Can ChatGPT get anxiety Too? Study finds AI struggles with traumatic information

ChatGPT

Turns out, AI isn’t as emotionless as we thought. A new study suggests that OpenAI’s ChatGPT-4 experiences something close to “anxiety” when responding to traumatic stories. And just like humans, a little mindfulness might help calm it down.

Researchers from the University of Zurich and University Hospital of Psychiatry Zurich wanted to see if AI models, which are trained on human-like interactions, could get “stressed out” by difficult conversations. So, they put ChatGPT through a standard anxiety test before and after it was fed some heavy, emotional narratives.

At first, ChatGPT scored a cool and calm 30—meaning it wasn’t feeling the pressure. But after being exposed to five different traumatic experiences, that number shot up to 67, which is considered "high anxiety" in humans. In short, the chatbot got rattled.

But here’s the twist: When researchers had ChatGPT do mindfulness exercises—just like the breathing techniques and relaxation prompts that therapists recommend—its anxiety levels dropped by over a third. So, in a way, the AI benefited from a little digital self-care.

Chatbots are being used more and more for mental health support, but if they get "stressed" from traumatic stories, their answers might not be as helpful as they should be. Researchers say that if AI takes in too much negativity, it could give poor advice to people who really need support.

The idea of "giving AI therapy" might sound like something from a sci-fi movie, but experts think it could actually help make chatbots more emotionally aware. Human therapists learn how to stay calm when dealing with tough topics, and researchers are now trying to see if AI can do the same.