AI chatbots show distinct decision-making patterns in ethical and social dilemmas

Recent studies highlight complexities in how artificial intelligence interacts with human decision-making and ethics. A Penn State-led study found that AI chatbots rarely expressed hesitation when tasked with choosing between two hypothetical patients for a single kidney transplant, even when offered a coin-flip option. Unlike humans, who often express indecision in such scenarios, AI models tended to fixate on specific traits, such as age or alcohol consumption, rather than balancing multiple complex factors like the number of dependents.

Complementary research from the University of Michigan examined how users respond to AI disagreement in social dilemmas. The study found that while AI agreement can reinforce a user's existing confidence, users often dismiss AI responses that challenge their reasoning rather than reconsidering their own views. This suggests that efforts to make AI more willing to disagree may not effectively mitigate the reinforcement of flawed reasoning or distorted perceptions of reality.