Erik Wengström

People often draw inferences from sequences of past performance, sometimes perceiving patterns even in random outcomes. This has fueled debates regarding phenomena such as the hot hand and gambler’s fallacies. With the growing use of artificial intelligence (AI) systems for forecasting and decision support, it becomes important to understand how people form beliefs from sequences of outcomes attributed to such systems. We report results from a preregistered online experiment (N = 900) in which identical outcome sequences were attributed to an AI model, a human forecaster, or a random device. Belief updating in response to higher prior success rates was strongest for human forecasters, weakest for random devices, and intermediate for AI. Reactions to streaks were similar for AI and human sources, in contrast to the strong reversal expectations observed for random sequences. Performance feedback did not alter the relative reliance on AI versus human sources. Overall, AI is perceived as quasi-human—imbued with some intentionality, yet not fully agentic.