placed in a dish as a loose aggregate, something remarkable happens over ~24 hours:
- The cells do not die (as isolated cells typically would)
- They reorganize — moving, adhering, and arranging themselves into a compact, roughly spherical or ovoid body
- Cells with cilia (hair-like projections, normally used to move mucus across skin) reorient those cilia outward, creating a coordinated propulsion system
- The resulting structure, roughly 500–700 micrometers across, moves through fluid in a directed way
No blueprint was given. No morphogen gradient was applied. The cells used their existing genetic toolkit — developed for one purpose (being frog skin) — to build a novel functional organism suited to their new context.
This is a profound demonstration of context-dependent agency: the cells are not executing a fixed program, they are solving the problem of "what should I be?" given their current situation.
Stage 2: The AI-Designed Body Plans
In the 2020 paper, Bongard's team used an evolutionary algorithm running on a supercomputer to design optimal xenobot body shapes for specific tasks (like moving in a particular direction or pushing objects).
The process:
- Simulate thousands of random arrangements of frog cells in silico
- Evaluate which arrangements best accomplish a target behavior
- Select the best performers, mutate them, repeat
- Take the winning design and build it in real life using microsurgery on real frog cells
Strikingly, the real xenobots closely matched the behavior of their simulated counterparts — confirming that the cells were indeed executing coherent, predictable behaviors, not random activity.
But the spontaneously self-assembled xenobots (without AI design) were arguably more interesting, because they required no external design at all.
Stage 3: Kinematic Self-Replication (2021)
The 2021 PNAS paper reported something that genuinely shocked the scientific community.
The setup: Researchers placed xenobots in a dish containing a large supply of loose, dissociated frog cells — essentially raw cellular material.
What happened: The xenobots spontaneously began to gather the loose cells, sweeping them together into piles using their ciliary motion.
Those piles of gathered cells then self-assembled into new xenobots.
Those new xenobots could, in turn, gather more loose cells and produce another generation.
This is kinematic self-replication — replication through motion and behavior rather than through chemical template copying (like DNA replication). It is an entirely different category of reproduction from anything previously documented.
Why "Kinematic" Is the Key Word
In biology, replication almost always means template-directed chemical copying: DNA unzips, complementary bases are added, you get two copies. The information is in the molecule.
Kinematic self-replication is different: the information is in the behavior and shape of the organism. The xenobot "replicates" not by copying its molecules but by doing things in the world — moving, gathering, organizing — that result in a new entity similar to itself.
This is closer to how a whirlpool "reproduces" (by creating conditions for new whirlpools) than how a bacterium reproduces. But xenobots are vastly more complex and directed.
Von Neumann had theorized kinematic self-replicators in the 1940s as a thought experiment. Xenobots are the first biological instantiation.
The Shape Discovery: C-Shape Matters
A critical finding from the 2021 work: the default spherical xenobots replicate poorly — they gather some cells but the process is inefficient and quickly dies out after 1–2 generations.
When the AI evolutionary algorithm was again applied — this time to optimize for replication efficiency — it discovered that a C-shaped or pac-man-shaped xenobot was dramatically more effective at gathering and corralling loose cells into the interior of the C, where they could aggregate and self-organize.
This C-shape does not exist in frog biology. It was discovered by an algorithm, then built, then confirmed to work. The frog cells — never having "intended" to replicate this way — executed the replication faithfully when given the right body shape.
The Philosophical Payload
The xenobot work raises several deeply unsettling and important questions:
1. What is an organism? Xenobots are made of frog cells, carry frog DNA, but are not frogs. They have never existed in nature. They exhibit goal-directed behavior (moving, replicating) that no frog exhibits. Are they a new species? A new category of being?
2. What is reproduction? Xenobots don't pass on genetic information to their "offspring" — the daughter xenobots have the same genome as the parent, inherited from the original frog embryo. What's being transmitted is form and behavior, not genetic novelty. This blurs the line between reproduction and construction.
3. Where does the "goal" come from? The xenobots were not programmed to gather cells and replicate. This behavior emerged from the interaction of cell-level properties (cilia, adhesion, motility) with the physical environment. The goal-directedness is real — measurable, reproducible — but it wasn't designed in. It emerged.
Levin argues this shows that agency and teleology can be genuine properties of matter — not imposed from outside by a designer or genome, but arising from the dynamics of sufficiently complex, self-organizing systems.
Connecting Planaria and Xenobots: The Unified Theme
Both experiments point to the same conclusion:
| Planaria | Xenobots | |
|---|---|---|
| Substrate | Adult flatworm tissue | Embryonic frog cells |
| Context change | Chemical disruption of polarity signals | Removal from embryonic context |
| Response | Adopt and maintain new body plan | Self-assemble novel body plan |
| Memory/Goal | Bioelectric target state | Emergent morphogenetic goal |
| Key finding | Morphological memory above genome | Goal-directed behavior without programming |
In both cases, cells are doing something more than executing a genetic program. They are reading their environment, computing a response, and pursuing a target state — even when that target state has been altered or is entirely novel.
This is Levin's core claim made flesh: biology is cognitive all the way down, and the gap between "mere matter" and "minded agent" is not a gap at all, but a continuum we are only beginning to measure.
Posts
