"AI" can't even perform as well as humans (despite plenty of promises) in a field like radiology. The idea of an AI family doc system or ER doc system actually making diagnoses (instead of being a glorified productivity tool*) is downright hilarious. Lots and lots of luck interpreting barely coherent, contradictory and often misleading inputs from patients, dealing with lost records or typos, etc.
Doctors don't get paid the big bucks for rule-based solutions based on rote memorization. They get paid the big bucks to understand when it's inappropriate to rely on them.
* which IS a worthy goal to aspire to and actually helpful
> "AI" can't even perform as well as humans (despite plenty of promises) in a field like radiology. The idea of an AI family doc system or ER doc system actually making diagnoses (instead of being a glorified productivity tool*) is downright hilarious. Lots and lots of luck interpreting barely coherent, contradictory and often misleading inputs from patients, dealing with lost records or typos, etc.
I think the future of that might be with wearables like the Apple Watch. While it probably won't replace doctors wholesale, applying ML to the data gathered from various sensors continously seems like a much better promise to me.
> They get paid the big bucks to understand when it's inappropriate to rely on them.
An automated system could record and analyze more outcome and biometric data than a group of doctors, over time obtaining more experience about when to apply or not the various medical rules. Human experience doesn't scale like a dataset or a model.
I bet some diagnostics could be correctly predicted by a model that a human can't understand, especially if they require manipulating more information than a human can hold at once.
Doctors don't get paid the big bucks for rule-based solutions based on rote memorization. They get paid the big bucks to understand when it's inappropriate to rely on them.
* which IS a worthy goal to aspire to and actually helpful