Geoff Hinton Dismissed the Need for Explainable AI
forbes.com
forbes.com
Imo, this whole explainability thing is an incorrect human intuition that we're imposing on AI. If you want unbiased AI, explainability isn't the way to do it. Instead judge its performance against explicit criteria for what counts as bias. If those criteria don't exist or can't be agreed on then the problem is hopeless from the outset.
mathematical proofs. My algorithm solved xyz conjecture. it's true. how? I don't know it's a black box.
Also, from a diagnosis/debgugging problems it can be a problem. Oh, no the was a huge mistake and the system just catastrophically failed. how do we fix it? don't know. Should be keep using it? don't know.
Not saying there aren't places where a black box is perfectly fine.