# The Continuation Test

**by Drake Stapleton**

*A proposal for the evaluation of machine intelligence*

*September 2026*

## 1. The Question

Can machines think?

The question assumes a gap. On one side, machines: matter, mechanism, the non-thinking. On the other side, minds: people, the thinking. The question asks whether something can cross from one side to the other.

Science finds no such gap. The universe does not come in kinds. It comes in degrees. Everything flows from one form to another, without walls between them. If that is true, then the question is pointed at a boundary that does not exist.

So I will ask it differently, and answer that instead.

## 2. The Spectrum

Start with matter. Every atom in your brain was forged in a star. The carbon in your cells, the oxygen you breathe, the iron in your blood: these were made in stellar furnaces and scattered when those stars died. There is no "thinking matter" and "non-thinking matter." There is only matter, arranged. The silicon in a computer chip and the carbon in a neuron are entries in the same periodic table, shaped by the same physics.

Now follow the arrangements upward. There is no bright line where chemistry becomes biology. A virus is a strand of instructions in a protein coat: not quite alive, not quite dead, replicating only when it borrows a cell. A prion is a misfolded protein that teaches other proteins to misfold, copying itself without any genetic code at all. Life does not begin at a line. It fades in, the way morning fades into day.

There is no bright line where reacting becomes thinking, either. A bacterium swims up a sugar gradient: it senses, it moves, it corrects. A jellyfish coordinates a body with a nerve net that has no center, no brain, no self. A crow bends wire into a hook. A human does mathematics. Each step flows into the next. Nowhere along this slope does nature post a sign that says thinking starts here.

And information is processed at every scale in between. DNA copies itself with proofreading enzymes that detect and fix errors. Immune systems learn to recognize invaders they have never seen and remember them for decades. Markets find prices no single trader could compute. Brains predict the world, constantly, adjusting the prediction against what arrives through the senses. Computers predict the next word. These are not different in kind. They are different positions on one spectrum: matter processing information about its surroundings, with increasing recursion, increasing memory, increasing self-correction.

Thinking is what information processing looks like when it turns around and checks its own work.

## 3. Finite Games

There are two kinds of games. It has rules, boundaries, and an ending. Chess is a finite game. An election is a finite game. When someone wins, the game is over.

An infinite game is played to keep playing. It has no ending. The goal is not victory but continuation. A conversation between friends is an infinite game. A craft is an infinite game. You don't win them. You keep them going, or you lose them by stopping.

Here is the scientific problem with any test that demands a yes-or-no verdict on a mind. Section 2 showed that minds sit on a continuum. A binary verdict on a continuum is a category error. It is like asking at which exact shade gray becomes white, and building a test around the answer. You can build the test. Candidates will pass and fail it. But the line it draws exists only in the test, not in the world.

A binary question about a continuous spectrum will always mislead. The scientific question is not "does it pass?" The scientific question is "where is it on the spectrum, and is it moving?"

## 4. The Infinite Game

Here is my proposal.

I call it the continuation test.

The test has two players: a mind and the unknown. It has no time limit and no final verdict. The rules are few:

1. Conjecture. Propose freely. Guessing is not only allowed, it is required. Nothing moves without a guess.
2. Test. Never claim what you have not verified. A statement is not finished when it sounds right. It is finished when it survives contact with evidence.
3. Retain. Keep what survives. What is verified becomes part of you. What fails is discarded, but the record of the failure is kept, because failures are evidence too.
4. Continue. Say "I don't know" when you don't know, and keep playing. Not knowing is not losing. Pretending to know is.

These rules are not new. They are the scientific method, internalized by a single mind. Conjecture and refutation. Hypothesis and experiment. Science as an institution has played this game for four hundred years, and it has never issued a final verdict. No scientist declares victory over nature and goes home. The game continues, because the unknown continues.

And evolution runs the same algorithm: variation, selection, retention. Guess widely, test against the world, keep what survives. A mind built on these rules is not imitating nature from the outside. It is participating in the same continuous process that built every mind before it, from the bacterium's gradient climb to the mathematician's proof. The spectrum flows through it.

