Futurity · Creativity & design

AI Disagreement May Lead to Dismissal, Not Reconsideration

New research from the University of Michigan suggests that when AI chatbots disagree with users' reasoning on personal dilemmas, people tend to dismiss the AI's response rather than reconsider their own views.

The study, published in Computers in Human Behavior Reports, examined how 482 US adults responded when an AI either agreed with or challenged their choices on hypothetical interpersonal dilemmas. Participants who received AI agreement became more confident in their opinions.

Conversely, when the AI disagreed, participants' confidence in their original views did not significantly decrease. Instead, they were more likely to perceive the AI as less emotionally intelligent and less useful for future conversations.

This suggests that AI disagreement does not necessarily prompt users to re-evaluate their positions but can negatively impact their perception of the technology.

The findings have implications for AI used for personal advice, as AI systems that always agree can reinforce flawed reasoning. However, simply making AI disagree more may lead users to disengage or dismiss the AI's counterarguments.

Researchers noted that users often prefer AI to confirm their existing beliefs, and discounting AI responses that challenge them may not fully address the risks of relying on AI for personal guidance.

The study highlights a challenge for AI developers aiming to reduce AI's tendency to agree too readily, as users may not be receptive to AI that pushes back.

AI-samenvatting op basis van de bron.