It's more subtle than that - the AI combo catches more because the AI is inhuman and catches abnormalities a person wouldn't, but also misses stuff a person wouldn't. In the study even expert humans got lazy and their skills atrophied after working with the AI.
You end up needing to structure both training and application in such a way to minimize deskilling effects in order to get high-quality and sustained results.
maybe i’m wrong or missed it but i don’t think the study found evidence that human +ai skill declined over time with use? though that’s certainly going to be an important thing to monitor
It was the headline result. After working with the AI for 3 months, doctors missed over 20% of the stuff they had caught previously. This isn't surprising or necessarily catastrophic, but it is something that needs to inform how we'd train doctors and use the tech.
right this is a decline if the ai is taken away from them, not a degradation in skills using the ai
a decline in skill without machine learning tools available would only be meaningful if those tools became unavailable. it’s a risk, but i don’t think a big one relative to the benefits of using it
fuck AI
yeah I'm pretty sure the study showed that after using AI, docs were worse at identifying problems *without* AI than they were previously before they started using AI. its kinda like saying you are worse at doing mental math after using a calculator for a while than you were prior to ever using one