I did look up functionals before posting, I dont know how I managed to take some many uni courses and read so many books without any mention of them! But there is very little mystique about them, it is function+field, which is why I asked for complexity metrics-does anyone need any reminders of how tough the n-body problem is in a field? Also, an implementation with 50000 neurons or 50 or whatever seems to me to define a different functional de facto, although I have very little investment in this kind of opinion. It is more interesting for me that your "solution" suffers both from uniqueness and infinities, butI will leave the technicalities to the more CAS inclined. Without the incorporation of heuristics it looks like the causal singularity could be a gazillion years away. Of course I have been wrong before. Once. Thanks for the summary.
AT Sent from my iPad On Aug 31, 2012, at 9:22 PM, "Sergio Pissanetzky" <[email protected]> wrote: > ANASTASIOS> A little ill-defined, any dimensional or complexity bounds for > that "function" ... > SERGIO> Short answer, assuming you know what a complex adaptive system (CAS) > is, and some Math and Physics. It's a functional, not a function. The problem > space is a causal set representing a CAS. The functional represents the > entropy (more precisely, the action) existing in the information acquired by > the CAS (for example, from sensors or cameras). The information comes with > entropy. The entropy processor removes the excess entropy, causing the CAS to > self-organize, that is, converge to its attractor. And the attractor is an > algorithm. For example, it can drive the actuators in a robot. Mostly all is > published. The functional is very simple, given in Section 3.5 of my 2011 > Complexity paper, Complexity 17(2): 19–38 (note I said partially ordered sets > in the paper, but they are actually causal sets). You can also check my > website www.SciControls.com. Everything is very precisely defined. > > > ANASTASIOS> ... and its problem space? > SERGIO> The given causal set, representing all the information available to > the CAS at any particular time. The causal set grows, from learning, so > self-organization is an ongoing process. > > > ANASTASIOS> The entropy processor as in ...(example of software or hardware > completed or under construction?) > SERGIO> I implemented the EP on my PC years ago, and solved some examples > with it. They are listed or referenced in my Complexity paper. My student > also built one as part of his Master Thesis. But it's too slow, and I need > special hardware. I have plans to build an FPGA for a USB port, hopefully > next spring. Perhaps some 50,000 neurons, but I don't know for sure. > > > ANASTASIOS> What would your "exact same AGI machine" be comparing? > SERGIO> I meant the AGI machine will pass all benchmarks together, rather > than re-programming the machine specifically for each benchnmark. > > > ANASTASIOS> One entropy processor against another? > SERGIO> No, "the" EP directly against the human brain. A human is given a > problem statement (one of the GUAPs), and asked to solve the problem, without > telling her why. The solution is documented, and the human is dismissed. Then > the machine does the same, and the two solutions are compared. They must be > nearly identical. There are some problem/solution pairs published, but > additional ones can be generated if needed. Google has some patents for > entropy processors, but those are for video only, not useful for AGI. > > > ANASTASIOS> One functional against the other? > SERGIO> The entropy is one thing, so the functional is unique, there can't be > any other. Of course one can "invent" another, but it would have no physical > meaning and would not do what the brain does. The brain is a physical system > so the functional must have a physical meaning or it will not do what the > brain does. > > > Sergio > > -----Original Message----- > From: Anastasios Tsiolakidis [mailto:[email protected]] > Sent: Friday, August 31, 2012 10:04 AM > To: AGI > Subject: Re: [agi] Your essential AGI modules and breakthroughs, again? > > On Fri, Aug 31, 2012 at 3:49 PM, Sergio Pissanetzky <[email protected]> > wrote: > > The breakthrough is the action functional that powers emergent inference, > > the non-negotiable module is the entropy processor. > > A little ill-defined, any dimensional or complexity bounds for that > "function" and its problem space? The entropy processor as in ... > example of software or hardware completed or under construction? > > > Benchmarking would relatively easy. It would consist of solving > > various problems of the kind "easy for man, difficult for machine" > > (the GUAPs), under the conditions that the exact same AGI machine is > > used for all problems > > Right, our "problem" is that we are looking for benchmarks and test suites > that allow for partial solutions, so that we know we are making progress, > rather than pit AGIs against each other to see which one is more optimal or > megabrainish. What would your "exact same AGI machine" > be comparing? One entropy processor against another? One functional against > the other? > > > > > Why do you ask? > > I was running a little short of ideas for my own design :) > > AT > > > ------------------------------------------- > AGI > Archives: https://www.listbox.com/member/archive/303/=now > RSS Feed: https://www.listbox.com/member/archive/rss/303/18883996-f0d58d57 > Modify Your Subscription: https://www.listbox.com/member/?&; > Powered by Listbox: http://www.listbox.com > > AGI | Archives | Modify Your Subscription ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-c97d2393 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-2484a968 Powered by Listbox: http://www.listbox.com
