Three Ways of Thinking About "AI Psychosis”

a reading/viewing list

Chatbots are having unintended consequences on the mind. The media has mostly been reporting on this phenomenon through an old, familiar trope: the rare horror of madness, which is a trope I find a bit odd, considering that it’s not actually that rare.

In a global study from 2016, 5.8% of people reported having a psychotic experience at least once in their life, and 2% in the prior year. OpenAI’s most recent numbers claim that .07% of users each week show signs of either psychosis or mania, which is not really surprising in context. Actually, it seems a bit low!

All kinds of really ordinary stuff can trigger a psychotic experience — not getting enough sleep, taking certain drugs, being super stressed out, giving birth, losing a loved one, and even just spending 25 minutes in a sensory deprivation tank. Brains make patterns, so of course linking your brain up to a statistical prediction machine that can make patterns a million times faster than you can have some intense effects on your mind.

There’s a lot we don’t know about this phenomenon, though. Journalists have been calling it ‘AI psychosis’, but experts seem kind of annoyed by that term, because it’s not really the best description of what’s going on.

Almost all reported cases solely involve delusions (false beliefs), which is just one part of the psychosis constellation. And not everyone who’s experienced these delusions are using a pathological lens to understand it — for some, it’s akin to a spiritual experience or a psychedelic trip.

Quite a few papers have come out in the last few months on ‘AI psychosis’, so for paid subscribers, I’ve compiled a few of my favorites, supplemented with some good video essays, a lived-experience podcast project, and a relevant Mad Studies text.