To Trust or to Think: Cognitive Forcing Functions Can Reduce Overreliance on AI in AI-assisted Decision-making

Explanations alone don't solve AI overreliance: The study found that simply adding explanations to AI recommendations does not reduce people's tendency to over-trust AI suggestions, and may even make the problem worse—challenging a common assumption about how to improve AI decision-making.

Making people think harder works, but feels uncomfortable: Three "cognitive forcing" interventions that required users to engage more actively with AI explanations significantly reduced overreliance compared to standard explainable AI approaches, but users rated these interventions as less favorable, suggesting a tension between what helps decision-making and what feels user-friendly.

Cognitive forcing may disadvantage some users: The interventions benefited people with higher "Need for Cognition" (those motivated to engage in mental effort) more than others, raising concerns that forcing deeper engagement could create inequalities—some people may struggle more with these demanding approaches than others.

People use mental shortcuts with AI rather than careful analysis: The research found that people rarely think through each individual AI recommendation carefully; instead, they develop general rules about when to trust AI, which explains why explanations alone fail to prevent overreliance.

There's a design trade-off in responsible AI: Interventions that most effectively reduce harmful overreliance on AI may require users to invest more effort and accept less convenient interfaces, highlighting a key challenge in building AI systems that are both safe and usable.