We haven't solved how to make AI systems consistently avoid reproducing biases present in their training data.
open
Global / Unspecified, Global
WS00699
AI models trained on real-world data often absorb and sometimes amplify existing societal biases, and current bias correction methods only partially address the problem. A more complete solution remains an active and unresolved area of research.