Everyone at my firm wants to jump in and create some shiny new AI tool to take to market, but rarely understand what that requires. So I frequently find myself pushing back against premature AI projects. I usually argue that we first need to build a simpler application that provides value out of the box in order to drive data into a single repository which we can then use to train algorithms on. I keep repeating the phrase "ML should be a version 2 feature."
I think this article identifies the root of the problem. No amount of marketing hype can replace a legitimate value proposition for users. And with rare exceptions, ML alone isn't going to provide that value.
The real takeaway is that "shiny! AI!" does not sell products. Meeting a need and not requiring users to change their behaviour sells products, AI or not.
Autonomous vehicles are bounded for the trough of disillusionment? I hope not :)
Shiny AI only works when it's augmenting something passively.