I see a lot of references to machine learning as the key thing to turn on the "smartness" in processes.
Having worked with AI (for predicting financial time series) way long before machine learning became mainstream cool via the Siri joke, I can honestly say that that moment is very, very, very far away.
See, machine learning is great for some domains (especially pattern recognition - most popular being voice, sound, video, images, handwriting etc) but incredibly poor for other domains (facing the unknown, grasping new and infrequent concepts, etc).
So, unless someone has an idea about how to train a machine to come up with a completely original decision when confronted with a major business crisis... don't hold your breath until our business processes become automagically autonomous.
As an example, during our financial time series forecasting business, our simulations showed that any probable gains realized by AI (neural networks, genetic algorithms) during normal "periods" were more than wiped away by a Black Swan event.
I apologize if this came a bit too strong, but there is a Murphy's law that goes: "If you hold a hammer, everything around you will look like a nail". Same goes for AI - I don't think this will ultimately work for business processes.
The only thing smart about a Smart Process is the way it has been designed (including ideas like plug-in flows and ad-hoc tasks and external app triggers and whatnot), and that's inherently a human task.