EFTA02575966.pdf

From: Joscha Bach Sent: Wednesday, October 23, 2013 3:10 PM To: Sebastian Seung Cc: Joi Ito; takashi ikegami; Ari Gesher; Kevin Slavin; Martin Nowak; Greg Borenstein; Jeffrey Epstein Subject: Re: MDF Hi, Sebastian, what an honor to meet you! > I call this theory the "hierarchical perceptron," and credit it to > Fukushima's Neocognitron (1980). His work in turn was inspired by > discoveries and speculations of neuroscientists Hubel and Wiesel > (1962). Today's deep learning architectures for vision continue in > =his tradition. By the way: Kurzweil's most recent book mostly covers his rediscovery =f the Neocognitron, too. I believe that it is a crucial part of the story of the mind (with =espect to hierarchical self-organization), but not the whole story, of =ourse. > In the 1990s, my research was focused on machine learning, which can > =e seen as a rebranding of the pattern recognition camp of Al. As far > as = can tell, AGI is an attempt to revive and rebrand the reasoning > camp =f Al. I'm sympathetic to this goal. If I were starting over in > AI today, I'd study reasoning rather than join the pattern recognition > bandwagon. I perceive AGI (including BICA, cognitive systems etc.) as a quite =eterogenous bunch of approaches, which are only united by their =ommitment to the original goal of AI, i.e. building/understanding minds =nstead of building applications. Most of the people in the field would =cknowledge the need to combine associative, parallel, distributed =rocessing with some localist processes that facilitate language, =nalytical deduction, planning and so on. We need to to all of it, in a =ommon framework. > That being said, AGI will have trouble succeeding because it is > following the scruffy tradition. Perhaps the main failing of this > tradition is its refusal to define objective (and preferably > quantitative) measures of success. The question of good benchmark tasks is haunting Al since its inception. =sually, when we identify a task that requires intelligence in humans =playing chess or soccer or Jeopardy, driving a car etc.) we end up with = kind of very smart (chess-playing, car-driving) toaster. That being =aid, Al was always very fruitful in the sense that it arguably was the =ost productive and useful field of computer science, even if it fell =hort of its lofty goals. Without a commitment to understanding =ntelligence and mind itself, Al as a discipline may be doomed, because =t will lose the cohesion and direction of a common goal. > Minsky is bitter because he failed to turn his field into a science. The field is there alright, still feeding truckloads of tenured =rofessors, huge conferences and so on. I think that Minsky's =rustration is much more fueled by the disregard of the contemporaries =or the "right" ideas, and the deterioration into EFTA_R1_01747742 EFTA02575966 methodologism, =ntellectual cowardice and the focus on short term goals (in some sense: =recisely what it means to become a science). > I'm reminded of the saying that "A science is any discipline in which > the fool of this generation can go beyond the =oint reached by the > genius of the last generation." You can stand on the shoulders of giants, or on a very big pile of =warfs: it works either way. But we need to pick the right questions. (I =m currently quite enamored with a developmental approach, i.e. working =long child performance with respect to story and scene comprehension.) Cheers, Joscha <?xml version=.0" encoding=TF-8"?> <!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.O.dtd"> <plist version=.0"> <dict> <key>conversation-id</key> <integer>270440</integer> <key>date-last-viewed</key> <integer>0</integer> <key>date-received</key> <integer>1382541011</integer> <key>flags</key> <integer>862375014S</integer> <key>gmail-label-ids</key> <array> <integer>6</integer> <integer>2</integer> </array> <key>remote-id</key> <string>364477</string> </dict> </plist> 2 EFTA_R1_01747743 EFTA02575967