Indeed.
I think the problem with current marketing is that you need a problem that it is applicable to at the point of deciding to use it which is a very narrow subset of reality. Most problems can be solved with little or no intelligence. Actually even better than that, when you apply intelligence to something, be it human, machine learning or otherwise, it actually compromises the result significantly because it removes all determinism. You can't reason always as to why something happened and that's not something people can cope with.
I actually watched a commissions matching company burn through 200,000 financial commissions records with their "AI" system which got about a 50% hit rate and 70% certainty which they said was remarkable. For AI it was. But their competitor had a 95% hit rate with 90% certainty with no intelligence at all. No AI meat there; just string matching in SQL and a stemming algorithm.
The worst thing about the "AI" based system was that the rate of failure changed, sometimes better, sometimes worse as it was reacting to real data entered by dumbasses. On a bad day it created more work for humans to unpick than it did not existing.
Microsoft like to sell it as a use case for everything and project that. Same thing they did with their WS-* bollocks for distributed architecture (WCF, WPF, WWF etc).
It's not a bag of new tricks, it's a monkey with a gun if you put it in the wrong room.