This doesn't seem like a good cost-benefit analysis for AI. Not sure what that would look like, but it seems like making some attempt to quantify the benefits would help.