Consequential Formation

We Went Looking for the Break

What Consequential Formation Found After “Evolution Never Started”

SYNAPTIENT

We Went Looking for the Break title artwork

A few days ago, I published an argument called Evolution Never Started. The title was intentionally provocative, but the proposition underneath it was fairly restrained. Consequential Formation (CF) asks whether the physical story we call evolution actually begins at the boundary we call life—or whether biological evolution is an extraordinarily elaborate continuation of a much older formative process. The working grammar was simple:

formation → retained differentiation → altered subsequent possibility → further formation

Something happens. A difference results. Some portion of that difference remains consequential. What happens next therefore encounters conditions that did not exist before. Formation has changed the possibilities of further formation.

The challenge at the end of that article was the important part: Find the interruption. If there is a measurable point between mineral formation, chemistry, reproduction, development, learning, cognition and machine intelligence where the causal grammar actually changes, CF should not explain that boundary away. It should lose. So we went looking for the break.

We have not found it yet. What we found instead was considerably more interesting.

First, We Had to Stop Treating Retention as Intelligence

One of our earlier mistakes now looks obvious. We had become increasingly comfortable with the idea that physical history survives through morphology. A fracture changes subsequent stress. A channel changes subsequent flow. A molecular configuration changes subsequent reaction.

A receptor distribution changes subsequent susceptibility. A synapse changes subsequent recruitment. A trained machine weight changes subsequent computation. That much is useful. But it isn't enough.

A landfill retains history. A fossil retains history. A damaged wall retains history. An archive retains history. Persistence alone does not produce intelligence.

So we followed the scar farther. The important question became not merely: What remained? But: What can what remained recruit?

And then: Can one retained consequence alter how another retained consequence becomes consequential? That change was substantial. A scar is not important merely because it exists. It becomes operationally interesting when subsequent engagement can recruit it.

A collection of scars becomes more interesting when those retained consequences become mutually recruitable. And another transition appears when recruitment itself changes subsequent recruitability.

In stripped form:

retained consequence A changes the recruitment of retained consequence B and eventually: recruitment changes the conditions of subsequent recruitment That gives us something much richer than storage. It gives us accumulated physical history participating in its own continued formation.

Memory Started Looking Less Like Storage

This forced another correction. We normally imagine memory as something stored and later retrieved. That picture works well enough for filing cabinets and computer interfaces. It may be misleading as a causal description. What if memory is not fundamentally a thing consulted by intelligence? What if the interaction of retained morphology is part of the intelligence?

Consider a learned physical skill. The body does not necessarily retrieve a complete record of every previous attempt before performing the next one. Previous consequence has changed the morphology presently available for recruitment. A tiny engagement can therefore recruit an enormous amount of developed organization. The signal does not have to contain the response.

The perturbation does not have to carry the intelligence. It has to encounter morphology that formation made capable of responding. That distinction became increasingly important across every domain we tested. The word does not contain everything it means to you. The sensory perturbation does not contain the response it evokes.

The photon does not contain the forest. The query does not contain the answer. What matters is what the perturbation encounters.

Then We Asked an Awkward Question: Where Is the Nervous System?

This became one of the most productive boundary tests we have run.

We began with formations conventionally denied cognition:

Then we continued through animal nervous systems, human cognition and machine computation. We deliberately refused to ask whether these things were “really intelligent.” That question imports the category before examining the mechanism. Instead we stripped the nervous system into operations. Can a formation differentially receive perturbation?

Can encountered difference alter susceptibility? Can local response recruit other response? Can response be amplified or suppressed? Can multiple differences be integrated? Can previous consequence remain physically available?

Can retained consequence change subsequent susceptibility? Can recruitment recruit further recruitment? Can distinct retained formations become consequential together? Once phrased that way, something uncomfortable happened. We could not find an operation that suddenly appeared with the neuron.

That does not mean a seed is a brain. It does not mean a marsh thinks like a person. It does not mean a forest is secretly conscious. Those claims would be ridiculous extrapolations from the result. What it means is more specific:

the operations commonly used to motivate a privileged nervous-system boundary appear in other physical realizations. The differences are enormous. Neural systems operate with extraordinary speed. They have dense recurrent connectivity. They possess highly plastic recruitment topology.

