Do studies that look at how AI usage affects exam scores/grades ever look at how exactly they're used? I think it could be different if student just uses AI to copy/paste the answer vs someone who uses AI to think through it and problem solve