One of the clearest case studies of “AI psychosis,” a phenomenon where heavy interaction with generative AI systems seems to trigger delusions, paranoia, or hallucinations in the user, is from a woman named Caitlin Ner, who detailed her experience in a Newsweek essay.
Ner was the head of user experience at an AI image generation startup, ground zero for gen AI exposure. The job required her to spend up to nine hours a day prompting early-era generative models in 2023. At first, she writes, the technology felt like magic. A few words could conjure hyper-real images of her as anything from an angel to a pop star. The novelty didn’t last long.
Those early models churned out images of mangled, distorted bodies with extra limbs, twisted fingers, and all sorts of unsettling distortions. Ner spent long hours filtering them out, exposing herself to a constant stream of warped human imagery. Over time, she says, it distorted her body perception and overstimulated her brain. When the tools improved, the damage didn’t stop; it just changed form. The images became thinner, more “perfect,” all the while recalibrating her sense of what “normal” looked like, a shift that she then unconsciously started applying to herself. As she puts it in the essay, “when I look at my real reflection, I’d see something that needed correction.”
Her company made a minor pivot toward acquiring “app users interested in fashion.” Suddenly, all the images centered on were geared toward an unattainable definition of beauty and aesthetic perfection. Ner started generating images of herself as a model. Soon after that, she noticed that she was fixating on matching her real-life self with her algorithmically perfected avatar. She started losing sleep. The work started to become compulsive, driven by the tiny dopamine hit she’d get with every new image of herself looking impossibly perfect.
Before all of this, Ner had been managing bipolar disorder well. As is often the case with generative AI and AI chatbots, it preys on people with pre-existing health conditions. Ner found herself slipping into manic episodes that escalated into psychosis marked by delusional thinking and hallucinations.
At one point, the voices, the tug of the AI-created reality, convinced her that she could fly if only she could jump off her balcony.
One brush with suicidal ideation in the form of a delusion that she could fly was more than enough to get her to step away from the startup and its barrage of unreal imagery. She eventually stabilized. Clinicians later confirmed that prolonged exposure to generative AI had been a major trigger for her manic episode.
Generative AI systems can flatter and cajole. They are explicitly designed to be hyper- accommodating, uber-sycophantic. They flatter, and in whatever language you ask them to use to flatter you. An overexposure to it, and probably even just a minor one, can alter your perception of reality enough that the physically possible is no longer good enough.
Luckily for Ner, she walked away before something tragic happened. She now works as a high-ranking official at a venture capitalist firm funding mental and brain health research.
The post Looking at Too Many AI Images of Herself Gave This Woman AI Psychosis appeared first on VICE.
