I think this blog post provides a very interesting and considered overview:
http://multiverseaccordingtoben.blogspot.hk/2014/03/lessons-from-deep-mind-vicarious.html Ben's the organiser of the Artificial General Intelligence conferences, and is a major voice in the community seeking AGI. Dileep gave a talk at the last one, but it alone seems to be missing from the archive on YouTube.. Incidentally, Jeff, Numenta, and HTM figure very prominently in many talks in Ben's circle of co-workers. One of his closest collaborators is basing his work on a system derived from early generations of HTM. There's a great interview in Ben's book about this: http://hplusmagazine.com/2014/03/20/new-from-humanity-press-between-ape-and-artilect-limited-time-free-ebook/ (free version of the book by following the PDF link) The book is a great and ranging read, highly recommended. Dileep says in the talk I have seen that his system does not use anything like NuPIC at the layer/region level, and that it mimics the neocortex in using hierarchy and in a few other chosen aspects. I'm sure he's incorporated many things he learned working at Numenta as well, but it seems he has decided that Jeff's path is not for him. Whatever the hardcore think of Vicarious and Dileep himself, I think any encouraging publicity is good publicity. I've noticed a completely different attitude among tech people recently when I tell them what I do. These mega-investments subtly change people attitudes towards the mission of using neuroscience to build useful systems. As someone once said, the difference between a lunatic and an eccentric is just his bank balance.. I personally don't see the point of what Vicarious is doing. Sure, if they discover some combination of heuristics which solves some set of problems, and they make some money out of it, good luck to them. But it's going to be just another "magic box", not a transparent scientific discovery. Contrast that approach with Stephen Wolfram. Sure, he wants to make some money to fund his many researches and to create a viable process for advancing his inventions. But he wants everyone to benefit, and his commercial success is only one metric of how well he spends his time and the worth of his decisions. Jeff is very explicit about commercial exploitation. His view is that the smell of money will attract the perfect combination of investors and talent, and will speed innovation in ways not possible by either closed commercial development or academic research. One way I see his role is as a provider and curator of the hard science of HTM theory, and it's our job to help convert that into technological reality, both open source and commercial. Regards, Fergal Byrne On Wed, Mar 26, 2014 at 6:27 PM, David Ragazzi <[email protected]>wrote: > Good observations, I only disagree in a one point: > > > However, we can assume, can we not, that Dileep most likely continued > with and expanded upon the work he was doing while part of Numenta? While > his approach was certainly more "East Pole" AI that Numenta's more > connectionist architecture, it too was based on biology (neocortex) as well > > I'm not sube about this... From I saw in Dileep's peer-reviewed papers > (some co-authored with Hawkins) the essence of the predictions was bayesian > inference using matemathical formulas to perform this. The recent > "connectionism" approach adopted by Numenta tries perform similar > function*, but it doesn't use bayesian formulas, on the contrary: it > BEHAVES in a bayesian manner, which is really NICE.. This to me is more > realistic than any based-rules inference.. > > David > > PS: I sincerelly don't know who first suggested this approach, i.e. the > own Dileep, Subutai, or other.. but I personally I consider this person (or > people) as true genius.. > > > > > > On 26 March 2014 15:03, Dean Horak <[email protected]> wrote: > >> Yes, the comments are interesting. >> >> The Yann LeCun/Dileep George exchange was back in October of last year. >> >> Then, last week Facebook did virtually the very thing that Yann chastised >> Dileep for. >> >> "Facebook announced last >> week<https://www.facebook.com/publications/546316888800776/> that >> it has developed a program called "DeepFace," which researchers say can >> determine whether two photographed faces are of the same person with 97.25 >> percent accuracy." >> >> Now (presumably after Yann took the time to actually investigate what >> Vicarious was doing) we see this big investment on the part of Zuckerberg >> (among others). >> >> Do you think Zuckerberg made this investment without first consulting his >> in-house AI guru (LeCun)? Doubtful. >> >> As far as a biological architecture, it's hard to say since Vicarious