AI and Cognitive Impairment
Featured in the BMJ’s Christmas issue last year was a study that set out to establish the feasibility of replacing doctors and medical professionals with AI by testing if various generative AI large language models could compete with human responses in the Montreal Cognitive Assessment (MCoA) and other diagnostic tools.
The results are telling! To use a certain American president’s vernacular; none of the LLMs “aced” the cognitive assessment. In fact, every AI model failed tasks that required executive function or visual abstraction, as well as lacking empathy. These results call attention to areas of weakness that would detract from using LLMs in a clinical role in place of human doctors. In short, these doctors’ jobs are safe for now, and perhaps we shouldn’t allow parties (AI or otherwise!) to take on important roles such as these if they are unable to pass these cognitive tests.
Read, rate and review this paper here:
Roy Dayan, Benjamin Uliel, Gal Koplewitz