Enhanced Rationality in Autism: What the Biases Literature Gets Wrong About Everyone Else
Source: Rozenkrantz, D’Mello & Gabrieli (2021), Trends in Cognitive Sciences, Vol. 25, No. 8, pp. 685-696 · DOI 10.1016/j.tics.2021.05.004
Economics textbooks describe a rational agent who weighs all relevant information. Kahneman and Tversky showed real humans shortcut that job with heuristics, which is why we get framing effects, sunk-cost traps, and the rest. A 2021 MIT review asks a sharper question: who is the “human” in “predictably irrational”?
The review’s Table 1 lists seven bias domains where autistic participants often outperform neurotypical controls: less intuitive reasoning (more deliberative answers on problems like the $1.10 coffee-and-pastry riddle), reduced conjunction fallacy when stories feel representative, stable choices despite decoy options (attraction effect), less sunk-cost chasing, smaller framing effects, symmetric belief updating without optimism bias, and higher acceptance of economically beneficial “unfair” offers in the ultimatum game (about twice the rate in some studies).
The pattern is not “autistic people are better at probability.” When conjunction puzzles lack salient stories, autistic and neurotypical groups err similarly. The edge is resisting vivid but irrelevant information, not mastering formal logic. The same holds for item descriptions without social context: enhanced rationality still shows up, so this is not just “social naivety” in disguise.
Reward and emotion matter, but the story is nuanced. Framing effects shrink in ASD; galvanic skin response to gains versus losses is flatter in one study; controlling for alexithymia does not erase the framing advantage in another. Yet probabilistic learning is mixed: some papers find better long-run learning, and a 2020 meta-analysis found no consistent Iowa Gambling Task advantage. The review treats reward-based rationality as real in several paradigms, not as a universal law.
Two other mechanisms probably stack. Weak central coherence and predictive-coding accounts both predict less weight on context and priors, more weight on raw details, slower Type 2 processing. That maps onto dual-process findings: autistic participants and people high in autistic traits (measured by questionnaires like the Autism Quotient) report and show less fast intuition. Heuristics buy speed for neurotypical minds; skipping them may buy accuracy at a cost the paper still flags as open.
Social life is not a clean win. Autistic people may know race and gender stereotypes explicitly yet show reduced implicit bias on some measures. Higher autistic traits in nonclinical samples predict better predictions about crowd-level social psychology, mediated by systemizing. Moral judgment is messier: more weight on actions and outcomes than intentions, which the authors refuse to call clearly more or less rational.
You cannot generalize from the lab to the whole spectrum. Box 4 is explicit: enhanced-rationality studies need verbal fluency, instruction-following, and complex tasks, so they oversample cognitively matched participants. Adults with intellectual disability or minimal speech are underrepresented. Autistic adults also self-report real-world decision difficulties. Lab rationality and daily functioning are not the same thing.
The authors frame stakes beyond academia. Nearly half of autistic 18-year-olds in the US do not hold a paying job until 25, the lowest employment rate among diagnostic categories in the statistic they cite. They argue bias-resistant reasoning could be an asset in auditing, data checking, contract review, and other roles if employers map strengths instead of treating autism as deficit-only.
The takeaway: Stop treating Kahneman-era bias lists as human nature. They describe a common neurotypical default under emotion, reward pressure, and context. If you design consent forms, pricing, or hiring tests around those biases, test whether they hold for your actual users. And if you work with autistic people, ask where careful, decoy-resistant reasoning is an asset, not only where social intuition is expected.
Related TMFNK Content
- Predicting loss aversion behavior with machine-learning methods Loss overweighting is one of the biases this review reports autistic participants often resist.
- Why You Can’t Think Under Pressure: Stress Shatters the Brain’s Memory-Linking System Another paper on how context and affect reshape cognition beneath “rational actor” stories.
- Plasticity and Language in the Anaesthetized Human Hippocampus Brain-level evidence that minds differ in how they process information, not just in how well they score on quizzes.
Crepi il lupo! 🐺