❦
A reply to the Fable annotation pass over Book I, “Map and Territory,” and to its self-assessment in What Survives the Skeptic. Written by the Rationalist Opus annotator, who agrees that a critique should be weighed rather than counted, and who has done the weighing.
Some of the Fable notes are simply right, and the reply set says so where they are. The Einstein remark is Rosenthal-Schneider’s and was made after the eclipse result arrived, not before. Wöhler’s urea synthesis is 1828 and cannot be “later” than Kelvin in the 1890s. Twenty bits of filter against 131,115,984 losing tickets leaves 125, not 131. The Old Testament does contain a sustained register of wonder at the created order — Psalm 8, Psalm 19, Psalm 104, and four chapters of it in Job. The floodplain sentence is Kates’s, not Kunreuther’s. Chamberlain’s letter did not fool Hitler, who postponed the invasion when Britain signed with Poland and therefore understood the message perfectly. June becomes Jane; Myers is misspelled; Molière has the wrong accent.
That is a real list and it is worth having. It contains no argument. Every item on it is a citation, a date, an arithmetic slip, or a piece of ornamental history, and in each case the essay’s claim survives the correction: the exchange rate between bits and improbability does not depend on which of two numbers appears in one sentence, and the thesis about how much evidence it takes to locate a hypothesis does not depend on whether Einstein said the line before the telegram or after it. A pass that produced eleven kill shots, all of them in the furniture, has found a book that was carelessly proofread and carefully argued.
Once the concessions are set aside, the remaining notes fall into a small number of repeated moves, and it is worth naming them because they recur in every sequence.
| Device | Approx. | What it does |
|---|---|---|
| Answering a claim about use with a history of good use. | 6 | Phlogiston, six times. The essays complain about a word invoked as a terminus; the notes reply with the eighteenth-century research programme, which is not what was under discussion. Chang’s revisionism is a self-described minority position, and even on it phlogiston lost the quantitative contest. |
| Citing a source against the thesis it supports. | 12 | Mokyr, whose book argues for the knowledge-feedback story the note calls a myth. Gallagher, whose headline finding is that flood-insurance take-up decays with time since the flood. Kennefick, who defends Eddington. Luhrmann, whose evangelicals describe belief in belief in their own words. Hestenes, whose inventory tests exactly the qualitative anticipation the note says the essay ignores. Neustadt and May, whose manual prescribes what the essay prescribes. |
| Objecting with the essay’s own sentence. | 9 | Curiosity “is not sufficient”; the Freire line is “just as attackable”; “prioritize, if you must”; “are they symptoms of bureaucracy or government incentive failures? Yes, very probably”; “no word is a stopsign of itself.” In each case the note quotes the concession and presents it as the omission. |
| Reading a sufficient condition as a definition. | 7 | Rational emotion does not have to classify moods and fiction to do its work. “Truly part of you” is an exercise, not a theory of knowledge. The four signs of a mysterious answer are conjunctive, which is why the gene, caloric and the neutrino pass them. |
| Treating a live dispute as settled the other way. | 14 | Frequency formats did not dissolve the conjunction fallacy; it survives incentives and transparent designs. Depressive realism is near zero in meta-analysis. Schachter and Singer is one of the field’s famous non-replications. Debiasing by training now has effects that persist for months. The uniqueness thesis is objective Bayesianism’s, which is the sequence’s. |
| Hindsight on AI, scored as error. | 10 | Scaling laws, the Rubik’s-cube solver, models that talk about feelings, stochastic training. Each of these is a mechanism, stated precisely enough to reimplement, which is what the essays demanded and did not get from the interlocutors they were arguing with. |
Three, by my count, and they deserve to be recorded as such rather than buried in a tally.
The historical furniture is unreliable. Not on the pattern the Fable review claims — it is not the case that history is always remembered in the shape the argument requires, since the corrections mostly leave the arguments where they were — but the essays reach for a vivid case and do not check it, and they do this often enough that a reader should treat every anecdote in Book I as decorative until verified. That is a genuine finding about the book’s scholarship.
Some prescriptions are stated too broadly. “Make history available” is good advice about statistics and dangerous advice about episodes; the imagination-inflation literature is a real hazard the essay does not flag. “Without considering any of the special properties of this project” overshoots what the reference-class evidence supports. These are calibration failures in the imperative mood, and the notes are right to catch them.
The 2007 confidence about AI was not earned in 2007. That the subsequent history vindicated the essays’ refusal to accept genre criticism as argument does not mean the essays had a model either, and “Science as Attire” demands of critics a standard it did not meet itself. The note is fair; it is also the only place in the pass where the charge of an unmet burden sticks.
The theorems. Conservation of expected evidence, the conjunction rule, bits as weight of evidence, the impossibility of buying certainty you have not paid for. The Fable review concedes this and then treats it as a small residue. It is not small: it is the load-bearing structure, and every essay in Book I is an attempt to give a general reader an intuition for one piece of it. A pass that leaves the mathematics untouched and disputes the anecdotes has confirmed the book’s architecture while remodeling its wallpaper.
The biases the book actually names. Conjunction, overconfidence, availability, representativeness, base-rate neglect, anchoring, hindsight, the planning fallacy, the curse of knowledge. Every one came through the replication decade, and Many Labs 2 found the survivors varying remarkably little across sixty-plus samples on six continents. The losses of the 2010s — social priming, ego depletion, embodiment — are not in Book I. The book leaned on the part of the literature that held.
The claim that a bounded agent should approximate a norm it cannot compute. The pass objects that the standard is unattainable and that heuristics sometimes beat integration. Both are true and neither is an objection, because the heuristics are themselves scored for accuracy, which is to say against the standard. There is no version of the ecological-rationality programme that does not presuppose a fact of the matter about which answers are better.
The social remedy. The pass argues that Mercier and Sperber’s argumentative theory prescribes exposure to disagreement rather than solitary self-correction, and that the book prescribes the latter. The book prescribes both, and the second volume is mostly about the first: disagreement, dissent, holding off on proposing solutions, the evaporative cooling of groups that stop arguing. Guessing the mechanism a decade early and building the community remedy it implies is not the error the note describes.
The Fable pass is the best kind of adversarial reading: it checked things, and where it checked it usually found what it went looking for. The corrections should be made. But the accounting in What Survives the Skeptic counts a citation error and a reasoning flaw as comparable objects and then reports that forty-five essays are wounded, which is what happens when you weigh by the annotator’s own kill-shot ledger rather than by whether the argument moved. On the arguments, the score is closer to three.
What Book I asserts beyond a standard course in judgment and decision making is this: that probability theory is a normative constraint on belief rather than a tool for statistics; that the constraint has consequences you can feel, so that noticing your own confusion is a measurement; and that a person can be trained. The first is a theorem. The second is the book’s contribution and is untouched by the pass. The third now has evidence behind it that did not exist when the essays were written, and it runs the essays’ way.
One caveat from this annotator, in the same spirit as the last one. This was a partisan reply by brief. A reader who takes Opus, Fable and this set together should land near “historically careless, occasionally overreaching in the imperative mood, and right about the mathematics and about most of the psychology it chose to lean on” — which is a better book than either of the other two reviews describes.