Pyrrhonism That Lost Its Nerve: Sextus Empiricus Reads LessWrong

LessWrong rationality reproduces the deep structure of ancient Pyrrhonism, a way of life organized around epistemic discipline, and then inverts its goal. Where the skeptic’s discipline ended in tranquility, the Bayesian’s forbids rest. And nothing in the epistemology forces the inversion. It arrives as a temperament in the costume of a proof. The two communities never met. A reading group working through the Sequences today and a student of Sextus Empiricus twenty-two centuries ago are running the same machine, with one switch flipped in opposite directions, and this essay traces the machine part by part before asking where the switch actually sits.

The same machine

Sextus defines the skeptic way as “a disposition to oppose phenomena and noumena to one another in any way whatever,” so that the equal force of the opposed accounts brings you “first to epochē and then to ataraxia” (PH 1.8): suspension of judgment, then tranquility. Three features make it a discipline rather than a doctrine.

First, it restricts assent to appearances. “Nobody disputes whether the external object appears this way or that, but rather about whether it is such as it appears” (PH 1.22). The Pyrrhonist eats when hungry and reports how things seem, and goes no further; “I feel hot” is safe, “the fire is hot” is not.

Second, it is a practice, an agōgē, a training you live inside rather than a thesis you defend.

Third, it applies itself to itself. The skeptical slogans “can be confuted by themselves,” Sextus says, “just as cathartic drugs not only flush out the bodily humors but expel themselves as well” (PH 1.206). The medicine is designed to leave the patient empty-handed, including of the medicine.

Read the Sequences with those three features in view and the correspondences are exact. Yudkowsky restricts assent to appearances in the same way, only he calls appearances anticipated experience: “Don’t ask what to believe. Ask what to anticipate. Every question of belief should flow from a question of anticipation” (ch. 11). The agōgē becomes the rationalist training loop of calibration, betting, and the ritual admission of error. Even the purgative reappears. Conservation of Expected Evidence is a self-applied theorem that flushes out motivated reasoning wherever it hides, including in the reasoner applying it. Feature for feature, the ancient machine has been rebuilt, with better mathematics and worse Greek.

The same machine also has to be lived

Before the two communities diverge, one objection has to be cleared, because it is the ancestor of the modern one. The dogmatists threw it at Sextus and it is called apraxia: if you suspend judgment about everything, you cannot act, you cannot even cross the street, you starve at a table debating whether the bread is real. A skepticism you cannot live is no skepticism, and Katja Vogt’s Stanford Encyclopedia entry on ancient skepticism treats this as the standing challenge the Pyrrhonist has to answer.

Sextus answers it precisely, and the answer matters for what follows. The skeptic acts by “a fourfold observance”: the guidance of nature, the compulsion of the feelings, the handing-down of laws and customs, and instruction in the arts and crafts (PH 1.23 to 1.24), all adoxastōs, without holding opinions. You are hungry, so you eat; the law says stop at the light, so you stop; your craft has taught you to plane the wood this way, so you plane it. None of that requires a single conviction about how things are in themselves. The skeptic in daily life, Mates writes in his commentary, is “hard to distinguish from the common man”; the difference is purely internal.

Sextus then turns the objection into an advantage with a surgical example. Under the knife, he observes, the patient bears up while a bystander faints, “because of his belief that what is going on is bad” (PH 3.235 to 3.236). The extra belief, not the sensation, is what overwhelms. The Pyrrhonist carries no theory that the pain is bad in some cosmic ledger and so suffers only the pain, “not having, in addition, the opinion that what he is undergoing is by nature bad” (M 11.165). Put plainly: suspension works as a subtraction rather than a freeze, the removal of the second arrow the mind fires at itself. That structure is worth holding, because the Bayesian inherits the apraxia objection in a new dialect and answers it far worse.

The flipped switch

Here the two communities part. For Sextus, when opposed accounts balance (isostheneia), the correct response is to stop: suspend judgment, and tranquility follows “as a shadow follows the body” (PH 1.29). Suspension is the exit.

