I'd look at what will be, rather than what is. I think that it's probably not controversial to say that AI is going to improve; these are early days. The question is to what extent.

If one is to assume that AI will improve very little over time, that ten years from now the kind of responses that you'll get generated by a computer ten years hence in response to a question will be about the same as they are today, then, yeah, it's probably an error to commit major resources to AI stuff or to expend resources acquiring training data for it.

But that assumption may not hold.