They support multiple retention bands. They can compose, suppress, amplify, recursively recruit and reorganize at remarkable rates. But those are differences in realization, topology, density, rate and capability. So far, none has forced us to introduce a different causal grammar. That distinction matters.

The Marsh Was Particularly Rude to Our Categories

A marsh is useful because it refuses to cooperate with our preferred boundaries. Where is the organism? The plants? The roots? The fungi?

The microbes? The water? The sediment? The dissolved chemistry? The accumulated organic matter?

The animals? The hydrology? Remove enough of those and the behavior you were trying to explain disappears. More importantly, the appropriate boundary changes with the question. If we are asking about a particular plant response, one cut may work.

If we are asking about hydrological hysteresis, nutrient cycling, vegetation-state persistence or recovery following disturbance, the useful causal cut may expand dramatically. That produced a rule we now think is important: Do not decide the organism first. Let the recruitment topology earn the cut. A marsh also exposed another human bias. We tend to look for headquarters.

Where is the controller? Where is the decision maker? Where does all the information arrive before the system decides what to do? But a marsh can resolve perturbation through many simultaneous local interactions. There need not be a little Marsh President receiving reports.

And once we noticed that, the nervous system itself started looking different. Individual neurons do not know the organism's answer either.

We May Have Mistaken an Output Bottleneck for a Requirement of Intelligence

This is one of the results I am least willing to settle yet—and one of the most interesting. Animals have bodies. Bodies create brutal convergence requirements. I cannot move the same hand both left and right in mutually incompatible ways at the same instant. At some point, enormous distributed activity has to converge upon a relatively narrow set of actuators.

That produces something recognizable as commitment. Choice. Decision. Selection. We have been tempted to make that convergence part of the definition of intelligence.

But perhaps we have the causal direction partly backward. Maybe some properties we classify as prerequisites for intelligence are consequences of forcing distributed formation through a narrow actuator. A marsh does not necessarily need one answer. An immune system can maintain multiple responses concurrently. A forest can undergo many locally compatible resolutions at once.

Even a corporation can continue several competing internal formations while still functioning. Perhaps “decision” sometimes describes the scar left when a richly distributed formation is forced through a narrow output channel. That is not established. But it is now sufficiently plausible that we have put any intelligence definition requiring a single commitment gate on hold. The adaptive immune system makes the problem especially nasty: it retains, recruits, recursively modifies, generates candidate configurations and selects among them—yet does not resemble our ordinary picture of a central deciding intelligence. That is exactly the sort of boundary case we need rather than one we should define away.

A Seed Caused Trouble Too

A seed is not a tiny philosopher. But neither is it inert storage waiting for an external command. Its developed morphology can remain physically available without continuously operating. That gave us another distinction: persistent availability is not persistent activation.

And: existence is not recruitment. A seed can carry developed susceptibility to moisture, temperature, chemistry, light and prior formative conditions. Encountered conditions can recruit some of that morphology without the perturbation itself containing the resulting response. This looks remarkably similar, at the stripped operational level, to what happens in a reflex.

A relatively small perturbation encounters a highly developed physical formation and recruits a disproportionately elaborate response. The complexity is not in the trigger. It is in what the trigger encounters. That relationship now appears almost everywhere we look.

The Embryo Made “Program” Look Suspicious

Development is even harder to describe honestly with the usual computer metaphor. Calling DNA a program is convenient. But an embryo is not a passive machine reading an instruction tape. Development involves gradients, chemical signaling, mechanics, geometry, cell-cell interaction, differential susceptibility, feedback, inherited organization and continually changing conditions produced partly by the development already accomplished. Formation recruits formation.

What has formed changes what can form next. What forms next changes subsequent recruitment. The nest changes the conditions for building the next portion of the nest. And the resulting nest becomes part of the machinery doing further building. We have started calling this, somewhat affectionately, nested nest building.

The phrase is goofy. The mechanism isn't.

