Ben,
BEN> If an AGI machine must be very simple, then why is the brain so bloody complex? SERGIO> Because of all it has learned. Acquiring information is not the same as learning. Information comes with entropy, learning is the process that removes excess entropy, causing the information to self-organize, develop meaning, and become understandable. Which 100M dots of light on the retina are not. There is no need to explain brain anatomy if the purpose is to understand brain function. Ants do things that machines can't, I would be content if I could understand just that, for now. There are those who believe that the anatomy of the brain resulted not only from frozen accidents but also from adaptation to its function. Instead of insisting with brain anatomy, it would be better to consider recent advances in Neuroscience that take into account the now well-know distributed nature of brain function over its frozen anatomy. Joaquin Fuster describes the brain as a distributed cortical/cognitive network where hierarchical structures composed of cognits represent knowledge, with the cognits being distributed over the entire cortex and not concentrated in modules. The 7-point list of his ideas he presents on page ix is nearly identical to my list of mathematical properties of causal sets. One can be converted into the other by simply replacing a few words. He didn't know anything about me, and I didn't know anything about him until somebody posted his name on the blog. BEN> A neuroscience text is 1000 pages, and each chapter is just a coarse overview of a certain neural mechanism, region or network... SERGIO> Just one more proof that self-organization can not be explained much less coded at the complexity level. It is better to try to understand the principles of self-organization, which turn out to be very simple, and then apply them and try to generate the complexity. The machine I propose can do only one thing: learn, meaning acquiring information, storing it in memory, and self-organizing it. To use it, you give it information and let it learn. Which looks a lot like a child at school. The idea that we should learn ourselves all we can, including learn about learning, then stuff all that into a machine, and then the machine will somehow become capable of learning, is preposterous. The reasonable approach would be to give the machine the ability to learn in the first place, and then let IT do the learning. The problem with that, is that nobody knew what the "ability to learn" is, so they just assumed that nothing was there. The history of AI shows that researchers who took matters seriously and confronted that reality, were led to desperation and decided to adopt desperate measures, such as simulating the cortex, reverse-engineering the cortex, simulating cognition, intelligent design, or a quantum mechanical cortex. But will the simulation reveal how learning works? No, it will only reveal that what we put into it. BEN> When we last discussed neuroscience (years ago, on this list), you made the claim that the brain's neuronal network is best modeled as a dag rather than a graph replete with cycles... which seems quite inaccurate to me... SERGIO> Just one year ago, almost to the date. Maybe I said brain, but I was (or should have been) thinking about causal systems. Everything in our world is causal, at least this side of the black holes. Being causal does not mean no loops, it means that they halt, so they can be unrolled. For the purpose of mathematical analysis, it is better to consider the loops unrolled, as in a dag or causet. Why are there so many loops in the brain? I don't know, but why couldn't they too be a result from evolution/self-organization? Why couldn't they have appeared to reuse resources? How do loops appear in a computer program? They appear when the developer notices that some code is repeated and decides to reuse it. But mathematical analysis with loops would be too difficult, so it is better to consider them unrolled, as they are in dags and causets. Loops are put in AFTER causal analysis is completed. Ben, I have been working during this year. I have now cleaned up the theory, meaning that it contains only the fundamental principles of Physics and the functional. Nothing else. It does not care about anything of the brain, for all that matters the brain could not exist and the theory would still stand. The brain doesn't either, care about the theory, that's what evolution does. But if we want to understand the brain then we need the theory first, the brain second. Ben, did you see my report on causal set that you requested? Sergio -----Original Message----- From: Ben Goertzel [mailto:[email protected]] Sent: Monday, September 10, 2012 4:31 PM To: AGI Subject: Re: [agi] Discovering physical dimensions Sergio, > No, an AGI machine must be very simple, capable only of learning without accumulating entropy, not one that expects us to force-feed all that I, me, myself into it. > > Sergio > If an AGI machine must be very simple, then why is the brain so bloody complex? A neuroscience text is 1000 pages, and each chapter is just a coarse overview of a certain neural mechanism, region or network... When we last discussed neuroscience (years ago, on this list), you made the claim that the brain's neuronal network is best modeled as a dag rather than a graph replete with cycles... which seems quite inaccurate to me... -- Ben G ------------------------------------------- AGI Archives: <https://www.listbox.com/member/archive/303/=now> https://www.listbox.com/member/archive/303/=now RSS Feed: <https://www.listbox.com/member/archive/rss/303/18883996-f0d58d57> https://www.listbox.com/member/archive/rss/303/18883996-f0d58d57 Modify Your Subscription: <https://www.listbox.com/member/?&; ad2> https://www.listbox.com/member/?&; d2 Powered by Listbox: <http://www.listbox.com> http://www.listbox.com ------------------------------------------- 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
