Re:Chief scientist of major corporation can’t handle critici...
by s0rce on Wednesday November 23, 2022 @11:13AM
(#33721042)
I've definitely seen radiologists miss things (as a patient and a researcher) but I've also seen the behavior where if something is labeled or indicated the next person (or in this case the first person) might not give it a true 2nd look and could easily default to agreeing with the AI instead of actually checking things.
Overall I think this would be a net benefit as image analysis can help with a lot of these activities, just needs care to not inadvertently remove the humans you still need.
[ Parent
| on HN ]
Re:Chief scientist of major corporation can’t handle critici...
by bluGill on Wednesday November 23, 2022 @11:50AM
(#33721588)
Right. Until we trust the AI 100% (at which point we get rid of humans) the AI input shouldn't be given until after the human makes a call. However then the AI gets above 80% (or something like this, UX has better numbers) trusted humans know the AI will look will just skip looking themselves.
The above is a common problem that UX research is interested in. I'm not sure how much it is solved, but it goes well beyond medical fields.
Re:Chief scientist of major corporation can’t handle critici...
by mistrial9 on Wednesday November 23, 2022 @01:08PM
(#33722523)
sadly, this will never appear completely in practice, I predict. Refer to information theory roughly, to support that. That is, pure signal is not often occurring in real systems. Some systems will favor signal quality at the expense of time and fewer false positives and negatives, but the majority of cases in the real world will favor less-effort, less-time decisions that are expedient, or worse, favor obfuscation of the process to cover the sins of the humans benefiting/profiting directly from the show. Also, systems that just are not actually working well, will be sold for quick profit using pressure marketing and forced contracts, at least I think so...