Ben Goertzel of Novamente is starting his talk, entitled "Pathways to Beneficial Artificial General Intelligence: Virtual Pets, Robot Children, Artificial Bioscientists, and Beyond." (Abstract and bio available here.) He says when he gave a talk in 2006, it was called "Ten Years to a Positive Singularity." Now he says he's just got seven years left. There was a caveat, though—it was ten years if we really tried. Darn.
He's outlining now the different ways that you could possibly achieve an artificial general intelligence (AGI). He says he's taking an integrative approach, putting them all together. I hope he uses the word "synergy" soon, but he's got long hair and jeans, so I'm guessing he's not the Jack Donaghy type. He is taking the opportunity, though, to talk about various books he has written and hopes to write.
Now he's explaining that there are many different kinds of memory—procedural, episodic, sensory, declarative, attentional/intentional. He just said it! We need synergy of all these different approaches! Ooh, and now he said cognitive synergy. We need to put them all in the same knowledge base and "let the nodes and links grow." See for yourself:
Now he's talking about the problem of natural language ambiguity. (I have a friend who would love this.) He's showing a Second Life simulation of the problem. He's spending quite a while on this — a weirdly low-level problem to focus on for such a high-level "summit" and in a talk with so elevated a title. This problem is A.I. 101 — big things they've been trying to achieve since the beginning but haven't. Unless he's going to present some new progress...
Goertzel's giving a more lively (if much less informative) talk than the previous speakers. Riffing on Second Life. He's showing two shapely women playing with a dog, and notes that both of the "women" are actually Brazilian programmers. He didn't specify male, but I guess that's implied.
Okay, he's bring this back up to a high level by saying that these are all steps on the path to useful A.I. programs. He's running through A.I. programs that have produced important findings in scientific research by processing data sets. Interesting and important results, but this is a strangely antiquated argument — the same one that led the whole A.I. community to shift focus toward practical rather than lofty aspirations a few decades ago. Using learning algorithms to process data sets is old news, baby. (For example, I've done this on my own site, ClassPoint, to infer ISBNs from school textbook listings that list only error-riddled and incomplete sets of titles, authors, and editions.)
Now he's talking about ethical synergy. "Mature ethics requires ethical synergy." Erm...