KI übersieht laut Studie fast ein Drittel aller Brustkrebserkrankungen

Now. It will only get better

Now let’s see human performance on the same data.

Two reasons the title is misleading: First, the article itself states that after changing the method they used AI with, they were able to correctly reevaluate about 80% of the missed cases. Second, it doesn’t compare that sample with human radiologists. It’s not clear if by using those same images, actual doctors would have done better or not.

How many does a human expert miss?

I mean, I love to shit on AI as much as the next guy, but maaaaaany years ago, when I was developing natural-language software, we hit a certain accuracy threshold (let’s say 95%) that no software was able to get past. I suggested that three of our PhD linguists do the same task. They got about 97%, which meant that the software got 98% of what a human could do.

i just wanna point out that scans interpretation itself is a skill. There’s clear cut cases but there’s also borderline cases. Even if AI helps with “initial”interpretation while the finalised report gets process, It would greatly reduce workload.

I wouldn’t expect much from AI, it’s relatively new. But I would really love to know the human perfomance on the same scale