I would say that we cannot hope for AI to be better than the training data that it is fed. I've seen it be much worse.

While I am not a data scientist, I've spent many years working for companies that employ them. Training data quality, the algorithms and heuristics used, and particularly the attributes of the training data (usually selected with a great deal of experimentation over time) are all critical to any kind of artificial intelligence.

Also, machine learning and "deep learning" are closer, in my opinion, to marketing terms than useful distinctions.