anyone who knew how markov bots work
There's nothing wrong with that technology, per se. It's the training corpus that dictates sucess/failure. Was the plan to blindly crawl the cesspool that is the Internet for that raw data a bad business decision? Most decidedly so.
It turns out that the most expensive part of AI is training. Both in terms of resources (power and equipment) but also the labor involved with validating the inputs. I did some fiddling around with semantic nets a few decades ago. But I was restricted to a library of engineering documents (at Boeing) that had been vetted to some extent by groups of experts (for a rather loose definition of that term) before release. The results worked well, but had no hope of advancing to the status of AGI. And of course, the tools built had no hope of reaching outside markets due to the inclusion of company proprietary knowledge.
This last point raises another problem wiith today's AI. Why should I allow my expertise, which I depend upon for a competitive edge, to fall into the hands of my competition? The AI bots won't find much useful stuff on my public facing web site. And I'm not stupid enough to host on cloud services. Owned by outfits with interests in AI services.