What Biology Already Knows
If biology knows when to stop, what exactly does it already know?
The question after probability
Stopping is the deeper act.
Growth is easy to imagine as an instruction. Completion is harder.
“Most Likely” asked how one continuation becomes history. This essay begins one step later. A living body does not merely continue. It develops, repairs, scales, and then—at some precise, astonishing point—stops.
A salamander does not grow a limb forever. A flatworm fragment does not keep adding heads and tails. An embryo does not remain a field of multiplying cells. The activity ceases when an anatomy has been achieved. That fact is so familiar that it can disappear inside the word development. But stopping is not the absence of development. It is development’s most exact judgment.
To stop correctly, a system must distinguish incomplete from complete. It must register not only what to do next, but whether anything remains to be done at all. The mystery is not simply how cells build. The mystery is what lets them recognize enough.
What Levin’s experiments show
The route can change while the architecture survives.
Michael Levin and his collaborators study how cells coordinate growth and form through biochemical, genetic, mechanical, and bioelectrical communication. Their perturbation experiments make the hidden flexibility of development visible.
01 / Corrected faces
Misplaced parts take unfamiliar paths.
In Xenopus tadpoles, researchers induced craniofacial abnormalities and tracked jaws and other structures over time. The parts did not simply replay their ordinary movements from shifted starting points. They moved along different axes and gradually normalized their shapes and relative positions. Development recovered the face without recovering the usual route.
02 / Triggered tails
A brief signal releases a whole sequence.
During a period when tadpole tails normally fail to regenerate, a short manipulation of sodium transport could initiate regeneration of a complex tail—with multiple tissue types arranged at the right scale and orientation. The intervention did not specify where to place every nerve, vessel, or muscle. It altered a high-level physiological condition, and the tissue completed the rest.
03 / Rewritten worms
The same genome can carry a different completion.
In planaria, temporarily disrupting bioelectrical communication can produce two-headed animals. After the treatment is gone, later fragments can regenerate the altered form again in plain water. Other bioelectric interventions can reset normal one-headed anatomy. The DNA sequence did not change; what changed was the pattern the collective tissue would rebuild.
A small thought experiment
Perturb the sequence.
The intermediate states move. The terminal criterion does not. This is a visual analogy, not a model of the biology.
Try the control. The path deforms; the last state remains at the same boundary.
Not a little picture in the cells.
To say that biology “knows” an anatomy is not to say that a cell contains a miniature diagram of the finished animal.
Levin’s language is deliberately drawn from computation and cognition because the behavior being studied is distributed, adaptive, and sensitive to error. Cells exchange electrical and chemical information. Tissues respond to differences between local conditions and larger patterns. The genome supplies essential components and constraints. Bioelectrical networks help coordinate what those components do together.
None of this requires a tiny architect hiding inside the embryo. Nor does it show that tissue foresees the future. The experiments establish something more disciplined and, in its way, more surprising: anatomy is not fully explained by a rigid procession of local steps. A living collective can arrive from the wrong starting position, compensate for disturbance, scale its work, and cease when a coherent form has been restored.
If biology were merely following a recipe, why would it recover the destination instead of the steps?
A recipe is loyal to sequence. Change the order, remove an instruction, move an ingredient, and the failure should propagate. Yet the perturbed tadpole face does not insist on repeating its ordinary movements. It finds other movements. The triggered tail does not wait for a scientist to enumerate every downstream construction. The planarian’s regenerative pattern can be rewritten without rewriting its genome.
These results do not abolish molecular mechanism. They demand a richer account of it. The relevant cause may be a network capable of holding large-scale pattern information, comparing present anatomy with that pattern, and recruiting many lower-level processes until the difference is reduced. “Architecture” names the coordination that a list of parts cannot name by itself.
The boundary
Levin’s experiments show robust correction and reprogrammable pattern memory. They do not establish the Immutable Past. Here the science ends, and my philosophy begins.
The future cannot contain the finished body.
The future is pure potential. It contains every continuation that remains available and no completed thing. A future frog is not hidden somewhere ahead, pulling cells toward itself. Until an event occurs, it has no settled existence. It may be likely, constrained, or nearly inevitable. It is still potential.
Completion belongs to the Immutable Past.
That sentence changes the direction of the puzzle. Development is not a search through the future for a destination that already exists there. It is an execution toward a completion already defined. The definition is not a later object exerting a backward force. It is the criterion by which the execution can become one finished history rather than endless activity.
The future offers paths. The architecture limits which paths can count as completion. The living system can tolerate detours because the detour is not the identity of the process. A jaw may move along an abnormal axis. A tail may begin after an artificial electrical trigger. A regenerative collective may even be given a different terminal pattern. What matters is not fidelity to a prescribed journey. What matters is whether the unfolding can close.
Definition is not prediction.
A prediction ranks what is likely to happen next. A definition states what the completed thing must be to count as this thing. The distinction is crucial.
In “Most Likely,” nature appeared as a prediction machine: it overwhelmingly converts the highest-probability continuation into history. Development adds another layer. Probability can favor the next cellular move, but probability alone cannot explain why a sequence ends at an organized body instead of merely continuing through more probable moves. Something must constrain the series as a whole.
That something need not be a conscious intention. It can be embodied in the relations among cells, voltages, genes, tissues, forces, and geometry. But whatever its mechanism, its work is architectural. It preserves the difference between one more step and completion.
This is why the planarian experiments are philosophically severe. Change the bioelectrical state and the same genome may execute toward a different body plan. The intervention does not merely make the old construction go wrong. It can change what the tissue treats as finished. Biology’s “knowledge,” then, cannot be reduced to a parts inventory. It includes a criterion for the whole.
Development is closure under disturbance.
If the ordinary path were sacred, perturbation would only produce failure. Instead, living systems often display a more interesting loyalty. They abandon the familiar route to preserve the architecture. They improvise locally in order to finish globally.
This is not freedom from constraint. It is constraint operating at a higher level. The system can vary the means because the completion remains defined. It can absorb surprise because it does not mistake its most recent state for its final form.
And when the form is complete, the work enters the only place completion can exist: the Past. What had been potential becomes immutable. The cells do not arrive at an object waiting in the future. Their coordinated action makes a completion historical.
Research notes
Vandenberg, Adams & Levin (2012)
Quantitative study of craniofacial normalization after induced perturbation in Xenopus tadpoles.
Tseng et al. (2010)
Experimental induction of complex tail regeneration through a transient sodium current.
Durant et al. (2017)
Long-term editing and resetting of planarian regenerative anatomy through bioelectric perturbation.
The summaries above are mine. “Know” and “memory” describe observable regulatory capacities at the level of living systems; they should not be read as claims that individual cells possess human-like awareness.