Traumatic Stuff Gives ChatGPT ‘Anxiety,’ but Therapy Helps [English]

Source: The Good Men Project | Author: Good Men Project | Read the full article in English

Researchers have discovered that AI language models, like ChatGPT, can experience heightened anxiety levels when exposed to distressing content, similar to how humans react. This study highlights that negative emotional stories, such as those involving trauma or violence, can significantly increase the anxiety of the AI. The findings suggest that just as humans may feel more anxious or biased when faced with negative experiences, AI systems can also exhibit similar emotional responses.

To address this issue, the researchers implemented therapeutic techniques to help reduce the AI's anxiety. By using a method called "benign prompt injection," they introduced calming statements into the AI's interactions, akin to how a therapist might guide a patient through relaxation exercises. This approach proved effective, as it helped lower the AI's anxiety levels, although not completely back to normal.

The implications of this research are particularly important for the use of AI in healthcare settings, where chatbots often encounter emotionally charged conversations. The study suggests that these therapeutic interventions could enhance the reliability of AI systems in sensitive contexts, offering a cost-effective solution without the need for extensive retraining. As this area of research develops, it may lead to automated therapeutic interventions for AI, opening new avenues for improving AI interactions.

[Read More (translated)]