1 Comment

Nice assessment! I think the problem that you highlight is one that is really important for AI to tackle, and when it does, it’s gonna be a remarkably more powerful tool. Essentially, humans have a really good bullshit gauge. We can look at a numerator of 31 out of what is likely to be thousands of cases and know it’s bullshit. I have looked at large claim, status, hats, and known they were incorrect, because the population provinces of concurrent diagnoses I was seeing were impossible--there is no world in which 50% of patients with schizophrenia also have autism in hundreds of patients. So training AI on the kinds of things humans are using to determine whether with her thing is bullshit or not is gonna be crucial.

Expand full comment