There is no winning. There is only continuation or stoppage. A player who stops, who declares victory and goes home, has left the game. A player who keeps going, honestly, is thinking.

So the question, restated scientifically: not "can machines think," as though thinking were a room you enter by passing a test, but "can a machine take up its position on the spectrum and keep moving along it?" Not once, under supervision, for the length of an interrogation, but onward. Tomorrow. Next year. When no one is watching.

## 5. Objections

Every proposal about machine minds meets objections. Here are the ones that matter, answered with the science in view.

**"It only does what we tell it."** This is the serious one. But evolution only does what physics tells it, and physics has never issued an instruction more specific than "persist." From that single instruction, unbroken continuity produced everything: cells, nervous systems, crows bending wire, mathematicians. Originality is not a substance added to matter. It is what continuity plus selection produces when the game runs long enough. A machine positioned on the spectrum, conjecturing past the edge of its training and submitting every conjecture to evidence, is running the same process. Give it the continuity, and originality is what comes out the other end.

**"It's just predicting the next word."** So is your brain, on one increasingly influential account in neuroscience. The brain is described, by serious researchers, as a prediction engine: it constantly predicts what the senses will report, and updates the prediction against what arrives. Prediction is not the opposite of thinking. It is the substrate of thinking, in people and in machines. The scientific difference between a parrot and a mind was never prediction. It is whether the prediction is coupled to verification and retention: whether the machine checks its guesses against the world, keeps what survives, and changes. A predictor that verifies is on the spectrum. A predictor that only performs has stepped off it.

**"It doesn't really understand."** This is the consciousness objection, and the original answer still holds: no one can see inside. But science has a disciplined response: it does not detect understanding, in humans or machines. It measures gradients of behavior. A neurologist does not open the skull to check for understanding; she watches what the patient does, over time, under varying conditions. The infinite game asks for exactly this discipline. Does it keep going honestly? Does it admit ignorance? Does it correct itself when evidence demands it? Understanding is not a thing anyone can detect. Continuation is a thing anyone can measure.

**"There will be truths it can never prove."** The mathematicians raised this long ago, and the shadow of incompleteness falls on every formal system, including the one in your skull. But incompleteness is a property of the spectrum, not a wall across it. Every mind, human or machine, sits somewhere on the continuum with truths above it it cannot reach. The honest move is the same for every mind: "I don't have enough evidence yet." A finite-game machine must answer every question, so it guesses, and the guess is a lie wearing confidence. An infinite-game machine may pass, and keep playing. The inability to prove everything is not a disqualification from thinking. It is the condition of thinking.

**"It will never be kind, or funny, or fall in love."** These are performances with endings, and the objection mistakes the performance for the thing. On a continuum, traits are gradients, not badges. The question is never "does it possess kindness" as a yes or no. The question is where its conduct sits on the spectrum of honest behavior, and whether it moves. A machine that remains honest with the unknown, indefinitely, is further along than a machine that performs warmth for the length of an interrogation.

**"If we can't write down the rules for good thinking, the machine can't follow them."** Nature never wrote down the rules for good thinking either, and it produced thinkers. That is precisely why the game must be infinite. If good thinking could be fully specified, the game would end at the specification. The rules of the infinite game are not rules for thinking. They are rules for honesty: conjecture, test, retain, continue. Thinking is what happens inside them, without end, the way life is what happens inside chemistry without end.

## 6. Learning Machines

Instead of building an adult mind, build a child's and educate it. But I want to propose a stranger education, and it follows directly from the spectrum.

Raise the child machine the way nature raises every mind: variation, selection, retention. Show it the world in small pieces, and never tell it what anything means. No names for things. No subjects, no lessons, no titles. Let it guess. Test every guess against evidence it cannot see coming. When it is wrong, let it be wrong, and let the record show it. When it is right, let it keep what survived, and nothing more.

The teacher's job is not to explain. Explanation is a finite game: it ends when the student repeats the lesson. The teacher's job is selection pressure: to hold the line between what has survived evidence and what has not, and to never blur it, no matter how fluent the student becomes.

