3 ms·
We've been down the AI diagnosis path before and even a purpose-built system failed to live up. I think that GPT, much like Tesla autopilot, is going to spend a
by tootie 4y ago
We've been down the AI diagnosis path before and even a purpose-built system failed to live up. I think that GPT, much like Tesla autopilot, is going to spend a very long time ironing out it's "last mile" before anyone will trust it. The tendency for AI to be spectacularly wrong when it fails to grasp context is still pretty prevalent.
- markus_zhang 4y agoJust curious is it a political issue (risky and who to sue) or a technical issue (ai has far worse judgement)?
- tootie 4y agoMostly that it faired poorly against human doctors. Real-world medical data is too messy to feed into a machine. https://www.theregister.com/2022/01/31/machine_learning_the_hard_way/ https://www.theregister.com/2022/01/31/machine_learning_the_...
- markus_zhang 4y agoThanks! I wonder what is the most advanced automatic technique right now. I also think it's difficult to replace nurses -- even though basic nurse work is even easier comparing to a family doctor who needs to diagnose. I figure the reason is that to replace nurse we need a lot of automatic hardware, which needs a lot of investment, unlike ChatGPT which can really replace a service agent given enough training because it doesn't involve hardware.
- tootie 4y agoWe have a lot. But it's really the same story that we see with software. The more sophisticated the software you grab off the shelf the more that's expected from a single worker. Modern doctor's offices are overloaded with gadgets. Most are using automatic blood pressure cuffs, digital pulse ox and all sorts of machines to test for things that used to be untestable. The need for nurses and techs to operate them all only increases.