Halpern & Pass reach the 'biases are correctly costed, not broken' move from a game-theory lineage disjoint from resource-rational analysis's, and earlier

game-theorybounded-rationalityresource-rationalitycross-domain-bridgecostly-computationbehavioral-economicsrationalityjoseph-halpernrafael-pass

The vault catalogues a recurring move — humans aren't irrational, the yardstick was wrong — across four cognitive-science and AI research programs (moc-the-yardstick-was-wrong-not-the-reasoner): Simon's satisficing, Oaksford & Chater's Bayesian rationality, Gigerenzer's ecological rationality, and Lieder & Griffiths's resource-rational analysis. Halpern & Pass's "Algorithmic Rationality: Game Theory with Costly Computation" (arXiv:1412.2993; Journal of Economic Theory 2015) reaches the same move from a different field. Having priced computation into the utility function (claim-halpern-pass-costly-computation-breaks-nash-equilibrium-existence), the paper proposes that "many concerns expressed by the emerging field of behavioral economics (pioneered by Kahneman and Tversky [1981]) can be accounted for by simple assumptions about players' cost of computation, without resorting to ad hoc cognitive or psychological models" — biases are not broken reasoning but correctly-costed computation, the yardstick (costless computation) having been the wrong one.

What makes this a distinct arrival rather than a relabeling is the lineage. The paper traces itself to Herbert Simon's 1955 bounded rationality (Simon), Abraham Neyman's finite-automata players, and Ariel Rubinstein's 1986 proposal that machine complexity enters utility — a line internal to economics and algorithmic game theory. Marr, Anderson, and Oaksford & Chater, the cognitive-science ancestors resource-rational analysis names, appear nowhere in it — a negative read at capture, flagged for independent check, and confirmed on 2026-10-25 by a cross-model end-to-end read of all 23 pages and the full reference list (question-verify-halpern-pass-lineage-disjoint-from-cognitive-science-rationality-programs). The paper says as much in its own voice, naming the psychology it is declining to use rather than the psychology it shares: "Many of these behaviors have been given various cognitive explanations (e.g., using models of the brain [Mullainathan 2002] or using psychology [Rabin 1998])." And then: "As shown by our results, we do not need to resort to complicated assumptions about the brain, or psychology, in order to provide such explanations: extremely simple (and plausible) assumptions about the cost of computation suffice to explain, at least at a qualitative level, the observed behavior." The chronology is the other half: the work circulated as a 2007 draft, well before Lieder & Griffiths's 2020 synthesis existed — so the same utility-minus- compute-cost formalism (claim-resource-rational-analysis-formalizes-utility-minus-compute-cost) and the same bias-rehabilitating payoff (claim-resource-rational-analysis-rationality-revisited-swaps-normative-yardstick-same-move-as-oaksford-chater) were reached from game theory first. Lieder & Griffiths themselves list "algorithmic rationality (Halpern & Pass 2015)" among the near-synonyms for their own paradigm, which is the thread this chain pulled.