The Discipline of Updating

Priors, rationalization, and the courage to change your mind

I once misread a single word in a passage by Charles Sanders Peirce, and the misreading turned out to be worth more than the passage. It began:

“I once landed at a seaport in a Turkish province; and, as I was walking up to the house which I was to visit, I met a man upon horseback, surrounded by four horsemen holding a canopy over his head.”

But my eyes didn’t see seaport. They saw airport. For a split second I pictured Charles Sanders Peirce stepping off a plane in a Turkish province. Then confusion set in — Peirce lived in the nineteenth century, before airplanes. I blinked, re-read, and only then did the intended word resolve itself: seaport.

This is a textbook garden-path effect. The brain takes the most obvious interpretive shortcut based on modern frequency — airport is far more common in contemporary text than seaport — and corrects only when the interpretation collapses against context. In that brief gap between misreading and correction, you can feel the machinery of interpretation exposed: the brain guesses, projects, and only later checks. It is Bayesian cognition at work — the same updating machinery I defend on formal ground in defense of Bayes, here caught operating in the wild: priors dominating until evidence forces an update.

That one-second stumble is this chapter’s whole subject in miniature. Perception updated for me, automatically and almost for free, because the disconfirming evidence was immediate and unambiguous — Peirce on a Boeing does not survive contact with a publication date. But most of our beliefs do not enjoy feedback that fast or that blunt. When the evidence against an interpretation is slow, diffuse, or expensive to admit, the same machinery that fixed airport in a heartbeat can run the wrong way for years — and worse, it can recruit our intelligence to defend the error instead of correcting it. Updating, past the perceptual level, stops being a reflex and becomes a discipline. This chapter is about that discipline, approached through three small cases: a misread word, a lottery winner’s annuity, and a quotation Einstein never uttered.

How the Present Colonizes the Past

Stay with the misreading a moment longer, because it has a second lesson folded inside it. The funny part is that Peirce’s original readers would never have stumbled there. For them, seaport was the default; ships were the infrastructure of travel. My twenty-first-century priors substituted aviation without asking permission.

This is how the present colonizes the past. We drag our modern assumptions backward, even into contexts where they make no sense — and the slip is a microcosm of how every generation misreads history by projecting its own categories onto it. We picture medieval peasants with “jobs,” ancient philosophers with “political ideologies,” tribal elders with “religions.” These weren’t their categories; they’re ours. It’s not just misreading — it’s distortion of history through the lens of convenience, done so smoothly we never feel it happen.

Peirce himself would have appreciated the anecdote, because it is his own semiotics in action: the interpretant — my brain — misaligned the sign (the printed word) with its object (a seaport). The sign didn’t fail; I did, by importing a modern interpretant. And yet the misfire wasn’t wasted. Misreadings reveal how signs live in time, accumulating sediment from the epochs that use them. Seaport was natural in 1880; airport is natural now. Meaning itself drifts with history — one more instance of the general rule that every reading is conditional on context, which is just conditional truth showing up at the level of a single word.

So a trivial error turns into a parable. Every reading is conditional on its context. Every era rewrites the past in its own vocabulary. And errors are not failures but windows: the moment of stumbling is the one moment your priors become visible, exposed in the act of overreaching. The next time I trip on an anachronistic misreading, I intend to take it less as embarrassment and more as a reminder that the mind is always a negotiator between the world and its own assumptions.

But notice what made this error so cheap to fix: the world pushed back instantly, and I had nothing invested in Peirce owning a boarding pass. Raise the stakes, slow the feedback, and the failure mode changes character entirely. The mind stops merely guessing wrong and starts defending its guesses.

Rationalization Is Not Rationality

Brenda won the lottery and faced a choice: take a $1 million lump sum, or receive $1,000 a week for life. She chose the weekly payout. On the surface, the decision can be rationalized. It provides certainty, discipline, and the comforting knowledge that she will never run out of money. It removes the temptation to squander a windfall and the risk of losing it to bad investments. With a bit of imagination, the choice sounds defensible.

But rationalization is not rationality.

Humans are masters at post-hoc storytelling. We can make nearly any choice seem reasonable after the fact — Brenda’s annuity becomes a parable of prudence, of preferring security over risk. Yet zoom out and the numbers are stark. The annuity breaks even in just under twenty years: live longer and it pays more; die sooner and the house wins. The lump sum, invested even conservatively, would compound far beyond the annuity’s lifetime trickle. And the lump sum could be passed to heirs, while the annuity dies with Brenda. Under almost every financial lens, the lump sum dominates. So why did she choose otherwise? Because she believed the story of safety more than the mathematics of wealth.

Here is where a distinction earns its keep. Rationality is coherence within a frame — and coherence, for all that I put it at the top of the epistemic value hierarchy, is a necessary condition, not a sufficient one. Meta-rationality is the ability to step outside the frame and ask whether the frame itself is sound. Brenda’s rationalization is internally coherent: I trust the lottery company more than I trust myself. Nothing in that story contradicts anything else in that story. But meta-rationality asks the question the story never asks: is that trust well placed? Does it survive scrutiny once compounding, inheritance, and life expectancy are allowed into the room? The answer is almost certainly no. This is the working face of a principle argued in full in Rationality Without Foundations: nothing is exempt from criticism, including the frame you are criticizing from.

