Steve,
Your path 2 may not work. In my previous posting I explained why. It's because one would need to know where to look. Without some understanding of the big picture, that's difficult to do. Note that I have actually followed path 2. I simulated refactoring. Why? because I knew, from some understanding of the big picture, that refactoring would be a good place to look. I also explained what I mean by "big picture." Take flying, for example. Without a principle, and without knowing where to look, we would still be trying to produce the best possible bird wings in our laboratories. In this case, the big picture were Newton equations, and the place to look the shape of wings. STEVE> Mathematicians note apparent needs for various functionality, that the wet lab people then look for, and the numerical analysis people evaluate. Note that this relationship is already alive and well in producing present AGI approaches. SERGIO> That's what I need. I noted some needed functionality, and I want wet lab people to start looking into. Also computer people to start looking into. I don't know where these relationships are in place, but I seem to be looking in the wrong place. STEVE> How to you start in trying to figure out how completely unknown systems work, when you don't understand the problems that they are solving and you have no access into their operation? SERGIO> Maybe, just maybe someone has already done this. Sergio From: Steve Richfield [mailto:[email protected]] Sent: Friday, June 22, 2012 5:44 PM To: AGI Subject: Re: [agi] Prediction Did Not Work (except in narrow ai.) Sergio, You have eloquently stated a MISunderstanding that I often comment on. I will once again attempt to pull the spaghetti apart... On Fri, Jun 22, 2012 at 12:32 PM, Sergio Pissanetzky <[email protected]> wrote: I have also insisted that the brain is the only known intelligent system, and that we have a lot to learn from it. But I've also said, the "lot" does NOT include the implementation of the brain. There are TWO separate paths. Neither one can be successfully followed without some exploration of the other: 1. Understanding what is happening, so we can simulate it. 2. Simulating what we see happening, so that we can understand it. So, regardless of whether your ultimate goal is understanding so you can write better code, or synapse-by-synapse simulation, most of the work needed is identical. All we need is to understand the principles, Of course, the BIG word in the above statement is "All". I personally believe that some machinery could be fairly directly constructed that would answer the vast majority of outstanding questions. For some unknown reason (Ben, here is your big chance to explain), there seems to be nearly universal resistance to even speculating on how a few million dollars could answer most of the remaining questions needed to build an AGI. I think the answer is that AGI people can't bear to even consider that their path to success might be blocked by a present lack of knowledge. Are there OTHER opinions here as to the mental defect that underlies this resistance? and then we can start using the principles in new and creative ways, such as an artificial system, without ever having to simulate the brain in all its complexity. Here we agree. We almost certainly need to simulate SOMETHING, but a fruit fly's brain might be enough. A mouse brain would be more than enough. Sorry if you don't like this, but this is a cornerstone for me. For the sake of principles, I don't need to know very much about the 200 types of neurons. I suspect that there are dozens of forms on unknown mathematical systems at work here. Otherwise, why would these many different types of neurons ever evolve? Sure, people might figure these systems of mathematics out on their own during the next 1,000 years or so, but I would like to see things move faster than that. No, computers do NOT have to compute things in the same way that we compute them, but computers DO need to somehow compute the same things by SOME method. These systems of mathematics are now unknown, so there is no end of the spaghetti hanging off of the plate from which to start. How to you start in trying to figure out how completely unknown systems work, when you don't understand the problems that they are solving and you have no access into their operation? Note that many neuroscientists know about AGI, whole-brain simulation, etc., but I have yet to meet one that thinks that there is any reason to expect these to succeed in coming centuries, short of some sort of Manhattan Project like effort to start figuring out how we actually work. This is clearly not possible from external observation - we must look inside, and do so with equipment having MUCH greater capabilities than present equipment. That doesn't mean I disdain that vast knowledge. Quite on the contrary. All I am saying is that, for now, it is not needed. ... and you say that based on what? Once the principle is set, then, not now, then it will be the right time to start examining neurological knowledge. By you, not me. I'll help all I can. People work in teams in Science, you know. I remember a 3-way discussion that included William Calvin and Cathryn Graubard who has done a LOT of neurological research - probably more than William. I explained that a particular inhibitory synapse that Cathryn had characterized had EXACTLY the right non-linearity needed to be processing the logarithms of probabilities. Cathryn objected, as her team had already found another entirely different equation that also matched her the observations to within the precision of her measurements, so why the heck even consider my equation? William stepped in and tried to explain that the equation that Cathryn had found didn't represent any useful computation, while my equation was exactly what was needed to perform computations that were suspected of being performed (in this case, to control the manufacture of lobster poop). Cathryn then commented that no one was even looking at computational efficacy. This particular experience drove home, at least to me, that there are three entirely different skill sets, populated by three entirely different types of people, and ALL THREE are needed to make headway: 1. Wet-lab biological skills 2. Mathematical skills 3. Numerical analysis skills (including some of the present AGI efforts) Mathematicians note apparent needs for various functionality, that the wet lab people then look for, and the numerical analysis people evaluate. Note that this relationship is already alive and well in producing present AGI approaches. Wet lab people note interesting structures and functionality, that the mathematicians look for some use for. There has been a GREAT impediment here, because the wet lab community has discouraged publishing things that people cannot PROVE are happening, so lots of useful speculation is simply lost - unless you hang around the right people. Numerical analysis people keep everyone honest - can their proposed functionality actually work? Ultimately, their work product will be the AGI that everyone has been waiting for. Steve AGI | <https://www.listbox.com/member/archive/303/=now> Archives <https://www.listbox.com/member/archive/rss/303/18883996-f0d58d57> | <https://www.listbox.com/member/?&; ad2> Modify Your Subscription <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
