Confirmation bias — AI Dictionary

A model's confident mistakes get folded back into training and reinforced instead of corrected.