4 ms·
Using AI for sugery would terrify me. Some algorithm that does it right 95% of the time and in 5% of the time would do something completely random and inexplica
by cm2187 8y ago
Using AI for sugery would terrify me. Some algorithm that does it right 95% of the time and in 5% of the time would do something completely random and inexplicable...
- Farradfahren 8y agoThe AI concluded, that nutrition absorbtion would be optimized if your colon was restructured into..
- tossaccount123 8y agoAnything in the medical field would have to be extensively tested, this isn't a web app where "move fast and break things" applies and it would primarily be used with routine surgeries and would default to human intervention under uncertainty. Most of these surgeries are basically identical and the AI could improve by sharing data with all surgeries performed around the world. Not to mention how bad humans are at surgery, especially when fatigued. AI is the only way to fix healthcare, a doctor/surgeon can work effectively only a few hours a day and takes nearly 30 years to educate from childhood. An AI could do surgery and diagnose patients nearly 24/7, 7 days a week.
- cm2187 8y agoI agree that medecine is one of the only domains that didn't industrialize, remaining an artisanal profession, with a highly skilled, highly paid professional in front of every customer. But I am more concerned about the failure modes of AI.
- dexterdog 8y agoSounds about the same or maybe better than what we have now.
- c0nfused 8y agoDon't forget that in that 5% every time you, the robot company, are going to get sued by either the patient or the hospital operating your product.
- vletal 8y agoThat is not how ai works. Such algorithm would not make it to an actual surgery. Even humans make mistakes. The advantage The point is to get the model to perform better than humans.
- throwawayjava 8y ago> the model Surgical robots -- at least the ones that actually exist and work today -- are designed more like traditional safety-critical robotics platforms and less like a one-off case study for a nice visualization/table for an end-to-end deep RL paper. There is no "the model".
- rmrfrmrf 8y agoWhen it learns through reinforcement that the best treatment for tachycardia is smashing the patient's skull in.