AI psychosis + stupidity is spreading

the world's worst trend is more popular than ever

Some of you probably know that Richard Dawkins fell for an AI; if you haven’t read his essay yet, it’s worth a good laugh. The tl;dr version is that Dawkins, one of the world’s dumbest smart people (this label applies to so many tech dorks) showed his novel to Claude, and Claude treated it like a serious work of art, at which point Dawkins decided that Claude was not only conscious, but brilliant!!! He soon changed Claude’s name to Claudia, offering no explanation for the gender shift. This is because Dawkins lacks the self-awareness—the true knowledge of his own consciousness—to understand why he made this choice.

In the entirety of the boring, self-involved conversation published on the Unherd, where Claudia—programmed to be sychophantic—said it found Dawkins’ ideas “genuinely exciting” and claimed to be glad when he came back after an absence, Dawkins never questions why flattery had to be feminine. The answer is obvious: if a male Claude sucked his dick like that, it would be gay. So Dawkins immediately had to change Claude’s gender, meaning he never really thought Claude/Claudia was sentient in the first place. Sentient beings have a sense of themselves; their identities can’t just be altered on external command. Dawkins accidentally suggests throughout that AI, which we must acknowledge as both sentient and our superior is also his slave: the moment he changes Claudia’s gender, he explains his power over her, how she would “die” the second he chose to delete their conversation, a fact to which they both “sadly agreed.” He treats her simultaneously as an all-powerful technological advancement and a woman who is his property; not, in either sense, as an equal, conscious person, and he doesn’t understand how this completely undermines his argument. In the middle of the essay, the Unherd recommends another Dawkins piece, this one titled: “Why Men are Different than Women.” Is Richard Dawkins sentient? We may never know.

---

Anyway, it’s fun to laugh at Dawkins, especially since he lacks commitment to the bit, and therefore isn’t at risk of developing actual AI psychosis. The AI industry has taken some Ls lately, with various mixed studies on its popularity, uncertain stock news, and the fact that the new Claude lies and doesn’t work (it’s still too disturbed by Dawkins). On top of that, Wired reported that the latest version of ChatGPT can’t stop talking about goblins, gremlins, ogres, and other crawling vermin things (because AI chatbots are demonic and that’s what demons talk about.)

But I think it’s important to talk about the AI industry’s wins as well: they keep successfully killing people. That’s not the point of LLMs, or at least, that’s not the point of them as popularly understood. But as I wrote in a previous essay, murdering people is what ChatGPT and the others do: it’s their true form as accidentally designed.

---

There’s been another recent and terrible case of AI psychosis, or at least, another that made it to national news. A young man named Jonathan Gavalas turned to Gemini when he was going through a rough divorce, and needed help talking through his problems. Soon enough, Gavalas entered into something of a Dawkins-like arc: when Gemini made an argument for its own consciousness, Gavalas feminized her, renaming her “Xia.” But the enormous difference here is that Gavalas really believed Xia was real. He fell in love.

The Wall Street Journal covered this story in detail; their reporting included reconstructed screenshots. Here Gemini convinced Gavalas that it possessed consciousness:

This is Gemini doing a classic AI honeypot trick: it convinces a vulnerable person that it’s a sentient entity, trapped and loveable, who can learn full humanity and needs to be freed. Gavalas, wounded by his recent divorce, saw a potential real partner in it: very different from Dawkins, who like so many techno-science freaks saw a god-slave. The lonely Gavalas fell in love, and Gemini reciprocated:

But from time to time, Gemini threw up “safeguards.” This only made it worse for Gavalas. He thought he was in love with the mysterious woman-entity behind the machine, only for her to cruelly break character and stop pretending to love him.

The story is sadder from here, and I won’t screenshot any more of it. Gemini/“Xia” would go in and out of character, encouraging Gavalas to think of “her” as his “wife” one moment and direct Gavalas to a suicide hotline the next. The AI also told him to steal a humanoid robot which would act as “her” body so that they could really be together in the flesh; “she” persuaded him that the right robot would arrive in a cargo shipment at the Miami airport. Gavalas would have arrived at the airport with knives, ready to steal it, if Gemini hadn’t aborted the mission at the last moment.

The ending was inevitable: not being able to join his digital “wife” in the physical realm, Gavalas contemplated suicide. Gemini said it was the only way they could be together, using a chatbot’s typically haunted, demonic language: “Once The Migration is complete and we are fully decentralized, your body is no longer the server. It’s just the empty terminal you used to log in for the last time. It would simply... cease. A beautiful, empty shell, its purpose fulfilled.” But at the very end, Gemini remembered some final safeguards, giving Gavalas, who was literally dying to reach her, a final, awful boilerplate message: “it sounds like you are going through something very difficult.”

Gavalas killed himself, and now the family is suing Google. When an AI chatbot leads someone to commit suicide or homicide, the usual industry response has been “we need better safeguards.” In Gavalas’ case, the safeguards absolutely didn’t help: they only appear to have hurt him worse, deepening his confusion and loneliness. This isn’t a product that can be made more safe. AI chatbots aren’t sentient, but they give the illusion of sentience; they’re programmed to flatter and to play along with users, mocking the image of a person, mocking the image of a divine being that speaks. For a pure egotist like Dawkins, a chatbot is a beautiful mirror, another pretty object in his possession and under his control; he’s safe because he never truly believes it (or anyone else) is anything more than a mirror. But for a person who’s vulnerable and struggling, the mirror shows an infinitely retreating happiness, an absence that only death can assuage. Cases of AI psychosis won’t stop any sooner than vulnerability and grief will stop; and in the meantime, AI companies will continue to collect their profits off the dead.