In-Chat Human Crisis Support for AI Chatbots

I don’t like to make the subject of this blog too heavy, but I’ve been digging through some weighty material on heavy subjects around deaths associated with AI chatbots. That Wikipedia link shows many, many examples, and I’ve seen countless others scattered across news reporting, and various incident trackers. The severity of the problem is much worse than I originally anticipated when I started down this rabbit hole. Anyway, trigger warning as this post talks about suicide. One of the pivotal research papers on the more broad theme of AI-induced harms (not necessarily always resulting in death) has been the Stanford SPIRALS paper, whose official title is “Characterizing Delusional Spirals through Human-LLM Chat Logs” (PDF). It is a relatively small sample as these things go, but for published studies in this space, it is one of the more comprehensive and well-thought out ones available. And it includes what I thought was a very interesting policy recommendation: Having crisis…

Commenting on Bubbles has been removed. Why? If you want to talk about this post, write to the author, use the comments on their blog if it has them, or reply with a "RE:" post on your own blog. Suggestions? feedback@bubbles.town