All AI Models are Wrong, but Some are Optimal