Last night I was at a networking/social/office party. My {locate engineer; move to engineer; socialize with engineer} loop was preempted by a manager from a small globally-distributed network of contractors. Somewhere in the midst of conversation, I let it slip that my PhD focus was AI/ML. He asked me a question from my "Questions I've been asked for which I have not a satisfactory answer but should probably investigate eventually," list, and my inarticulate response prompted me to sit down and hammer out this post. The question was,

When are we going to have real AI?

I've talked about the matter of 'real AI' in an earlier post. See AI is not Dead. I'm going to skip the pedantry of the question and dive into what I think is the spirit. "When are we going to have a synthetic intelligence system which is at least as cognizant as the least intelligent 'intelligent' human?"Predictions are always hard, especially about the future. Computer science predictions in particular have much less precision than their mathematical underpinnings much otherwise suggest. There were a few predictions from Cornell which stated AGI (artificial general intelligence) would emerge in the 1980's, then the 1990's. Rather than the zero-effort output of "now+40 years, lol!" I'll try and enumerate for the lay person (with tidbits for the journeyman) what we think is necessary for AGI.What do we have?Recent trends in AI mark a shift from the old way of manually building feature detectors and hierarchies to automatically learning features and