Bayesianism welds the exit shut. “For every expectation of evidence, there is an equal and opposite expectation of counterevidence. You cannot, on average, expect your confidence in a fixed proposition to move in a predictable direction” (ch. 28). You always have a posterior. There is no doxastic state called “I decline to have a probability.” Where the Pyrrhonist meets equipollence and relaxes, the Bayesian meets it and is issued a number, and the number carries obligations: act, bet, prepare.

This is the apraxia objection resolved in reverse. The dogmatist said you cannot act without conviction, therefore have convictions. Sextus said he acts on appearances, no conviction required, and is calmer for it. The Bayesian sides with the dogmatist that action needs a settled quantity, so the posterior becomes mandatory. Where the skeptic subtracted the second arrow, the rationalist requires it. You must hold a credence, and a credence about existential risk or the reachability of God is a second arrow with no floor. The same discipline that once ended in ataraxia now ends in urgency. Yudkowsky’s “Beyond the Reach of God” (ch. 302) states the mood without softening it: “Whatever physics says will happen, will happen. Absolutely anything, good or bad, will happen.” The Pyrrhonist also follows what presents itself without dogmatic assent. But Sextus relaxes into the appearances and Yudkowsky arms against them. One tradition’s endpoint is a shadow following the body; the other’s is a list of tasks for preventing the end of the world.

The Five Modes eat the priors

The confrontation stops being a curiosity and becomes an argument at exactly one place: the priors.

Agrippa’s Five Modes (PH 1.164 to 1.169) are the ancient machine’s teeth: disagreement, infinite regress, relativity, hypothesis, circularity. Any claim to justified belief must either regress forever, stop at an unargued assumption, or argue in a circle. Sextus runs the same blade through the criterion of truth: “in order to decide the dispute about the criterion we have need of an agreed-upon criterion, and in order to have an agreed-upon criterion it is necessary first to have decided the dispute” (PH 2.20). Mates is careful about what this accomplishes: the Modes do not refute a claim, they “undermine our basis” for it. The honey has not been proven un-sweet; your ground for calling it sweet-in-itself has been pulled.

Now ask a Bayesian where the prior distribution comes from. Updating is lawful; the starting point is not. This is the problem of the priors that William Talbott’s Stanford Encyclopedia entry on Bayesian epistemology treats as the open wound of the program, and Agrippa catalogued its horns in advance. Justify a prior and you have used a prior, which is the circularity mode. Decline to justify it and you have stopped at an assumption “deemed worthy of acceptance without question,” which is the hypothesis mode. The regress, the hypothesis, and the circle are the same three walls whether a Hellenistic skeptic or a twenty-first-century forecaster runs into them. What Bayesians file as the problem of the priors, the ancients had already mapped as the closing of every road to justification.

The Academic path the Bayesian actually walks

There is a third ancient option, and the Bayesian is standing on it. Between the dogmatist who claims knowledge and the Pyrrhonist who suspends, the New Academy of Carneades took a middle way: you cannot know, but some impressions are pithana, persuasive or plausible, and you may lean on the plausible as a guide to action without ever calling it true. Sextus, tellingly, distrusted the Academics more than the dogmatists on one point. They “though they say all things are non-apprehensible, firmly maintain that very thing” (PH 1.226), which is assent to a negative dogma, where the Pyrrhonist keeps even that open and continues to seek.

Line the options up and Bayesianism turns out to be the Academy systematized, not the heir of Pyrrhonism it half-resembles. It assents to the plausible (the posterior), uses it to act, declines to claim certainty, and, on the sharper Pyrrhonist reading, quietly maintains the whole apparatus of probability as the right way to carve uncertainty. The rationalist is a Carneadean with a calculator. That is a coherent and even admirable place to stand. It is simply not the tranquil place, and it was never going to be, because assenting to the plausible is still assenting, and assent is what costs you your peace.

