The only times I've heard him mentioned the impression was negative and that he didn't understand any of the actual science.
People hear "machine learning" and they think it is about machines that know how to think. Machine learning is actually just optimization of high dimensional functions. If this language were used it wouldn't sound as sexy, but no one would think machines are going to take over the world.
AI isn't magic. It's really just clever search techniques and mathematical optimization.
There are still many, many things that we don't understand about the brain. Even the things we think we understand, we're not always 100% sure of. Recreating an actual intelligence will be difficult.
What's your point? Nobody said it's magic. The fact that it isn't magic (and that its tremendous complexity far surpasses our current ability to understand it) supports the notion that it won't suddenly spring into existence. If we placed some primordial sludge in a petrie dish overnight, we wouldn't worry that a sentient creature will have materialized. And if we program a computer to optimize numerical functions, there is just as little evidence (perhaps less), to suggest that the computer will somehow gain sentience.
There's a very fine line between AI futurist and best-guess scifi writer. Most "AI thought leaders" are scifi writers, not technical researchers. They take preconditions, generate a story, think how it could happen given plausible technology, then market that as soon-to-be-fact.
It's an entertaining society and endlessly fun to read, but still complete fiction based on internal brain states of individuals and not necessarily based on real world interactions.
Also see: Eliezer Yudkowsky — great writer, fun to read, but largely scifi thought experiments masquerading as research.
