It will take at least until they take a wholly different approach to “AI”. Until they make something that has some concept of what it is saying, you’ll continue to get things much like you get today–a probability-based response that amounts to a series of symbols it thinks are a good reply to the series of symbols you entered. It has no way to validate itself nor even a concept of validation of output, so its validity will always be in question and the complexity of what it can do limited.