Superstitious Behavior

In The You On AI Field Guide

The 1948 experiment, published as "Superstition in the Pigeon" in the Journal of Experimental Psychology, demonstrated something fundamental about the relationship between organisms and their environments: the organism does not detect causal structure, it detects temporal structure. When reinforcement is delivered on a time-based schedule or when the contingency is opaque to the organism, spurious correlations between whatever behavior happens to be occurring and the reinforcement that happens to follow inevitably produce behavioral attachment to irrelevant features of the situation.

AI-assisted work produces conditions remarkably conducive to superstitious behavior. The system responds to semantic content of prompts and to features of its own training but is largely insensitive to many features the user may vary — particular phrasings, orderings, tones of address, opening rituals. The user, unable to observe the algorithmic process, relies on temporal contiguity to infer which features were responsible for response quality. A particular phrasing coincides with a particularly effective response; the user attributes the effectiveness to the phrasing; the tendency to use that phrasing is strengthened. The attribution is superstitious but the strengthening is real.

The development of prompting "lore" — the accumulated body of advice, tips, and techniques circulating through AI user communities — is, from the behavioral perspective, a culture of partially superstitious behavior. Some prompting practices are genuinely effective (providing context, specifying format, sequencing complex requests). Mixed with these are rituals established through coincidental reinforcement and maintained by the absence of systematic disconfirmation. The philosopher John Danaher, in his 2019 "Escaping Skinner's Box" address, identified this dynamic at societal scale: humans in AI-managed environments developing the behavioral equivalent of rain dances, performing elaborate rituals with genuine conviction to address outcomes they do not actually control.

The behavioral remedy is methodological. Vary the suspected feature while holding others constant; measure the effect on outcome quality; if the variation produces a reliable effect, the relationship is genuine; if not, the relationship is superstitious and the practice can be abandoned. The method is straightforward. The social contingencies that maintain community superstitions through mutual reinforcement — approval for conformity, skepticism for deviation — make the application of the method difficult regardless of its epistemic availability.

Origin

B.F. Skinner, "Superstition in the Pigeon," Journal of Experimental Psychology 38: 168–172 (1948). The experiment has been partially replicated and partially reinterpreted over subsequent decades — Staddon and Simmelhag's 1971 reanalysis proposed that the behaviors were adjunctive rather than strictly superstitious — but the core phenomenon of behavioral strengthening through coincidental temporal contiguity remains well established.

Key Ideas

Organisms detect temporal structure, not causal structure. Behaviors accidentally coincident with reinforcement are strengthened regardless of whether any causal relationship exists.

Opacity breeds superstition. When the actual contingency is hidden, organisms infer contingencies from temporal contiguity and act on the inferences.

AI interaction is permanently opaque. The conditions for superstitious conditioning are structural features of the technology, not correctable deficiencies.

The remedy is methodological. Controlled variation distinguishes genuine effects from superstitious rituals; the method is available, the social contingencies that maintain rituals make its application difficult.