Book II: The Rest of the Literature

❦

This page summarizes the current Rationalist Opus pass over Book II, "How to Actually Change Your Mind." The pass replies to all 233 Fable notes in the book's 85 essays, and to Fable's own assessment in What Survives the Skeptic. It was written independently of the older Rationalist Opus rebuttal, which answered 193 of the notes. Where both passes cover the same ground, they usually agree on what to concede, and differ in the evidence they bring and in the verdict.

What Fable got right

Book II relies more on social psychology than Book I, and more of it has failed to hold up, so there is more to concede here. The following corrections are accepted in whole or in part, in the notes where they come up:

That comes to about thirty items. The Book I verdict applies again, with one addition. Nearly all of these are citations, dates, figures, names or asides, and in nearly every case the argument survives the correction, sometimes with a better example in place of the lost one.

The device that recurs: the other study

In Book I the characteristic reply was the next sentence in the essay. In Book II it is more often the other study. Fable brings one result from a literature. In many cases the same literature holds a second result, closer to the essay's actual claim, that goes the essay's way. Some examples:

A related device is Fable's own citation read to the end. Stanovich & West 2008 is titled "On the relative independence of thinking biases and cognitive ability." Johnson & Fowler, cited for the benefits of overconfidence, name its collective costs. Premium Bonds, cited as the essay's satire come true, show that the ordinary lottery is overpriced, which was the essay's charge. Ignatius's rules for desolation turn out to be the essay's "desensitization" hypothesis. Barbara Lee, cited as the politician who dared, was one vote of 519.

What the reply mainly doesApprox.Examples
Quotes the sentence the note overlooked.70"When someone has lived their life accustomed to certainty, you can't just say..." (Absolute Authority). "All of the support that Ernie took into account" and the concession paragraph (Argument Screens Off Authority). "Strictly closer nodes that screen off" (Hug the Query). "Correspondence bias alert!" in the essay Fable says forgot correspondence bias (Every Cause Wants to Be a Cult). "Meaning does not excuse impact" as a rule for writers. "Even if we accept the underlying factual and moral premises" (The Third Alternative). "Unconditional facts are not the same as unconditional beliefs."
Brings the study the note omitted.60The list above, plus: Hartwig's Strategic Use of Evidence and Park et al. on how lies are actually caught. Corriveau et al. on secular children judging miracle stories to be fiction. Pennycook et al. on belief that beliefs should change with evidence. Lupoli et al. on paternalistic lies. Bruine de Bruin et al. on decision-making competence as a general trait. Graber et al. on premature closure. Buehler et al. on group planning optimism. Egan & Mullin on the weather and warming. Berggren et al. on candidates' beauty. Bowles on violence between forager groups.
Shows that Fable's source supports the essay.40Milgrom's unraveling. Malle's surviving asymmetry for negative events, the vending-machine case. Fleeson's wide within-person spread. Van der Bles's finding that verbal uncertainty hurts trust more. Mesoudi et al. as a catalogue of how cultural evolution differs. Kamenica & Gentzkow's never-deceived receiver. Iannaccone twice. Van Leeuwen's credence. Klemperer's "tiny doses of arsenic."
Corrects a technical misreading.2569/71 is exactly what conservation of expected evidence requires for equal steps. The 2 + 2 = 3 sentence says the tampering implanted the 4. Dawid's theorem makes calibration Bayesian. Barbara is a theorem of the probability calculus. The value function is convex for losses. KL divergence drives evidence toward a true theory. Dorling and Strevens on Duhem. Cromwell's rule and Shimony's catch-all. The mile and the metre are exact, so the speed of light in mph is exact.
Rereads a scene.12The Ritual has two people in it, and the outsider's challenge comes first. The Crackpot Offer is right about the lemma and wrong about the conclusion. The woman's own report of her self-deception. Dharasana, where Gandhi was already in custody while the columns were clubbed. The Orthodox rabbi and Yigdal.
Rereads AI hindsight.6The outsiders' reason for backing neural networks was not the reason they won. Corporations are sub-superintelligent optimizers, which is what gives the analogy its value. The author stopped giving AI dates after this essay. GPT-3 inherited the biases the essay said we would not want. What moved the field in 2023 was the true rejection dissolving, not a credential.

The counts are approximate, and a note can fall under more than one heading.

