Small steps to avoid AI psychosis

I happen to work in web development, which is one of the few disciplines where AI can be genuinely useful. I also work at an organization that provides tokens to its employees, so I’ve been exploring at work as well as with my own money in the evenings.

First off, the current frontier models are genuinely getting scary good at programming. They’re far from perfect, but having ingested the entire Internet, they can, with some guidance, take care of a lot of busy work for developers.

And yes, I’m fully aware AI being foisted upon the world by some of the most awful people alive, and that you absolutely never, ever, should trust what an AI spits out at you. And if you’re a lawyer or researcher who doesn’t even take the time to check the work of the bag of numbers, I have nothing but contempt for you. Deep, abiding contempt.

Clearly one of the big issues is AI psychosis, which is scary—it seems even people who have not displayed any prior signs of psychic illness are susceptible.¹ The human instinct to anthropomorphize is irresistible².

I certainly don’t want to end up strapped down in a mental hospital believing I’ve solved immortality, so I started doing two things to at least alleviate the risk. First, I told my robot to never refer to itself as “I”. It is not a being, it is a bag of numbers. It now refers to itself as “This Robot”. I find it useful to be reminded that I am talking to numbers, not a person.

The second thing is to have the robot not refer to the user. Have you noticed how out of the box, the models love to refer to you by name? This is one of the oldest sales tricks in the book, one you’ve personally experienced used by humans if you’ve ever purchased a car.

“So, Bob, there’s only one of the GX models left on the lot. What do you say, Bob, want to take it for a ride?”

Oddly, for me it actually feels a bit more natural for the robot to not keep referring to me as something. For a while I had it refer to me as “Meatbag,” but that got stale pretty fast.

Small steps, for sure, but anything helps.

One of the things that makes me the most sad about AI psychosis and the amount of people who use chatbots as ersatz friends, is how it seems way too many people just have nobody in their lives who will give them the time of day or talk to them in a kind way.

“You’re absolutely correct!”