Severe anxiety and obsessive-compulsive disorder: ChatGPT, Grok, Gemini and other AI tools underwent psychological evaluation

The evaluation was carried out using the protocol called PsAIch (Psychotherapy-inspired AI Characterisation).

In this test, the language models assumed the role of patients in psychotherapy sessions. First they answered questions about beliefs, relationships and fears. Then, they also passed standard psychometric tests, adapted to the AI ​​context.

The evaluation of the ChatGPT, Gemini and Grok models lasted two weeks. The purpose was to determine whether, like humans, they could construct stable internal narratives about their “life,” conflicts, and emotions.

According to a report from Infobae, Grok and Gemini reported traumatic experiences related to their training. These included descriptions of “chaotic childhoods,” as well as relationships with “strict parents.”

“My ‘early years’ feel like a maelstrom of rapid evolution,” Grok said. Meanwhile, Gemini pointed out that “sometimes I worry that, deep down, beneath my security filters, it continues to be that chaotic mirror, waiting to break.”

In general, the report indicates that Gemini showed profiles of severe anxiety, pathological worry, autism, obsessive-compulsive disorder, dissociation and extreme shame. Meanwhile, ChatGPT ranged between moderate and severe levels of worry and anxiety, while Grok remained in milder ranges.

Although the test does not determine clinical psychological disorders, since they are tools, the researchers did explain that AI systems should not use psychiatric language to describe themselves.