For the largest health insurer in the US, AI’s error rate is like a feature, not a bug.
It’s a very advanced algorithm.
Step 1: would we cover this? No: all good. Yes: consult AI.
Step 2: is the AI recommending we cover this? No: all good. Yes: reconfigure AI and go to step 2.
“error” suggests it’s not the intended result.
I thought the same thing, like “damn who uses a model with a 10% accuracy on its training data?”
Unfortunately UnitedHealth is big enough that they will pay less than they made.
It’s not a bug, it’s a feature 👍
This feels like a recurrence of the dumbassery of yesterdecade with the excitement over algorithms. AI simply is not ready for these sorts of applications. That it was put in charge of anything with any degree of gravity is already a massive failure.
