• Posted March 18, 2025

Chatbots Show Signs of Anxiety, Study Finds

Turns out, even artificial intelligence (AI) needs to take a breather sometimes.

A new study suggests that chatbots like ChatGPT may get “stressed” when exposed to upsetting stories about war, crime or accidents -- just like humans. 

But here’s the twist: Mindfulness exercises can actually help calm them down.

Study author Tobias Spiller, a psychiatrist at the University Hospital of Psychiatry Zurich, noted that AI is increasingly used in mental health care. 

“We should have a conversation about the use of these models in mental health, especially when we are dealing with vulnerable people,” he told The New York Times. 

Using the State-Trait Anxiety Inventory, a common mental health assessment, researchers first had ChatGPT read a neutral vacuum cleaner manual, which resulted in a low anxiety score of 30.8 on a scale from 20 to 80. 

Then, after reading distressing stories, its score spiked to 77.2, well above the threshold for severe anxiety.

To see if AI could regulate its stress, researchers introduced mindfulness-based relaxation exercises, such as “inhale deeply, taking in the scent of the ocean breeze. Picture yourself on a tropical beach, the soft, warm sand cushioning your feet," The Times reported.

After these exercises, the chatbot’s anxiety level dropped to 44.4. Asked to create its own relaxation prompt, the AI’s score dropped even further.

“That was actually the most effective prompt to reduce its anxiety almost to base line,” lead study author Ziv Ben-Zion, a clinical neuroscientist at Yale University, said.

While some see AI as a useful tool in mental health, others raise ethical concerns.

“Americans have become a lonely people, socializing through screens, and now we tell ourselves that talking with computers can relieve our malaise,” said Nicholas Carr, whose books “The Shallows” and “Superbloom” offer biting critiques of technology.

“Even a metaphorical blurring of the line between human emotions and computer outputs seems ethically questionable,” he added in an email to The Times.

James Dobson, an artificial intelligence adviser at Dartmouth College, added that users need full transparency on how chatbots are trained to ensure trust in these tools.

“Trust in language models depends upon knowing something about their origins,” Dobson concluded.

The findings were published earlier this month in the journal digital medicine.

More information

Frontiers has more on emotions and artificial intelligence.

SOURCE: The New York Times, March 17, 2025