I disagree with LeCun but you have a good point. Telling you how much uncertainty there is in a result is a useful metric. But it is only part of what is required (by my definition of explainability).
My counterpoint is when a model trained on dog breeds misidentifies a breed with a high certainty factor, the model should be able to justify (explain) why it made that judgement.
My counterpoint is when a model trained on dog breeds misidentifies a breed with a high certainty factor, the model should be able to justify (explain) why it made that judgement.