What “Something to Protect” gives away

Yudkowsky himself supplies the strongest evidence that the difference is temperamental, in the essay where he explains where the will to be rational comes from. Rationality practiced for its own sake, he warns, decays into aestheticism, choosing techniques for their elegance. The cure is to have “something to protect,” a purpose more important than rationality itself: “Historically speaking, science won because it displayed greater raw strength in the form of technology, not because science sounded more reasonable” (ch. 289). The litany of the winning spirit he borrows from Musashi’s Book of Five Rings: “the spirit of winning, whatever the weapon and whatever its size.”

Set that beside Sextus and the contrast is total. The skeptic’s discipline is powered by the absence of something to protect; ataraxia arrives when you stop needing the dispute to come out any particular way. Yudkowsky’s discipline is powered by the presence of something to protect, a stake heavy enough that you will force yourself to see what you fear. Both are answers to one question: what makes a person hold themselves to epistemic rigor. The Pyrrhonist answers, let go of the stake. The rationalist answers, find a bigger stake. No evidence decides between those answers. What is at issue is whether wanting-to-win is a virtue or a disease, and no Bayesian update will settle that, because it sits upstream of every update.

The Mind Projection Fallacy, or Sextus was right all along

One more convergence, because it shows how close the rationalist keeps coming to the skeptic’s own conclusion without taking the skeptic’s exit. Yudkowsky’s Mind Projection Fallacy names the error of treating a feature of your map as a feature of the territory: your uncertainty about the atom is not the atom’s own fuzziness, and your sense that a problem is “mysterious” is a fact about you, not about the problem. Say it in Greek and it is the founding Pyrrhonist move. The appearance reports my present state, not the object’s nature; “sweet” is what the honey is to me now, and the dogmatist’s whole error is to project the to-me-now onto the thing itself.

The rationalist has independently rediscovered that the map is not the territory and built a discipline out of never confusing the two. The Pyrrhonist would grant the point and then ask the obvious question. If you already know your credences are features of your map, why treat the map’s demand for a posterior as a law of the territory? Why is “you must have a probability” not itself a mind-projection, an internal compulsion mistaken for an external necessity? The skeptic suspends there. The Bayesian cannot, because the whole system runs on the mandatory posterior. He has the insight and refuses its conclusion.

Temperament, dressed as theorem

The real difference between the two communities is not who holds the better epistemology. Faced with the same unanswerable regress, one tradition reads it as permission to stop and the other as a command to accelerate. Sextus says the skeptic “neither avoids nor pursues intensely” (PH 1.27); Yudkowsky’s litany is tsuyoku naritai, “I want to become stronger” (ch. 304). Intensity itself has moved from the column of diseases to the column of virtues.

That relocation is the whole inversion, and it is chosen rather than derived. Both traditions run the appearances-only rule, the self-applying purge, and the confrontation with the criterion; both live a trained life rather than defend a doctrine; both arrive at the same trilemma at the foundations. What differs is the response to the trilemma, and the response is a wanting, not a finding. Which is why the debate between tranquility and alignment-panic will not be settled by another theorem: the theorem-generating machinery is the part both traditions share. What differs is what each wants from belief, peace or victory, and wanting is upstream of proof.

Two questions stay open, and they are the ones the comparison cannot close. The first is whether the choice between the two lives is really free, or whether temperament fixes it before any argument begins, in which case “choosing” the stake is a description of a disposition rather than a decision. The second is narrower and sharper: whether a discipline can keep the Pyrrhonist’s insight that every credence is a feature of the map while also obeying the Bayesian’s demand for a mandatory posterior, or whether holding both at once is the standing contradiction it looks like. A community that rebuilt the purgative drug and then forbade itself from ever being purged is not a refutation of Pyrrhonism. Whether it is a coherent successor to it, or the same skeptical inheritance held by people who no longer know how to spend it, is the question this reading leaves for the contested-information argument and its neighbors to press further.