AI models misjudge rule violations: Human vs. machine decisions
Summary: Researchers discovered that AI models fail to accurately reproduce human decisions about rule violations. They tend towards harsher judgements. The reason for this is the data that these models are trained with. They’re often labeled in a descriptive way rather than a normative one, leading to different interpretations of rules violations.
This discrepancy can have serious consequences in the real world, including stricter sentences. Researchers suggest that dataset transparency be improved and the training context matched to the deployment context in order to improve accuracy of models.
Source:
AI Models Misjudge Rule Violations: Human Versus Machine Decisions
