BTW, Wisner-Gross will be giving one of the keynotes at AGI-14 in Quebec City in early August... I encourage y'all to come argue with him in person !!!
I don't think he's found the holy grail of AGI, but I do think his observations are interesting... I think causal path entropy (or something like it) would sensibly be included as one of the high-level goals of an AGI system... ben On Sat, Feb 22, 2014 at 4:20 AM, Bill Hibbard <[email protected]> wrote: > Yes, the paper at: > http://www.alexwg.org/publications/PhysRevLett_110-168702.pdf > is more detailed and quite interesting. > > An interesting project would be to investigate > the relation between this paper and AIXI. The > paper includes probabilities of future histories, > for a system interacting with an environment, in > a new definition of entropy, called causal path > entropy. > > Probabilities of future histories for a system > interacting with an environment play a major role > in the definition of intelligence in AIXI. It > would be interesting to see how close the relation > is between causal path entropy and AIXI. > > Bill > > > On Fri, 21 Feb 2014, Matt Mahoney wrote: > >>>> From: Tim Tyler [mailto:[email protected]] >>>> >>>> "Alex Wissner-Gross: A new equation for intelligence" >>>> >>>> - https://www.youtube.com/watch?v=ue2ZEmTJ_Xo >> >> >> On Mon, Feb 10, 2014 at 6:46 PM, Piaget Modeler >> <[email protected]> wrote: >>> >>> >>> I found it too vague. >> >> >> I did too, and the Entropica website wasn't any help. It just has the >> same video clip you saw on TED. However, I did find a more detailed >> explanation at >> http://www.alexwg.org/publications/PhysRevLett_110-168702.pdf >> >> Unfortunately, if you were looking for the holy grail of AI, you can >> keep looking. It doesn't shortcut the uncomputability of intelligence >> proven by Hutter's AIXI model. In the entropic model, the idea is that >> the optimal action of an intelligent agent is the one that maximizes >> future entropy. Of course entropy in the information theoretic sense >> is not computable because it depends on Kolmogorov complexity. >> >> However it might still be a useful principle, in the same way that >> Occam's Razor is useful to machine learning. We do know that >> computation requires energy. In particular, writing a bit of memory >> decreases the information theoretic entropy of a computer's state by >> up to 1 bit, and therefore requires a corresponding increase in >> entropy of the environment of kT ln 2 where T is the temperature and k >> is Boltzmann's constant. So it looks to me like the principle is to >> choose the action that maximizes expected future computation. >> >> -- >> -- Matt Mahoney, [email protected] >> >> >> ------------------------------------------- >> AGI >> Archives: https://www.listbox.com/member/archive/303/=now >> RSS Feed: https://www.listbox.com/member/archive/rss/303/3603840-9a430058 >> >> Modify Your Subscription: https://www.listbox.com/member/?&; >> Powered by Listbox: http://www.listbox.com >> > > > ------------------------------------------- > AGI > Archives: https://www.listbox.com/member/archive/303/=now > RSS Feed: https://www.listbox.com/member/archive/rss/303/212726-deec6279 > Modify Your Subscription: > https://www.listbox.com/member/?&; > Powered by Listbox: http://www.listbox.com -- Ben Goertzel, PhD http://goertzel.org "In an insane world, the sane man must appear to be insane". -- Capt. James T. Kirk ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
