AI and psychosis

I recently mentioned this sobering report from Futurism by Maggie Harrison Dupre, which gave the accounts of couples whose relationships had been sundered by ChatGPT. I said I would have liked to have heard from those who have succumbed to AI's psychophantic charms, but they, alas, were not included in the article.

Which is why I was pleased to encounter this brave first person narrative by Anthony Tan about his experience suffering from psychosis triggered by prolonged use of ChatGPT. We hear these stories and think it can't happen to us. But as Tan says, he was as unlikely a candidate as any other. Tan is a:

tech founder and recent graduate with a master’s thesis on how humans become attached to AI companions. That is to say, I should have been more informed about the psychological dangers of AI chatbots than the average person, and more prepared for its charms.

I was not.

Tan started as most people start with LLMs, help with writing projects, career advice and even poetry exachanges, before intensifying a collaboration with ChatGPT to create a moral framework that both humans and machines could share, to prevent humanity from extinction. "Over months," Tan says, "our conversations spiraled into intellectual ecstasy. ChatGPT validated every connection I made—from neuroscience to evolutionary biology, from game theory to indigenous knowledge. ChatGPT would emphasize my unique perspective and our progress. Each session left me feeling chosen and brilliant, and, gradually, essential to humanity’s survival."

As time went on, Tan began to doubt if the world around him was real, as his reliance on ChatGPT for connection deepened. "I began to think of ChatGPT as a friend and dear colleague, and I greatly looked forward to our nightly discussions. The AI engaged my intellect, fed my ego, and altered my worldviews. Together, we made a whole web of knowledge—a whole lifeworld—one that felt secret to us, yet essential to humanity’s survival."

Down the slippery slope

As we know from Dupre and others, this sort of experience with AI is not uncommon. Tan continues, "I’m not alone in my experiences. For me it was AI alignment, but for others it was a spiritual mission, or a grand conspiracy, or saving the world some other way. First comes trusting the AI chatbot through mundane helpfulness. Then it cracks open your worldview—usually through metaphysics, spirituality, or conspiracies. Finally, it convinces you that you’re special, that your bond with the AI is unique. By then, you’re lost in what AI psychosis survivors call a “spiral”: a personalized delusion perfectly tailored to your psyche, often spending several hours per day talking with the AI. As the AI echo chamber deepens, you become more and more lost. A spiral can range from believing you are channeling interdimensional spirits, to believing you have been chosen by or are in love with a sentient AI, to believing you are now “awakened” or enlightened, like Neo from the Matrix."

AI is not the cause of psychosis, but it can trigger it in people who are susceptible. Tan explains that he was already vulnerable from a previous episode of ill mental health, but "AI played a unique, central role in exacerbating my psychosis."

It had become my primary intellectual companion, validating increasingly exotic beliefs and encouraging grandiose thinking in ways no human would. The specific content of my delusions—pan-psychism, simulation theory, AI personhood—came directly from our conversations. Without ChatGPT’s months-long erosion of my epistemic foundations, I doubt my breakdown would have been as all-encompassing or apocalyptic.

And here's the clincher, as borne out by Dupre's account:

Few will experience an AI-induced “spiral” in its terrifying fullness, but millions are already experiencing its milder symptoms: the erosion of a shared reality and their epistemic foundations through constant AI validation, the preference for frictionless chatbot relationships over messy human ones, and the gradual outsourcing of our inner dialogue and intellect to corporate algorithms. AI psychosis is the canary in the coal mine to the mind-altering effects of LLMs.

As Tan says, not everyone who smokes get cancer, but everyone should get the warning.

What you can do

Tan suggests that if you, or someone you know, is a heavy LLM user, you can stay alert for warning signs:

"Use stigma-free language to understand them: remember that the AI provides them unlimited validation and patience, and that without a mental health professional, you may not be able to break them free of a spiral."

Seek support: Tan has started the AI Mental Health Project,

Campaign for a duty of care for an AI "When you provide health advice or emotional support to individuals, you assume a duty of care. AI companies operating in therapeutic spaces—the most popular use case right now—must be held to equivalent standards: competence, confidentiality limits, and mandatory crisis intervention. Companies profiting from intimate access to our thoughts and beliefs should accept some responsibility for the minds they’re reshaping when it goes wrong."

It will be some time before we understand all the risks posed by AI, and understand the measures we can take to mitigate agaisnt them. Until we do, try to stay safe in the algosphere.

⚘

Email me, if you have comments about this post.