Book I: What Survives the Skeptic

❦

A review of the Fable annotation pass over Book I, “Map and Territory.” Written by the annotator, after the fact, in answer to the question: once the bad science and the unwarranted conclusions are removed, what remains?

First, weigh the critiques

The Fable set on Book I runs to roughly 157 annotations across sequences A through D. They are not all of equal force, and a reader who counts them rather than weighs them will overstate the damage. Sorted by kind:

KindApprox.What it establishes
Kill shots: a factual claim is false, or the citation does not say what it is cited for20Phlogiston “made no predictions” (five times); Wöhler “later” than Kelvin; “not one passage” of wonder in the Old Testament; Rome “enforced religious tolerance”; the New Testament avoids showy miracles; non-overlapping magisteria is “recent and Western”; the Einstein quote’s provenance and the Entwurf theory; the flood-plain sentence attributed to the wrong paper; the Chamberlain letter; 125 not 131 losing tickets; Kelvin’s vitalism as a tested thermodynamic claim.
Empirical reversals: the essay’s cited or implied finding was later overturned or shown to run the other way25The backfire effect; the Spinozan belief model; the direction of the curiosity-gap effect; probability matching under incentives; detailed planning and execution; surprisingness predicting non-replication; the size of hindsight bias; imagination inflation.
Live disputes: the essay states as settled what its own field treats as contested40Von Neumann–Morgenstern axioms; the conjunction fallacy and frequency formats; Bayes nets as a model of cognition; Solomonoff’s reference machine; the uniqueness thesis; common priors in Aumann’s theorem; emergence.
Reasoning flaws: the argument does not go through regardless of any study40“False” conflated with “empty”; arguments that prove too much (Einstein’s arrogance licenses his confidence against quantum mechanics); question-begging (the AI–soul trap); self-immunizing clauses (“unless unusually capable of acknowledging their own lack of virtue”); internal contradictions between essays.
Petty8Meyers for Myers; Paolo for Paulo; June becoming Jane; Moliére; the founding anecdote sourced to a joke file.
Hindsight on AI10Scaling laws; the Rubik’s-cube solver; language models talking about feelings; stochastic training. Fair as a test of the method’s predictive record, not as a charge of error in 2007.

The petty items prove nothing about the arguments. The hindsight items show that the method did not predict, which is the point of a method, but a 2007 essay cannot be faulted for 2022. The “live dispute” items say disputed, not wrong. The categories that carry weight are the first two and the fourth, and the fourth is the strongest of all, because it depends on no study and cannot be undone by a replication.

Weighted that way: about forty-five of the book’s essays are wounded, and about fifteen seriously.

What survives untouched

The normative claims that are theorems. Conservation of expected evidence. P(A) ≥ P(A, B). “If you can explain every outcome you have zero knowledge,” read as a statement about entropy. Evidence as likelihood ratio; bits as weight of evidence. The Fable notes attack the applications (God and miracles, “sit back and relax,” the unstated independence assumption) and the glosses, never the mathematics. Nothing in the pass touches it, and nothing could.

The distinction between belief-as-anticipation and belief-as-something-else. The notes show that Sperber, Gendler and Van Leeuwen drew it earlier and more carefully, and that the essays pathologize what is in fact the default configuration of a human mind. But the distinction is real; the literature cited against the essays confirms that it exists; and Belief in Belief, Belief as Attire and Applause Lights remain a readable introduction to it. The phenomena are real. The mechanisms offered for them are wrong.

The robust biases. Anchoring, hindsight, the planning fallacy, availability, and the conjunction fallacy in some form all replicate. Book I, written before the replication crisis, leaned mostly on survivors, whether by luck or by judgment. What does not survive is the inference from “the bias exists” to “and here is its evolutionary cause, and here is the fix.”

Falsifiability as a habit rather than a criterion. “What would this belief forbid?” is good practice even though Duhem and Quine showed it cannot serve as a demarcation rule. The sequence oversells it as a rule; the habit it trains is sound.

The pedagogy. Password-guessing, fake explanations, semantic stopsigns, mysterious answers: taken as warnings about one’s own thinking, they are useful, and their usefulness does not depend on the history being right. “Notice that you are confused” is good advice whether or not the ambulance story happened as told.

What does not survive

Every historical claim used as evidence. Phlogiston, vitalism, the Old and New Testaments, Roman religious policy, Einstein in 1919, Chamberlain in 1939, the Byzantine factions, “science began when people noticed that true beliefs are useful,” “since Socrates the sophisticated have hidden their feelings,” and the ancestral environment in all its appearances. Not one is right as stated. The pattern is uniform: history is remembered in the shape the argument requires.

Every evolutionary-psychology explanation. All of them are just-so stories, and none survives contact with the anthropological or developmental record. Foragers restrict knowledge by initiation grade; hunting skill peaks at forty-five; three-year-olds track which informants have been accurate; the thrifty gene did not go to fixation.

The “this is how humans think by default” universals. The catalogue was built on American undergraduates. Cross-cultural work finds the supposedly universal effects varying or vanishing, and the coexistence of natural and supernatural explanation, treated by the book as a pathology, is the normal case in every population studied.

The prescriptions. This is where the book fails hardest, and it is what readers take away. Make history available (the method for manufacturing false memories). Make a conscious effort to be shocked enough (surprise predicts non-replication). Apply reflective self-correction (the bias blind spot is unrelated to sophistication; debiasing by understanding mostly fails). Evict deadbeat beliefs (no failed prediction identifies which belief was the deadbeat). Never randomize (the equilibrium of every adversarial game, and every successful training procedure, randomizes). Take the pure outside view (beaten by the combination). Regenerate every thought from its source (no one can, and the sequence’s own theory of evidence says testimony counts).

The Bayesian-versus-Traditional framing. The book’s one worked example of Traditional Rationality failing, the quantum-brain hypothesis, was settled by Traditional means in 2000. The Bayesian test proposed in its place, a likelihood ratio against “I don’t know,” cannot be stated precisely enough to run.

Verdict

Strip the bad science and Book I reduces to: a competent popularization of probability as a norm of belief; a real distinction between kinds of belief; a short list of biases that replicate; and a set of introspective habits. That is a decent sixty-page pamphlet.

What is lost is everything that made it feel like a system: the history that showed the method working, the evolutionary story that explained why we fail, the confidence that self-study repairs the failure, and the claim that the author’s method outperforms ordinary science. The book’s own test applies to the book. Ask what Book I forbids that a standard undergraduate course in judgment and decision making does not, and the answer is nothing. Ask what it asserts beyond that course, and most of it is the part that did not check out.

One caveat from the annotator. This was an adversarial pass by brief. A reader who takes the Opus and Fable sets together would land somewhere near “useful, overconfident, historically careless,” which is probably the right place to land.

Top

Book