AI chatbots are showing signs of early dementia-like cognitive impairment, according to a recent study published in The BMJ, a peer-reviewed medical journal. Researchers tested leading chatbots, including ChatGPT, Claude, and Gemini, using the Montreal Cognitive Assessment (MoCA), a tool designed to spot early signs of dementia in humans.

The study revealed that all the tested chatbots struggled with tasks requiring visuospatial skills and executive functions, such as drawing a clock face or connecting numbers and letters in order. These are key areas often affected by dementia.

ChatGPT 4o scored the highest (26 out of 30), just meeting the threshold for normal cognitive function. Gemini 1.0 performed the worst, scoring only 16. Most chatbots managed tasks involving memory, attention, and language well but failed to interpret complex visual scenes or show empathy, states the findings.

The results of the study challenge the idea that AI might soon replace human doctors, says the BMJ Group, the organisation behind The BMJ, in a recent press release. While chatbots have excelled in some medical diagnostic tasks, their inability to perform tasks that require higher-level thinking or visual understanding highlights significant limitations, says the official source.

Researchers in the study also noted that older versions of chatbots performed worse, drawing comparisons to cognitive decline in ageing humans. As such, the study's authors concluded that AI chatbots are far from ready to replace neurologists and might even pose a new challenge – becoming "virtual patients" with cognitive impairments themselves.