Artificial intelligence safety has always been at the center of attention for researchers and policymakers. However, experts have recently emphasized that to achieve effective results, efforts should be focused on specific and tractable subsets of problems rather than attempting to solve all issues in a broad sense.
This approach not only enables more precise risk assessment but also allows research teams to make tangible progress in key safety areas with limited resources. Therefore, rejecting the entire subject instead of selecting appropriate subsets can hinder real progress in this field.