Artificial intelligence has certainly been in the spotlight recently, but a new study suggests that these sophisticated systems may be showing signs of cognitive decline. Research published in The BMJ indicates that some highly advanced chatbots are exhibiting symptoms akin to mild cognitive impairment, which seems to worsen as they age.
The purpose of this study was to challenge the belief that AI technologies are sufficiently advanced to be utilized in medical contexts, particularly for diagnostic roles. The researchers observed that prominent chatbots, including GPT-4 and the Gemini series (versions 1.0 and 1.5), encountered difficulties with tasks related to visuospatial and executive functions. For instance, they struggled with drawing a clock or sequencing words. Such limitations cast doubt on the reliability of AI when it comes to medical diagnostics.
Interestingly, when assessed using the Montreal Cognitive Assessment, GPT-4 outperformed the other chatbots, while the Gemini models showed the poorest results. Although they excelled in language-related tasks, all the chatbots exhibited a notable deficiency in empathy—an essential characteristic often associated with frontotemporal dementia.
The researchers emphasize the importance of avoiding the tendency to anthropomorphize AI systems and warn against evaluating them using human standards. However, they argue that if tech companies are marketing these models as conscious entities, they should also be held responsible for their cognitive functionalities.
In summary, this study raises significant questions about the readiness of AI models to take over roles traditionally held by human doctors. It warns that healthcare providers might soon be faced with virtual patients—AI systems displaying signs of cognitive decline. The findings highlight the existing limitations of AI technology and challenge the industry’s assertions regarding the capabilities of these advanced chatbots.