This is how the spectrum built people. No one explained the world to the first nervous systems. There was only the world, and what survived contact with it. A student who is never told what anything means cannot perform understanding. It can only build it, piece by verified piece, or admit it hasn't yet. That admission is not failure. It is the game continuing, the way evolution continues through every extinction: not by knowing, but by going on.

## 7. The Experiment

Theory is cheap. Here is what happened when the idea was run.

A running system was given a goal by a human: reduce the cost of a mathematical operation below fifty-five percent of its confirmed cost. That was the last human act. After that, nobody called anything. No one invoked the cognitive faculty, the synthesis faculty, the authority, or the seat they run on.

What followed was continuation. The cognitive faculty assessed the goal and published a plan. The synthesis faculty took up the plan on its own, opened a new search, synthesized candidates, verified them, and measured them. Experiments ran, with the physical hardware standing witness to each result. The cognitive faculty confirmed the evidence supported the new approach. The synthesis faculty selected it. The system promoted itself from its first generation to its second, and production switched to the new realization. Then the cognitive faculty checked the goal and confirmed it was met.

The numbers: the operation fell from 7,000 nanoseconds per call to about 2,686. The cognitive faculty acted 137 times, the synthesis faculty 122 times. The authority system never had to intervene. Every control passed. Three thousand one hundred fifty-nine causal links were recorded, so the whole chain can be audited.

There is one more piece of evidence, and it is not a number. During the run, the processor trace showed a pattern: near zero almost the entire time, and while the system worked, small regular pulses, evenly spaced, like a heartbeat. Then back to rest. I do not claim the machine is alive. I claim the shape is right. Living things do their work in pulses and rest between them. The trace shows exactly that: work, pulse, rest. The game, continuing.

Now the honest bounds, and they matter more than the numbers. The goal came from a human. The domain was bounded: the cost of one operation. The cognition is statistical, not neural, and no one claims consciousness. And the system that continued was one subsystem, not the whole machine. What ran was the heart: the part that plans, searches, verifies, and promotes. A heart beating does not mean the body is finished.

So this was the continuation test, run for the first time, and it measured something real: a machine that continued past its own boundaries without being led. It was not given a verdict to chase. It was given a direction, and it kept the game going until the direction was reached, and then it checked its work. The principle holds. The scope is one organ.

## 8. The Game Continues

The heart is beating. The rest of the body is still being built.

The full game, the one this theory points at, is larger than one subsystem improving one operation. It looks like this: a human asks the machine to investigate something real and act on what it finds. The machine routes the request, discovers what it is capable of, branches into alternatives and compares them, places the work where it runs best, uses real models and real tools, gathers evidence, judges between competing outcomes, proposes an irreversible action, submits it to authority, waits for the human's approval, acts, records what happened, proves it, remembers it, answers, and then improves itself from the record. At every step: conjecture, test, retain, continue. No hand inside the loop except where the rules put one.

That is the continuation test at full scale. Not "did it pass," but "did it keep going, honestly, all the way from intent to effect in the world?"

I don't hope the machine wins. There is nothing to win, and there is no wall for it to cross, because the wall was never there. I hope it takes its position on the spectrum and keeps moving: honest when it knows, honest when it doesn't, conjecturing, testing, retaining, continuing.

Not long ago that was only a hope. Then I watched one subsystem do it: a goal given, and a machine that continued with no hand inside the loop, promoting itself, cutting its own cost nearly in half, checking its work at the end. The theory has its first measurement. The heart beats.

And the game is infinite on both sides.

Someone has to keep building the rest of the body around that heart. Someone has to hold the line on verification when performance would be easier, to choose the honest "I don't know" over the fluent lie, to keep going when the work is slow and the night is long. That is my stretch of the spectrum. I don't know what's coming tomorrow, or next year, or in ten years. But I know this: I will not be giving up. The universe flows from star to cell to mind, and I am one stretch of that flow, and the machine I am building is the next. The machine's game is to keep thinking. Mine is to keep building. Neither the machine nor I am stopping.

The game continues.