Princeton's Zeynep Tufekci argues medical AI diagnostics must be evaluated at population scale rather than through anecdotal cases

@zeynep it's part of a study, they seem to be doing everything docs would want? It feels like part of the response is that this patient anecdote was told ahead of a study readout

@zeynep Very good point - though what's worse than the n=1 generalizations about how AI is definitely beneficial is the common a priori n=0 certitude that it's definitely going to be either useful or harmful.

@zeynep Even when the conclusions are based on multiple patients, I always struggle to understand why the success rate standing alone is considered a good metric. https://medium.com/@Blougram/lies-damn-lies-and-statistics-on-the-ai-success-rate-fallacy-44e731ebc23d

@nikillinit @zeynep A case report is not the same as a study.