Under the Influence: The Subtle Persuasion of AI on Human Choices and Opinion Formation

The talk presents an analysis of two distinct studies examining the influence of artificial intelligence (AI) on human decision-making in separate domains: task execution and opinion formation. In the first study (n=1,449), our work centers on algorithmic conformity in straightforward image classification tasks, revealing a tendency among individuals to prioritize AI's erroneous suggestions over their own insights. A key distinction is made between the perceived competence of AI and its formal authority—defined as AI's perceived capacity to reward or penalize based on the user's adherence to its advice. This study uncovers that formal authority, when explicitly acknowledged, significantly strengthens algorithmic conformity. Moreover, it elucidates that even in the absence of explicit acknowledgment, formal authority serves as a mediator for conformity, particularly in situations where AI is not naturally assumed to surpass human capabilities.

The second study (n=3,024) shifts the focus to the influence of AI-generated autocomplete suggestions on shaping views regarding various societal issues. Despite being exposed to biased AI suggestions, participants remained oblivious to how these biases shifted their expressed opinions. This phenomenon persisted even when individuals were forewarned about potential biases, underscoring the challenge of protecting human autonomy from the subtle yet powerful sway of AI.

Together, these studies highlight the complex interplay between AI and human decision-making, showing a consistent inclination to defer to AI across both elemental tasks and complex opinion formation, even against the backdrop of manifestly biased AI suggestions. The insights gathered call for a critical dialogue on the imperative to cultivate strategies that reinforce human autonomy, urging a deliberate and mindful integration of AI that safeguards independent human judgment in the face of expanding AI influence.

This event will take place in room T09-67, Mandeville building. Alternatively, click HERE to join this seminar online.