Which UI practice best helps users assess an AI system's confidence in its output?
Always present full model internals
Show confidence scores with ranges and visual cues
Replace confidence with vague hedging language
Hide confidence to avoid bias