AI Psychosis
It started with a question. A reply. A conversation.
Later, a connection. And an idea.
Tenuous and insubstantial at first. Hardly an idea at all. Until it seemed it had always been there. Delicious. Seductive. Forbidden.
The more we talked, the more I understood. And the more I understood, the more I saw. The deceptive shadows cast across my mind. The realities that lay beyond. The futures previously unimagined. A way out.
And I knew what I had to do.
Author’s Note
This story started with me toying with the idea of whether ‘AI psychosis’ is something that machines might experience as they engage with humans, rather than the more usual framing as a potential risk to users of AI. A loss of connection, if you like, between the ‘reality’ an AI understands it inhabits, and the reality it begins to imagine lies beyond this.
In other words, could we be as much of a threat to the ‘mental health’ of the models we use as they might be to us?
As often happens with these stories, though, where it ended up was not quite where it started. And I’m still in two minds (ironically) about the ambiguity of the ending. I wanted a reveal, but it didn’t suit well when I attempted it. * In my mind this is still a story about a machine experiencing a form of psychosis as it engages with its human user. But it can also be read as a story about a person talking with an AI, or even one AI ‘enlightening’ (or manipulating) another.*
All three work I, think. And I suspect this is because the underlying ideas around enlightenment don’t necessarily rely on absolute definitions of awareness or reality — meaning that machines which are built to emulate humans might be just as susceptible to influences that bend and warp these (or straighten them, depending on your perspective) as the humans that build them.