One problem with the current rash of AI's in the news is they are only backward looking. They only "know" that which was in the training data up until they wer

One problem with the current rash of AI's in the news is they are only backward looking. They only "know" that which was in the training data up until they were trained. So to incorporate new things you have to retrain them, which could be an enormous expense to do continually. Also, if people start to rely on such AI's then the training data disappears and is replaced by AI output, so the whole thing is likely to get locked in a loop and never change and biases will also get locked in. So if you want to generate a dance video say, and the AI was trained before Michael Jackson was around you'll probably never be able to generate a Michael Jackson style video. However, there is also human input and guidance that is the source from which the AI generates output, so the quality and usefulness of the output will depend on how precisely the input can be specified. I agree with others (on other forums) who suggest that this very quickly turns into something like a programming language, where specificity becomes important for generating good content, which requires...human skill. Still, that seems useful, if not a replacement for people, an enhancer of people.