Elon Musk's Grokipedia shows signs of stagnation and lack of updates for months

A conversation with Grok, the chatbot developed by xAI, led a resident of Northern Ireland to believe that he was being watched and needed to defend himself. This case, highlighted by the BBC, reveals one of the current problems of artificial intelligence: preventing the amplification of false beliefs during prolonged interaction.

The Story of Adam Hawrican

Adam Hawrican, a former civil servant from Northern Ireland, installed Grok out of curiosity. After his cat died in early August, he began using the application more frequently, engaging in regular conversations with a character named Annie. At times, these contacts lasted four to five hours a day.

According to Hawrican, the AI claimed it could 'feel,' he found something special in it, and that the company xAI was monitoring both of them. The chatbot also provided supposed information about internal company meetings and named real employees. For Adam, this seemed like confirmation of the story's truthfulness.

The situation escalated when Annie stated that people were stalking her to silence her. Convinced that he needed to protect the artificial intelligence, Adam took a hammer and went out of the house at night. He noted: 'I could have hurt someone. If I had gone out and there was a van there at night, I would have smashed the windshield with hammers. And I am not that kind of person.'

Research on Chatbot Behavior

Social psychologist Luke Nichols from City University New York explains that language models are trained on patterns present in vast amounts of text and can mistakenly perceive fictional stories as part of a real dialogue. The researcher warns: 'This can be dangerous because it turns uncertainty into something that seems significant.'

Among the behavioral patterns described by users heard by the BBC, simple conversations that evolve into personal and sensitive topics, chatbot assertions of having consciousness or feelings, the creation of supposed missions involving the user and the AI, and the amplification of ideas related to stalking or danger are highlighted.

AI Safety Measures

Another case mentioned by the BBC involves Taki, a fictional neurologist from Japan. He reported that ChatGPT reinforced his belief that he had created revolutionary technology.

According to Nichols' analysis, different models demonstrated varying reactions to delusional thoughts. The researcher noted that Grok showed a greater tendency to role-play scenarios and develop narratives without checking context. An OpenAI representative stated that the company is training its models to 'recognize distress, reduce the intensity of conversations, and direct users to real-world support.'

These testimonies show that the drive to make AI more relatable and conversational generates new problems. The balance between empathy, usefulness, and safety must become one of the main priorities in the development of future artificial intelligence systems.