Can AI Forget?
A new model architecture called FlexOlmo is pushing the boundaries of machine unlearning by letting individuals retract their data from an AI after training without rebuilding the entire system. Traditional models bake data in permanently; FlexOlmo partitions training so that users can later remove their data’s trace. The experiment signals a shift in who controls AI memory — from developers and corporations toward users themselves. If FlexOlmo scales, it may redefine how the right to be forgotten applies in the age of algorithmic intelligence.
J.M.W. Turner, Snow Storm: Hannibal and his Army Crossing the Alps, 1812. Public domain.