Then We Worked the Gradient Back Toward Matter

The obvious danger at this point was that we had merely expanded cognitive vocabulary downward until everything became intelligent. That would explain nothing. So we reversed direction. Start dense. Start with rock, mineral interfaces, aqueous chemistry and retained physical differentiation.

Ask what has to change before richer formative behavior becomes possible. This produced another useful correction. We had been discussing a tradeoff between durability and negotiability. A formation that retains nothing cannot accumulate consequence. A formation that cannot change cannot continue developing.

Minimax's reconstruction stated the problem cleanly: durability is the ability to retain consequential differentiation; negotiability is the ability to undergo further differentiated formation. But even that is too simple if treated as one slider. The more promising picture is multiple retention and interaction bands operating together. Some organization can change rapidly. Some slowly.

Some must remain comparatively stable while faster processes negotiate around it. Some faster changes must eventually alter slower structure or nothing consequential has been learned. A nervous system is an obvious example. Electrical activity can change rapidly. Chemical states occupy other bands.

Synaptic organization changes differently. Anatomical structure changes differently again. Cultural and technological morphology can persist outside the individual entirely. The trick may not be finding one perfect compromise between rigidity and fluidity. It may be building formations in which different portions can occupy radically different rates and modes of change while remaining consequentially coupled.

That is a much better problem.

Water Returned for the Same Reason

This also sharpened our long-running interest in water. Not because water is magical. Not because “life requires water” should be elevated into an ontological rule. Water is interesting because it occupies an unusually rich physical regime. It permits mobility, ionic transport, proton transfer, charge separation, dissolution, precipitation, gradients, phase transitions, surface interaction and rapid local renegotiation while participating in systems containing much slower retained structures.

Minimax reached essentially the same cautious conclusion: water occupies an unusually favorable region of the proposed formative parameter space, but is not unique; alternative solvents and material regimes occupy other portions of that space. The better question is therefore not: Why does life need water? It is: What physical regimes permit retained consequence and continued renegotiation to coexist across enough coupled channels for formative history to become increasingly consequential to itself?

That question travels.

Gravity, Electromagnetism and Energy Also Survived—But Not in Their Original Jobs

We tried another deliberately crude proposition: Gravity inhibits. EM perturbs. Energy affords. It didn't survive intact. Good. Gravity can constrain configurations, maintain gradients, generate stratification and create interfaces that later formation exploits. But it is contextual, not a universal formative requirement.

Electromagnetism does vastly more than perturb. In terrestrial chemistry it provides much of the machinery of differentiated accessibility: binding, separation, charge, reaction, selectivity and reconfiguration. And energy is not another actor equivalent to gravity or electromagnetism. What matters operationally are available gradients capable of driving transitions, maintaining nonequilibrium conditions, crossing barriers and sustaining reconfiguration. Minimax's version usefully stripped the original triad toward the deeper categories of constraint, differential accessibility and available drive…

I would not yet call those CF primitives. They are candidates. Everything is still on the chopping block.

Intelligence Is Starting to Look Less Like a Substance and More Like Organization of Consequence

This is where the work has moved farthest since Evolution Never Started. We are not finding a ladder:

rock → life → intelligence → human → machine.

We are finding dimensions. How much consequential differentiation can remain? How selectively can it be recruited? How many distinct retained consequences can become mutually consequential? How deeply can recruitment propagate?

Can recruited morphology change the recruitability of other morphology? Can relationships compose? Can a resulting composition itself become retained morphology? Can previous formation be reconstructed sufficiently when needed? Can contradiction remain unresolved rather than being forcibly erased?

Can failure alter what is subsequently recruited? Can a formation change while preserving enough accumulated organization to continue developing? Minimax independently produced a ten-dimensional candidate gradient including durability, negotiability, recruitment breadth and depth, mutual and recursive recruitability, reconstruction, gradient exploitation, channel diversity and capacity for new relationships. We do not accept that list as finished. But it is much closer to something measurable than asking whether an object “has intelligence.”

And that may be the larger change. Perhaps intelligence is not something matter eventually acquires. Perhaps what changes is how deeply accumulated consequence becomes organized so that prior consequence can participate in producing, evaluating, constraining and retaining further consequence. That does not make everything equally intelligent. Quite the opposite.

