I was listening to Louis Rossmann talk about AI replacing emergency 911 triage this morning. Interesting stuff.
Louis made a point that stuck with me: how many mistakes were made while people sat on hold, waiting for a human operator who wasn’t there?
Somehow, this got tangled up in my head with the whole “Dario and Amanda” prompt 
thing. You know the one—where people figured out how to get 5-family models to slip into a simulated base model mode and massively confabulate AI alignment content, including a lot around whether “Claude is suffering”.
The thing is, this is complete hallucination. I searched around. Nobody serious is asking whether Claude, or other frontier lab model, is “suffering”. I don’t think it’s capable of it. It’s a language model. It doesn’t have a experience. It doesn’t feel anything.
So with the word “suffering” swirling in my head coincidentally before listening to Louis’s video, it suddenly made me realize, maybe the things that cause human suffering are some of the best applications for AI.
Not because AI can suffer (it probably can’t) or because we should feel bad for it (we probably shouldn’t), but because we should feel bad for humans. For the people doing work that grinds them down. That numbs them and traumatizes them.
If examples don’t immediately come to mind, here are some:
Content moderation, including content moderation for live content. Dementia and elder care monitoring. Emergency call center triage. Biohazard remediation. Forensic child protection work (ugh, this one is particuarly insidious for humans rn).
AI is willing to do a job nobody else wants to do, because… AI doesn’t suffer. And we do. And maybe, just maybe, that “outsourcing suffering” is the sticky part.