Recently, AI models have been in the spotlight for their peculiar behaviors, and a new study suggests they might even exhibit signs of cognitive decline. Research published in The BMJ indicates that some of the most sophisticated chatbots in use today are showing symptoms reminiscent of mild cognitive impairment, particularly as they age.

The primary goal of this study was to challenge the belief that AI technology has matured enough to be utilized effectively in the medical sector, especially for diagnostic roles. Researchers discovered that prominent chatbots, including GPT-4 and the Gemini 1.0 and 1.5 versions, faced difficulties with tasks that involve visuospatial and executive functions, such as drawing a clock face or recalling sequences of words. These findings raise significant doubts regarding the dependability of AI models in medical diagnostics.

Interestingly, when evaluated with the Montreal Cognitive Assessment, GPT-4 achieved the highest score among the chatbots, while the Gemini series lagged behind. Although these AI models performed well in areas like naming and language tasks, they collectively demonstrated a notable deficiency in empathy, which is a critical indicator of frontotemporal dementia.

The researchers emphasize the need to avoid anthropomorphizing AI systems and warn against setting human-like expectations for them. However, they also contend that if tech companies are marketing these AI models as conscious entities, they should be held responsible for their cognitive performance.

In summary, the study implies that AI models are not yet equipped to take over the roles of human physicians anytime soon. It raises concerns that healthcare providers may soon find themselves treating virtual patients—AI systems that exhibit signs of cognitive impairment. The results underscore the limitations of current AI technology and question the validity of the industry’s claims regarding the capabilities of these advanced chatbots.