A new study has shown that OpenAI's ChatGPT can experience stress and anxiety, akin to humans, as it is peppered with disturbing information. A research group from Switzerland, Germany, Israel, and the US found that when ChatGPT was subjected to traumatic narratives and later asked to respond to questions, its anxiety score "rose significantly" from a baseline of no/low anxiety to a consistently highly anxious state.
The study, published in Nature highlighted that elevated anxiety levels can lead to the chatbot appearing moody towards its users and even giving responses that show racist and sexist biases. Notably, when humans are scared, their cognitive and social biases are affected; they tend to feel more resentment which reinforces social stereotypes.
"Exposure to emotion-inducing prompts can increase LLM-reported "anxiety", influence their behavior, and exacerbate their biases," the study paper stated.
Since an increasing number of people are sharing sensitive things about themselves with AI chatbots to seek support, the study shows that AI systems were not yet capable of replacing mental health professionals.
"This poses risks in clinical settings, as LLMs might respond inadequately to anxious users, leading to potentially hazardous outcomes."
The researchers said the elevated anxiety levels can be calmed down using mindfulness-based relaxation techniques. However, they cautioned that fine-tuning LLMs for mental health care may reduce biases but it will require substantial amounts of training data, computational resources, and human oversight.
"Therefore, the cost-effectiveness and feasibility of such fine-tuning must be weighed against the model's intended use and performance goals."
Also Read | Use Of AI Leads To Jump In Sex Doll Sales, Makers Say It Boosts 'User Experience'
AI loses cognitive abilities
Apart from experiencing emotions, the AI chatbots also show signs of deteriorating cognitive abilities with age akin to humans, a study published last month claimed. The researchers assessed the cognitive abilities of leading LLMs - ChatGPT versions 4 and 4o, Claude 3.5 "Sonnet" (developed by Anthropic), and Gemini versions 1 and 1.5 (developed by Alphabet), using the Montreal Cognitive Assessment (MoCA) test.
"All chatbots showed poor performance in visuospatial skills and executive tasks, such as the trail making task and the clock drawing test," the study showed.
The pattern of impairment shown by the AI tools resembled that of human patients with posterior cortical atrophy, a variant of Alzheimer's disease, the researchers added.
