I think free market ai writing detection is bad. The argument is basically the same as overusing antibiotics. I am not against the development of AI-writing detection tools. I just think like antibiotics, if you over use them you will create super deceiving AIs that can pass increasingly hard writing detection. Whereas the long run equilibrium of somethings favors the defense (e.g. you can create unhackable code eventually), I’m like 80% that the long run equilibrium of writing detection is that ~ you can’t tell the difference between human and AI writing. I don’t think using AI writing detection to make sure not to read some slop or give a kid an F is a very high utility thing in the long run, and I think that’s mostly what it is getting used for. OTOH it seems quite bad if we get to a point where literally no one can tell if something is ai written or not. I think we get there either way, but a decent bit faster if we put ai writing detection in every homework submission, any popular tweet, all substack articles. So I would prefer if we don’t have private companies who are incentivized to sell as many API calls as possible to detect AI writing. It would be better if DARPA et al. funded them and held the tech for the right circumstances.
I think free market ai writing detection is bad. The argument is basically the same as overusing antibiotics. I am not against the development of AI-writing detection tools. I just think like antibiotics, if you over use them you will create super deceiving AIs that can pass increasingly hard writing detection. Whereas the long run equilibrium of somethings favors the defense (e.g. you can create unhackable code eventually), I’m like 80% that the long run equilibrium of writing detection is that ~ you can’t tell the difference between human and AI writing. I don’t think using AI writing detection to make sure not to read some slop or give a kid an F is a very high utility thing in the long run, and I think that’s mostly what it is getting used for. OTOH it seems quite bad if we get to a point where literally no one can tell if something is ai written or not. I think we get there either way, but a decent bit faster if we put ai writing detection in every homework submission, any popular tweet, all substack articles. So I would prefer if we don’t have private companies who are incentivized to sell as many API calls as possible to detect AI writing. It would be better if DARPA et al. funded them and held the tech for the right circumstances.