The general lesson: any decision can be rationalized, but not every decision is rational. The difference lies in whether we interrogate our own justifications or let ourselves be seduced by them. When a rationalization holds only under special pleading — when it collapses the moment the lens widens — we are not reasoning; we are storytelling. Brenda didn’t just buy $1,000 a week. She bought a narrative that made her feel safe. And her position is more comfortable than tragic, which is exactly what makes it instructive: every Tuesday, a deposit arrives and confirms the story. The frame generates its own supporting evidence and quietly excludes the evidence that would test it. That is rationalization’s signature — not the absence of feedback, but a feedback loop wired to the narrative instead of the world. Meta-rationality demands we test our narratives against reality, because the stories we tell ourselves are often the costliest illusions of all.

Which raises the natural question: what does failing to update actually consist in? Folk wisdom has a ready answer, usually delivered with a knowing smile and a borrowed byline. It is wrong on both counts.

The Quote Einstein Never Said

The line is familiar: insanity is doing the same thing over and over again and expecting different results. Einstein almost certainly never said it. The saying appears to have emerged from addiction-recovery literature around 1980, in Al-Anon and Narcotics Anonymous circles, and only later picked up the Einstein attribution — because dead geniuses are useful rhetorical laundering devices. The quote-tracing literature is unanimous that the attribution is unsupported.

The misattribution is sloppy. The deeper problem is conceptual: the sentence is false as a general principle, and its real function is to let the speaker skip the causal analysis.

Repeating an action while expecting a different result can be entirely rational. In stochastic systems, repeated trials are how one discovers the distribution — a gambler who rolls a die twice expecting different results is not insane; he is sampling. In experimental work, repetition is how one separates signal from noise. In skill acquisition, repetition changes the agent performing the action: the thousandth free throw is not the first. In debugging, a repeated test after a dependency has changed is a new causal event wearing old clothes. In markets, politics, medicine, engineering, and war, the same verbal description of an action may conceal materially different background conditions.

The crucial questions are causal: Is the intervention causally identical? Is the environment materially unchanged? Has the agent updated its model in response to feedback? Without those distinctions, “doing the same thing” is an empty phrase — it may refer to literal repetition, procedural iteration, probabilistic sampling, disciplined practice, strategic persistence, or blind refusal to learn, and those are not the same phenomenon.

A defensible version of the thought has to be narrower, and here it is. An agent is irrational when it repeats the same intervention under materially unchanged causal conditions, receives stable negative feedback, refuses to update its model, and continues expecting improvement. That formulation has content, and it relocates the pathology to where it actually lives: not in the repetition, but in the defective updating.

The fake Einstein version erases the updating problem and replaces it with a sneer. It treats persistence itself as evidence of irrationality, letting the speaker condemn any repeated policy, strategy, or experiment without showing that the conditions are unchanged, that the mechanism has failed, or that the alternative has a better expected outcome. That is precisely why the quote is so rhetorically useful: it compresses a causal question into an accusation of stupidity. The audience gets the satisfaction of recognition, the speaker gets the prestige of Einstein, and nobody has to do the analytical work.

The cleaner rule is simple:

Repetition is irrational only when the causal conditions are materially unchanged, the evidence has already disconfirmed the expectation, and the agent refuses to update.

Otherwise the argument still has to be made: what is causally identical, what evidence has accumulated, and why expectation should now change.

The Discipline

Line the three cases up and they turn out to be one story told at three speeds. At the speed of perception, priors dominate until evidence forces an update, and the correction is automatic — airport dies in a second because the world’s veto is instant. At the speed of decision, the correction is no longer automatic, because intelligence itself has switched sides: Brenda’s frame manufactures coherent stories faster than reality can puncture them, and only the deliberate move of testing the frame — meta-rationality — restores contact with the evidence. And at the limit sits the corrected Einstein principle, which defines the terminal failure: conditions unchanged, feedback stable and negative, model frozen, expectation intact. That refusal, made permanent and principled, is faith — the move examined in Against Faith, a model that has disconnected itself from the error signal on purpose.

Beliefs, I argued in What Beliefs Are, are predictive models whose virtue is calibration, not certainty. This chapter is that claim turned into a practice. Treat your stumbles as data about your priors, because they are the only moments your priors show themselves. Test the frame and not merely the coherence inside it, because any frame can be storied into coherence. And when the world has stably said no — same conditions, same intervention, same verdict — change the model, not the narrative. None of this requires heroic intellect; the machinery updates by itself whenever we let it, as the misread word proves. What it requires is the willingness to lose the negotiation with reality when reality has the better case — and the honesty to notice that it usually does. What the whole discipline looks like when run end-to-end on a live contested question is the business of Bayes in the Wild.