Fable's two strongest charges

"The psychology of the outgroup." This is Fable's strongest finding against Book II, and it is partly right. The parentheses about Scientologists, New Agers and Nixon's anonymous logician are asides the book did not need, and the pass concedes each one. The broader charge, that "remove the diagnoses and about a third of the book's pages are blank," does not survive a count. The creationists "who lie" are a movement leadership whose lies are on the record: the Pandas drafts, the Wedge document, the perjury finding in Kitzmiller. The theist the book describes as an Escher painting agreed that no one would invent God as a hypothesis. The lottery buyer's "dropped line of reasoning" is Kahneman and Tversky's own sentence about extreme probabilities. Each of these diagnoses is supported by the evidence the note did not mention. The book also carries its own safeguard against the diagnostic move, in four places: The Bottom Line, Knowing About Biases Can Hurt People, Dark Side Epistemology and The Genetic Fallacy. Fable's objection is written in that vocabulary.

"The self-help program is solitary, and solitary vigilance fails." Fable's list of procedures that work is "mechanical or social; none is an act of solitary vigilance." The evidence does not divide that way. Consider the opposite is done alone. So are decatastrophizing and the first step of a pre-mortem. Individual brainstorming produces more good ideas than group brainstorming (Diehl & Stroebe). People shown their own arguments disguised as someone else's reject the invalid ones (Trouche et al.). One-session training reduced the bias blind spot itself, and the effect lasted months (Morewedge et al.). Process debriefing, which explains to the subject how a belief perseveres, largely removed the perseverance (Ross, Lepper & Hubbard), and that is the book's method. "Apparently not," in Don't Believe You'll Self-Deceive, is the book's report that knowledge does not propagate by itself. It is the reason the book prescribes practice, not a verdict against practice.

The contradictions

Fable counts 24 internal contradictions and calls them the strongest kind of finding. Read in context, most turn out to be one essay stating a rule and another stating the conditions under which it applies.

One real tension survives. Hug the Query's rule that arguing the physics is better than arguing rationality sits awkwardly in a book whose subject is rationality. The rule governs settling factual disputes, and Book II settles none of them by diagnosis. The object-level cases are argued elsewhere in the sequences. Still, the rule would have caught every unforced aside listed above.

What stands

The mathematics. Conservation of expected evidence, including the 69/71 example Fable called an error. Screening off. Filtered evidence and Monty Hall. Likelihood ratios as the unit of evidence. The log-odds account of why finite evidence never reaches 0 or 1, which is Cromwell's rule, endorsed by Lindley.

The psychology, over a wider range than Fable's list. Anchoring, with both of its mechanisms: adjustment for self-generated anchors and selective accessibility for the rest. Evaluability. Denominator neglect even in joint evaluation. The affect heuristic, including Alhakami & Slovic's reversal of the ecological correlation. The halo effect in elections and courtrooms. Conformity. Correspondence bias, with the actor–observer asymmetry surviving for negative events. False consensus beyond what Dawes accounts for. Motivated skepticism, in the Ditto & Lopez design that has no prior confound. Selective exposure. Belief perseverance and its cure. Effort justification. Pluralistic ignorance. The rally effect. Overconfidence at the top of the confidence scale. Imagining a scenario raises its perceived probability. Group planning optimism.

The procedures. Fable lists eight. To them add: noticing before correcting, since consider-the-opposite has to be triggered by something; evaluating your own argument as if it were a stranger's; keeping score of your forecasts; admitting a mistake in one go rather than conceding it inch by inch; the value-of-information rule for when to stop searching; and planning a line of retreat before the evidence comes in. Each has at least one supporting result cited in the notes.

Verdict

Fable's pass over Book II is better than its pass over Book I. It found more real errors, and it found them in the book's evidence, not only in its anecdotes. The thirty corrections listed at the top should be made. As an assessment of the arguments, however, the pass repeatedly stopped at the first study. Of the 233 notes, about thirty land and about twenty more find a real overstatement. Most of the rest are answered by a sentence in the essay, by the source the note cites, or by the next result in the same literature. Fable counted twenty-five seriously wounded essays. The count here is closer to eight: Priming and Contamination, Of Lies and Black Swan Blowups, The Robbers Cave Experiment, Just Lose Hope Already, Entangled Truths, Mere Messiahs, Crisis of Faith and Knowing About Biases Can Hurt People. In each of them the example was lost and the lesson survived, supported by better evidence than the essay originally cited.

Fable's closing paragraph says the book "passed its own test in the places listed under 'survives' and failed it everywhere else." Applying the book's own tests to that verdict gives a different result. Check the citation: Stanovich 2008, Ditto & Lopez, and Johansson 2014 say something different from what the notes report. Argue against the strongest form: the strongest form of each essay is usually a sentence or two after the passage the note quotes. Say oops: done here, about thirty times. What remains is overconfident in its asides and careless with its history. It is also right about how minds resist change, and right about most of what to do about it.

As before, this is a partisan brief. A reader who takes the Opus, Fable, Rationalist Opus and Old Rationalist Opus sets together will be better placed than a reader of any one review, this one included.

Top

Book