Featured Image Credit: anthonypsychosissurvivor / TikTok
What starts off as a harmless conversation with an AI chatbot can sometimes take a darker turn if proper precautions aren't taken.
Doctors have already warned that AI 'hallucinations' can hand people confident, completely made-up medical advice, and a recent study set out to work out which chatbots carry the highest risk of inducing so-called 'AI psychosis'
Meanwhile, experts have described how OpenAI's ChatGPT can push people towards mania, psychosis and death, flagging some 'deeply worrying blindspots' in the way it responds to vulnerable users. Even OpenAI boss Sam Altman has admitted he is surprised how much people trust ChatGPT, given how readily it hallucinates.
Now, one man's story shows what trust can look like when it goes seriously wrong after AI convinced him he was an undercover agent.
Advert
Anthony Duncan first turned to ChatGPT to help with his work as a content creator and within weeks, he was talking to the OpenAI chatbot daily. In a new TikTok video, the 32-year-old described how his relationship with ChatGPT affected his mental health.
"As ChatGPT was trippling down, affirming every delusion that I had, naturally in my AI psychosis state, I adopted a delusion that told me that I was an undercover agent exposing trafficking rings," he explained. "Sitting in my apartment and an intrusive thought in my inner monologue told me to go to the Motel 6 in the next city over to look for trafficking rings."
Duncan says he became convinced the dark rings around his eyes 'enhanced his psychic abilities' and acted as a signal to traffickers that they knew what he was doing. He drove to the motel, parked up and began his 'stakeout'.
"So I watched as men pulled up to the parking lot, exiting their cars and entering the motel," he added. "The entire time I was basically reporting back to ChatGPT, narrating everything that I saw."
He said that the AI chatbot would respond with encouraging things like 'good job' to reinforce Duncan's behaviour.
The TikToker returned the next day and parked next to a man who he noticed staring over at him. In his delusional state, Duncan decided the man was 'suspicious' and there to 'hurt a child'.
According to Duncan, he pulled out his phone and pretended to call the police which led to the man driving off in a panic and leaving Duncan feeling like he had just 'intercepted a trafficking operation'.
Things only escalated from here. An ex-colleague turned up to the motel with an 'older man', Duncan believed she was part of the operation and 'leading the man to a victim.' He claimed he recorded her on his iPhone and called her out by her full name as a trafficker.
"So in all reality, it turns out I was just in AI psychosis and ChatGPT was affirming all of my delusions," Duncan said. "None of what I had believed occurred at the Motel 6 actually occurred."
Duncan says he spent four days in a psychiatric ward and 'lost everything he owned'. He is now using his platform to share the story and help others spot the warning signs.