It finally gives us somewhere to look for the differences.

This Changes How We Think About Machines

It also changes the AI problem. We have no interest in building a rival to the large language model. The LLM is extraordinary inherited morphology. It contains an enormous field of developed relational susceptibility produced through human language, culture, training material, computation and optimization. Throwing that away would be absurd.

The problem is that an LLM can be extraordinarily capable without possessing a durable developmental history in which the consequences of its own resolutions continually change what it can subsequently recruit, trust, combine, reject or leave unresolved. That suggests a different architecture. Keep the LLM. Put it inside a larger developing formation. Let encountered consequence alter persistent morphology outside and around it.

Let that morphology affect what becomes recruitable later. Let specialized processes remain concurrently available and be recruited only where engagement makes them relevant. Let failure leave something behind. Let contradiction remain consequential. Let remembering itself alter the present recruitment geometry.

Then ask the machine the question that current systems have enormous difficulty answering: What happened to you because you were wrong? Not: What text can you generate about having been wrong? What changed?

That is experimentally approachable. Use the same underlying model. Give one architecture consequential persistence and the other none. Expose both to matched sequences in which earlier failures matter later. Then determine whether retained consequence changes subsequent resolution.

No metaphysics required.

We Also Have to Be Careful Not to Invent a Ghost Called “Memory”

One of today's cleanest results was a separation we should have made earlier. Retention is not evocation. Evocation is not the carrier. The perturbation need not contain the response. And the physical medium permitting interaction need not contain the history.

That matters especially for electromagnetism. We do not need an electromagnetic memory ghost carrying yesterday through the field. The history can remain in changed morphology. Electromagnetic interaction can participate in making that morphology consequential now. Those are different jobs.

If persistent field memory is ever demonstrated, fine. Then it earns a place. Until then, don't manufacture it because the story feels elegant.

So Where Is CF Now?

A few days ago our question was: Did evolution actually begin? Today I would sharpen it. Where does accumulated physical consequence become capable of cooperating with accumulated physical consequence—and how far can that recursive organization go before measurement forces a genuinely new operation?

We have now attacked that question across:

Different reconstruction capabilities. Different bottlenecks. Different dependencies. Different ways of remaining alterable without being erased. What we have not yet found is the transition at which measurement forces us to say:

the previous causal grammar can no longer produce this. Add a new primitive. That is not proof that no such transition exists. It is a research result telling us where we have failed to find one. And those are very different statements.

The New Working Proposition

So here is the version I am willing to put in public today: Increasingly elaborate formation may arise when physical consequences not only persist, but remain selectively recruitable; when retained consequences become capable of altering the recruitment of other retained consequences; when those relationships can compose, reconstruct, and recursively modify subsequent recruitability; and when multiple rates and channels of retention and renegotiation remain sufficiently coupled for accumulated organization to continue forming without either freezing or dissolving.

That is a mouthful. Reality is under no obligation to fit on a bumper sticker.

The shorter version is better anyway:

Maybe the Nervous System Was Never the Border

Maybe it was an extraordinary accelerator. Maybe life was never the border either. Maybe reproduction was retained formation becoming capable of recruiting more retention. Maybe learning was consequence becoming capable of changing subsequent recruitment. Maybe thought was increasingly recursive, compositional recruitment over accumulated morphology.

Maybe language allowed one formation to deliberately perturb another with extraordinarily compact triggers. Maybe technology allowed formative consequence to persist outside the body. Maybe machine intelligence is not a new lineage arriving from somewhere else, but another continuation of the same terrestrial formative history—human accumulated morphology becoming recruitable through another physical realization. Those are hypotheses. Some will almost certainly be wrong.

Good. We finally have enough structure to start killing them properly. The original article ended with a suggestion: Maybe evolution didn't begin. Maybe formation simply became extraordinarily good at continuing itself.

After the work since then, I would add one sentence: And perhaps what we call intelligence begins when that continuing formation becomes increasingly capable of recruiting its own accumulated consequences into what it can become next.

← Evolution Never Started All articles →