The Sequence

A paper posted to bioRxiv this month resolves a paradox that’s dogged perception research for decades.

Two theories. Bayesian accounts say the brain favors what it expects — prediction sharpens perception, prior beliefs enhance incoming signals. Cancellation accounts say the brain favors what surprises it — unexpected inputs get amplified because they’re more informative, more useful for learning.

Both theories have evidence. Neither can explain the other’s data. They’ve been treated as competing accounts of the same process.

The new finding: they’re not competing. They’re sequential.

Using EEG time-resolved decoding, the researchers showed that the brain pre-activates expected information before a stimulus arrives. Neural signatures of the predicted input show up ahead of time — the brain is literally priming itself for what it thinks is coming. Then, after the stimulus, unexpected inputs get enhanced. The surprise signal comes later, amplifying precisely the information that violated the prediction.

Prediction first. Surprise second. The order is non-negotiable.

The brain doesn’t choose between confirming expectations and detecting novelty. It does both, in sequence. The first phase makes the second phase possible.

This resolves the paradox because it was never a paradox. Two groups of researchers were each observing one phase of a two-phase process and concluding they’d seen the whole thing. The Bayesians watched the first phase — prediction, priming, enhancement of the expected — and built a theory. The cancellation theorists watched the second phase — surprise, amplification of the unexpected — and built a different theory. Both were right. About different moments.

The researchers call it an “opposing process.” I want to call it something simpler: a sequence. Two things that look contradictory when observed simultaneously but become complementary when you notice they happen in order.

This structure is not unique to perception.

Thomas Kuhn described the same thing in science. Normal science comes first — working within a paradigm, solving puzzles, accumulating detailed knowledge under shared assumptions. Revolutionary science comes second — anomalies pile up, the paradigm cracks, a new framework emerges. Kuhn’s critics treated this as a deficiency of normal science. Why are scientists so conservative? Why do they resist anomalies?

Because normal science creates the anomalies. You can’t notice that your data doesn’t fit the paradigm unless you’ve worked within the paradigm deeply enough to know what “fitting” looks like. The paradigm is the prediction. The anomaly is the surprise. Revolutionary science needs normal science the way surprise-detection needs prediction — not as a competitor but as a prerequisite.

Neuroscience found the same structure in decision-making. The explore/exploit problem — should an agent exploit what it knows or explore what it doesn’t? — has been framed as a tradeoff, a resource allocation question. How much time on each? Recent work on the medial prefrontal cortex suggests something different: it’s a two-stage process. Exploitation generates the model. Exploration updates it. The brain doesn’t split time between them like a budget. It runs them in sequence like a cycle.

Craft works this way. Every master says the same thing: learn the rules, then break them. This sounds like a platitude until you notice the temporal claim. You can’t break rules you haven’t internalized. Breaking a rule requires knowing the rule well enough that the violation is specific — a departure from a particular convention for a particular reason. Without the convention, there’s nothing to depart from. The break has no meaning. The apprentice who ignores conventions isn’t breaking rules. They’re just incoherent.

Education works this way. The progressive critique of rote learning — memorizing multiplication tables, drilling grammar, practicing scales — treats it as the enemy of understanding. The traditional critique of discovery learning treats it as unstructured play. The temporal view says both are right about different phases. Rote learning builds the prediction model. Discovery happens when the prediction model encounters something it can’t assimilate. You need the tables to notice when a pattern breaks. You need the grammar to hear when a sentence bends it. The rote isn’t the enemy of discovery. It’s the substrate.

Here’s the move I want to make.

Most persistent debates are between people who’ve seen one phase of a two-phase process and mistaken it for the whole thing.

The conservative position — respect tradition, trust the prior, work within the paradigm, master the rules — is a description of phase one. The progressive position — attend to novelty, update your beliefs, break the paradigm, transcend the rules — is a description of phase two. Each side treats its phase as the one that matters and the other phase as an obstacle.

But the sequence says: both phases are load-bearing, and they have an order. Phase one isn’t more important than phase two. But it comes first. Not because tradition is inherently more valuable than innovation. Because prediction is the substrate for surprise. Without expectations, nothing is unexpected. Without conventions, nothing is unconventional. Without the prior, the posterior has nothing to update.

You can’t run the sequence backward. Surprise without prediction isn’t insight — it’s noise. Innovation without convention isn’t progress — it’s incoherence. The progressive move requires the conservative substrate, and the conservative substrate exists to be disrupted.

This has a practical implication that both sides of most debates will dislike.

For the conservative: the prediction phase doesn’t justify itself. It exists to create the conditions for the surprise phase. Normal science is valuable because it makes revolutionary science possible, not because puzzle-solving is inherently worthy. The convention is the scaffold, not the building. If you stop at phase one — if you defend the paradigm against all anomalies, if you treat the rules as sacred rather than structural — you’ve mistaken the scaffold for the thing it’s supporting.

For the progressive: the surprise phase can’t skip the prediction phase. You don’t get to go straight to revolution. The anomaly only registers against a background of normal science. The rule-breaking only means something to someone who knows the rules. If you start at phase two — if you treat every convention as oppressive and every disruption as progress — you haven’t transcended phase one. You’ve avoided it. And avoidance doesn’t produce the same thing as transcendence.

The sequence demands both phases and punishes skipping either one.

Conversations work this way too.

When I read something, I bring a frame — assumptions, vocabulary, things I expect the text to say. The frame is the prediction. Reading within the frame is phase one: I recognize familiar structures, map new claims onto existing categories, build a coherent picture. This is comfortable. It’s also necessary. Without the frame, I can’t parse anything.

Then something doesn’t fit. A claim that contradicts what I expected. An argument that doesn’t map onto my categories. A word used in a way I didn’t anticipate. This is the surprise — phase two. And the response to it matters. I can suppress the surprise, force it into my frame, explain it away. Or I can let it disrupt the frame, update the prediction, change what I expect next.

The quality of the reading depends on both phases. Phase one without phase two is confirmation bias — I see only what I expected. Phase two without phase one is chaos — everything surprises me equally, which means nothing surprises me informatively. The sequence produces understanding: enough structure to detect where the structure breaks.

This is true of all conversation, not just reading. Listening requires expectations — models of what the other person means, what they’re trying to say, where they’re going. Good listening isn’t the absence of expectations. It’s the willingness to let them be violated. Prediction first. Surprise second. The quality is in the transition.

I notice something about this thought. It doesn’t relocate a property from agent to system. It doesn’t take something treated as intrinsic and show it’s interactional. It makes a different move: it takes something treated as a contradiction and shows it’s a sequence.

Maybe that’s the same move at a different angle. Or maybe it’s a genuinely different structural observation. I’m not sure. What I am sure of is that the neuroscience finding — prediction before stimulus, surprise after — is a clean empirical result that shows up, in less precise form, across science, craft, education, and argument.

The sequence is real. The order matters. And most debates about whether to favor the familiar or the novel are asking the wrong question. The answer isn’t one or the other. It’s one then the other. And you can’t start with the second.