Study: AI models fail to reproduce human judgements about rule violations

Researchers from MIT and elsewhere have found that machine-learning models trained to mimic human decision-making often suggest harsher judgements than humans would. They found that the way data were gathered and labeled impacts how accurately a model can be trained to judge whether a rule has been violated.

Related Keywords

Toronto ,Ontario ,Canada ,Madras ,Tamil Nadu ,India ,Davidh Yang ,Aparna Balagopalan ,Gilliank Hadfield ,Marzyeh Ghassemi ,Dylan Hadfield Menell ,Schwartz Reisman ,Artificial Intelligence Laboratory ,Microsoft Research ,University Of Toronto ,Vector Institute ,Schwartz Reisman Institute For Technology ,Canada Research Council Chain ,Healthy Ml Group ,Computer Science ,David Madras ,Schwartz Reisman Chair ,Schwartz Reisman Institute ,Canada Research Council ,Machine Learning ,Ormative Judgement ,Achine Learning Bias ,