has >> been very closed lip about their technology (Recursive Cortical Network). >> However, we can assume, can we not, that Dileep most likely continued with >> and expanded upon the work he was doing while part of Numenta? While his >> approach was certainly more "East Pole" AI that Numenta's more >> connectionist architecture, it too was based on biology (neocortex) as well >> - though perhaps a bit farther removed than NuPIC. Then again, NuPIC is >> considerably biologically removed when compared to say, Spaun (Nengo), >> which is biologically removed from say the Blue Brain Project. How much >> biological realism is the right amount remains an open question. Hopefully, >> NuPIC is the correct balance, but that is still TBD. >> >> Dean >> >> >> On Wed, Mar 26, 2014 at 12:36 PM, Matthew Taylor <[email protected]>wrote: >> >>> (There is also an interesting discussion between Dileep George and Yan >>> in the comments!) >>> --------- >>> Matt Taylor >>> OS Community Flag-Bearer >>> Numenta >>> >>> >>> On Wed, Mar 26, 2014 at 10:34 AM, Matthew Taylor <[email protected]> >>> wrote: >>> > Here's what Facebook ML Director Yan LeCun had to say about Vicarious >>> > breaking CAPTCHAs recently: >>> > >>> > https://plus.google.com/+YannLeCunPhD/posts/Qwj9EEkUJXY >>> > --------- >>> > Matt Taylor >>> > OS Community Flag-Bearer >>> > Numenta >>> > >>> > >>> > On Wed, Mar 26, 2014 at 10:31 AM, Matthew Taylor <[email protected]> >>> wrote: >>> >> Well, we would know that if Vicarious open-sourced their algorithms. >>> ;-) >>> >> --------- >>> >> Matt Taylor >>> >> OS Community Flag-Bearer >>> >> Numenta >>> >> >>> >> >>> >> On Wed, Mar 26, 2014 at 10:15 AM, David Ragazzi < >>> [email protected]> wrote: >>> >>> Certainly a good news if we face Dileep George as a "son" or former >>> partner >>> >>> of Numenta, but a not so good news if we face Vicarious as a Numenta >>> >>> concurrent when the subject is research funds.. A question: Do >>> Zuckerberg >>> >>> knows about Numenta? I aren't trying take a position on this >>> question, but >>> >>> as far I know Vicarious uses a non biological architecture (i.e. a >>> non >>> >>> neural networks approach) to simulate neocortex which is not >>> insteresting to >>> >>> many of us and much less useful for neuroscientists interested in a >>> real >>> >>> biological simulation, not a simulation of the neocortex functional >>> >>> properties.. Well, maybe I'm wrong... >>> >>> >>> >>> >>> >>> On 26 March 2014 13:20, Jeff Fohl <[email protected]> wrote: >>> >>>> >>> >>>> I thought people on this list might be interested in this news >>> tidbit, >>> >>>> about Vicarious picking up a major round of investment: >>> >>>> >>> >>>> >>> >>>> >>> http://blogs.wsj.com/digits/2014/03/21/zuckerberg-musk-invest-in-artificial-intelligence-company-vicarious/ >>> >>>> >>> >>>> - Jeff >>> >>>> >>> >>>> _______________________________________________ >>> >>>> nupic mailing list >>> >>>> [email protected] >>> >>>> http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org >>> >>>> >>> >>> >>> >>> >>> >>> >>> >>> -- >>> >>> David Ragazzi >>> >>> OS Community Commiter >>> >>> Numenta.org >>> >>> -- >>> >>> "I think James Connonly, the Irish revolutionary, is right when he >>> says that >>> >>> the only prophets are those who make their future. So we're not >>> >>> anticipating, we're working for it." >>> >>> >>> >>> _______________________________________________ >>> >>> nupic mailing list >>> >>> [email protected] >>> >>> http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org >>> >>> >>> >>> _______________________________________________ >>> nupic mailing list >>> [email protected] >>> http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org >>> >> >> >> _______________________________________________ >> nupic mailing list >> [email protected] >> http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org >> >> > > > -- > David Ragazzi > OS Community Commiter > Numenta.org > -- > "I think James Connonly, the Irish revolutionary, is right when he says that > the only prophets are those who make their future. So we're not > anticipating, we're working for it." > > _______________________________________________ > nupic mailing list > [email protected] > http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org > > -- Fergal Byrne, Brenter IT <http://www.examsupport.ie>http://inbits.com - Better Living through Thoughtful Technology http://ie.linkedin.com/in/fergbyrne/ https://github.com/fergalbyrne e:[email protected] t:+353 83 4214179 Formerly of Adnet [email protected] http://www.adnet.ie
_______________________________________________ nupic mailing list [email protected] http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
