The findings challenge the assumption that artificial intelligence will soon replace human doctors, as the cognitive impairment evident in leading chatbots may affect their reliability in medical diagnostics.
When subjected to the Montreal Cognitive Assessment (MoCA), GPT-4o scored the highest at 26 out of 30, barely meeting the threshold of what's considered normal, while the Gemini family scored the lowest at 16.
While all tested chatbots excelled in naming, attention, language, and abstraction tasks, they struggled significantly with visuospatial and executive tasks like drawing a clock or remembering a simple five-word sequence.
The study illustrates that while AI models may perform well on many tasks, their cognitive limitations akin to mild dementia raise concerns about their suitability for medical applications.
Collection
[
|
...
]