If 95% is good enough for you, machine learning will probably get you there rather easily. With many of the really valuable use-cases, it's just not good enoug

If 95% is good enough for you, machine learning will probably get you there rather easily. With many of the really valuable use-cases, it's just not good enough. If 100% of the time you need an expert to tell if a sample falls within the 95% of successes or 5% failures, you're not adding any value. Even if you're bulk processing stuff that would've otherwise been ignored, somebody will have to deal with those signals. The net effect is more work, not less. In other words, would-be radiologists ought to stay in school.