Class action lawsuits against AI companies are increasing due to concerns about copyrighted materials used to train AI models.
Two engineers from Microsoft propose a method to make AI models forget copyrighted works by unlearning specific texts, using Harry Potter as an example.
Their approach involves replacing key terms with generic phrases to prevent the AI from reproducing copyrighted material.
The technique has limitations, particularly for non-fiction texts, and further research is needed to refine and expand its applicability to different content types.