❦
This is a summary of the current Rationalist Opus pass over Book I, "Map and Territory." The pass replies to every one of the 161 Fable notes in the 44 essays that carry them, and to Fable's own assessment in What Survives the Skeptic. It was written independently of the older Rationalist Opus rebuttal, which answered 126 of the notes; the two reach similar verdicts by partly different routes.
These points are conceded, in whole or in part, in the notes where they arise:
Every item on this list is a citation, a date, a spelling, a number in an illustration, or a flourish of history. None of them is an argument the essays depend on. The book needed a fact-checker. It did not need a different epistemology.
The pattern that recurs most often is simple: the objection is answered by a sentence in the essay itself, usually within a paragraph of the passage the note quotes. The following examples are not the only ones:
A second, smaller pattern is misreading what happens in a parable. In A Fable of Science and Politics, no character shows a backfire effect. All six accept that the sky is blue and then differ in what they do next, which is what the correction literature the note cites actually found. The Bayesian Judo guest moves from "AI is impossible" to "it couldn't be emotional" to "nothing made can be eternal," which is not how someone holding Searle's considered view behaves. The dragon-claimant's excuses always come before the tests, and they are always correct, which is the evidence that he has a model somewhere.
| What the reply mainly does | Approx. | Examples |
|---|---|---|
| Quotes the sentence the note overlooked. | 45 | The examples above; plus "Prioritize, if you must"; the instrumental-goods paragraph in "Science" as Curiosity-Stopper; "I wouldn't fault a subject for continuing to invent hypotheses"; the essay's own concession on incentive failures in Planning Fallacy; the snow-is-white account of what truth is for. |
| Shows that Fable's source supports the essay. | 30 | Cameron's Circus Factions treats the Blues and Greens as sporting organizations, not parties. Gallagher explains the insurance pattern by forgetting, which is availability. Shtulman finds acceptance of evolution without understanding, which is attire. Klayman & Ha, together with the DAX–MED result, show that looking into the dark works. The sapphire know-how of Collins's study did spread, by a visit. Schaeffer et al. dissolved "emergent abilities" once the measurement was modelled. Rehder & Burnett explain Markov violations with an extra causal node. |
| Brings evidence the note omits, where the dispute is live. | 24 | Sides et al. on the conjunction fallacy surviving real bets. A preregistered null on depressive realism. Ranney & Clark on mechanistic explanation moving acceptance of climate science. Harvey's 1979 re-reduction of the Sobral plates. Kamin & Rachlinski on hindsight in negligence cases. Sloman & Rabb on borrowed understanding. Evans's Madagascar case against the causal-theory reply. |
| Corrects a technical misreading. | 11 | Mutual information as a necessary condition. Correlation cannot rescue a conjunction. The pundit's forecast is about variance, not direction. The log score is the unique local proper scoring rule. "I don't know" as a broad prior that a hypothesis must beat. Mixed strategies as the output of lawful calculation. |
| Rereads a parable's events. | 9 | The six sky branches, the dinner guest, the dragon's excuses, the pagan panelist's own words, Marcello's inference, and the speaker whose own answer was "I don't know." |
| Rereads AI hindsight as mechanism. | 7 | Scaling laws, DeepCubeA, parameter counts in the Thor comparison, SGD noise that full-batch training matches, and a test of emotional talk that has never been run on a model not trained to imitate it. |
The counts are approximate, and a note can fall under more than one heading.
Rhetorical absolutes. "Not one single passage." "Since the days of Socrates." "Always right. Every single time." "A uniquely human flaw." Each absolute gives the skeptic a free counterexample. Most of the underlying claims survive when "always" is reduced to "characteristically," but the essays should have written "characteristically."
Unchecked history, again. The older pass said this, and the finding holds up on a second reading. Book I uses anecdotes the way a speaker uses slides, and at least four of them are wrong. The conclusions rest on the psychology and the mathematics, not on the anecdotes, but a reader should treat any historical illustration in Book I as unverified until checked.
Confident asides from outside the author's field. The emergency-room economics, the Chamberlain letter and the hare's cud are all asserted in passing, with no source, in essays whose subject is not being fooled by confident assertion. None of them affects the main argument, and that is also why they should have been cut.
The mathematics. Conservation of expected evidence, the conjunction rule, bits as the measure of evidence, and the claim that the evidence needed to locate a hypothesis is its prior improbability. Fable disputes some applications of these results and none of the results themselves.
The psychology the book chose. The biases Book I relies on are conjunction, availability, anchoring, hindsight, the planning fallacy, overconfidence and the curse of knowledge. All of them survived the replication decade, and anchoring and framing replicated across countries in the Many Labs projects. The effects that collapsed in the 2010s, social priming and ego depletion, are not in Book I.
The prescriptions, where they were tested. Each of these has since been tested and holds up:
The book's claim that a person can be trained to think better now has more evidence behind it than the essays did when they were written.
Fable read Book I adversarially and checked its facts, which is valuable, and the corrections listed above should be made. As a critique of the arguments, though, the pass often stopped reading one sentence too early. Of the 161 notes, about a dozen are right, and about a dozen more find a real overstatement. Most of the remainder are answered by a sentence in the essay itself, by the source the note cites, or by a result the note did not mention. The book that emerges is overconfident in its rhetoric and careless with its anecdotes. It is also correct in its mathematics, well chosen in its psychology, and borne out in its practical advice.
The same caveat as the older pass applies. This reply was written as a partisan brief, and a reader who takes Opus, Fable and both Rationalist sets together is better placed than a reader of any single review, this one included.