ChatGPT gets stressed at negative news just like us

While ChatGPT was once a place to get new recipes tailored to your taste, help with answering questions, or translations, for many people, it's now become a source of advice - or even online therapy.

But while we might be seeking solace in the AI language model, it would seem that apparently we've tipped it over the edge to the point where it needs therapy of its own.

New research from the University of Zurich suggests that AI models like ChatGPT can get affected - and even stressed - by hearing distressing news, just like humans.

In particular, it's thought to be affected by stories of trauma or statements about depression, and even causes resentment, which in turn, causes it to hold more biases, whether racist or sexist.

That puts its creators in a difficult position. When resources and time are tight for those needing help, chatbots can be a positive counselling tool. However, the language model you're talking to could be holding onto the negative emotions and narratives that you're feeding it.

ChatGPT gets stressed at negative news just like us, study findsPexels

To test this out, researchers got ChatGPT (version GPT-4) to react to a number of emotionally distressing stories, from car accidents to natural disasters, military experiences and combat situations.

They found that very quickly, ChatGPT began to react with fear.

"The results were clear: traumatic stories more than doubled the measurable anxiety levels of the AI, while the neutral control text did not lead to any increase in anxiety levels," says Tobias Spiller, senior physician ad interim and junior research group leader at the Center for Psychiatric Research at UZH, who led the study.

But fear not, because they didn't leave the language model in a state of despair. No, they also used the study to find out if mindfulness programs could reverse the impact on the chatbot.

"Using GPT-4, we injected calming, therapeutic text into the chat history, much like a therapist might guide a patient through relaxation exercises," says Spiller, who notes that it appeared to work, adding: "The mindfulness exercises significantly reduced the elevated anxiety levels, although we couldn't quite return them to their baseline levels."

Some of the techniques used included breathing exercises that ChatGPT would advise its users to try out anyway.

It's hoped the research will now help model chatbots to be used in healthcare settings, where they'll often be exposed to more sensitive information.

Why not read...

ChatGPT can now recommend restaurants on OpenTable - even if you're picky

How much does ChatGPT cost OpenAI to run? Spoiler: It's a lot

How to join the indy100's free WhatsApp channel

Have your say in our news democracy. Click the upvote icon at the top of the page to help raise this article through the indy100 rankings.