Where AI has consistently failed is creating a real thinking machine that can reason and adapt. That's not what we're building today though.

The problem is to many peoples jobs are to easy and highly specialized. They don't require strong AI - just good enough "dumb" algorithms, and we're getting really good at those. Doing paperwork by strict specifications. Driving a truck from point A to B. Assemble a circuitboard from a blueprint. As long as you keep a handful of humans in the loop incase something unexpected happens, this is all stuff robots can do today.

Unfortunately I don't know any AI researchers who have done work in the past 5 decades. Would you care to share what would happen?

In all honesty I have no idea. Would they laugh and say that "no way AI needs another 20 years"? Or would they say it's only a few years until everyone working with their hands is out of work? Kudos on the pretentious reply btw.

Reading about the AI Winter[1] will provide some useful background.

However, there is some evidence that it really is different this time. Big data makes a big difference to things like image recognition.

[1] http://en.wikipedia.org/wiki/AI_winter

I imagine they will say "I've heard that same prediction every week for all of those 5 decades."