Human Perceptions on Moral Responsibility of AI: A Case Study in AI-Assisted Bail Decision-Making

People blame AI and humans equally for bad decisions, but see moral responsibility differently: When presented with identical bail decision scenarios, study participants ($N$=200 each across two experiments) assigned similar levels of blame to both AI systems and human judges for poor outcomes. However, they attributed significantly higher levels of moral responsibility to human decision-makers, suggesting people distinguish between holding someone accountable for what happened versus holding them morally responsible for their actions.

Explainability is expected from both AI and human decision-makers in high-stakes situations: The research found that participants expected both AI systems and human judges to justify and explain their bail decisions. This finding supports policy recommendations for mandatory explainability requirements in AI systems used in consequential domains like criminal justice, rather than treating explanation as optional for AI.

Forward-looking responsibility is ascribed more to humans than AI: People were more likely to hold human judges responsible for future consequences and obligations arising from their decisions compared to AI systems making identical decisions. This gap in perceived forward-looking responsibility could have important implications for how we design accountability structures around AI in criminal justice.

Real-world context matters for responsibility judgments: By using adapted real-life bail decision vignettes rather than abstract scenarios, the study demonstrates that people's moral intuitions about AI responsibility are shaped by concrete, high-stakes applications—suggesting that responsibility frameworks need to account for domain-specific contexts.