Jim,
I deserved that. I have been disrespectful. May I try again? For your convenience, I repeat my ABCD exercise: (A) "The buttler is the assasin. He was at the scene, and he has a motive." That's a very coarse granularity, but it is already a causal set. Still, we can already take action. We can arrest the buttler and find more facts. (B) "The crime took place at 7:30." (C) "The buttler left the scene at 7:29." And we can draw a conclusion: (D) "The buttler was not at the scene at the time of the crime. You seemed to imply that one needed additional information, such as what is "butler", before drawing conclusions. I said conclusion, but I was actually thinking of the structural hierarchy of the causal set (sometimes I am at a loss for words, just like you with "problem"). Assume for now that ABCD can be formalized as a causal set. I know this needs to be discussed some more, but let's do it some other time. Then, the sentence between C and D would be: "And the top of the new hierarchy is:" Is it possible that your view of causal sets is static? There exists a direct map from a causal set and its corresponding hierarchy. The map is so strong that saying causal set is like saying hierarchy. The map is independent of the elements of the causal set (whether they have any meaning or any structure at all). But, when new information is acquired (this is an AGI I am thinking about, it learns or senses all the time, and it can also go after information), then the causal set changes and the hierarchy changes too. All of it, including that what I previously called "conclusion." And the granularity also changes. It is going to be more fine-grained than it was before. I am saying this to emphasize that causal sets support granularity. You can start with a very coarse granularity and refine it as you go. Which looks a lot like we ourselves do. Then, when you say "we need to know what is 'butler' ", that's alright, but you are only asking for more information. When that information is obtained, and if it is obtained at all, there is going to be a new causal set and a new hierarchy. The so called "conclusion" will change too. If you go to my Pissanetzky(2009a) <http://www.scicontrols.com/ReferencesForThisWebsite.htm#Pissanetzky2009a> and scroll down to page 27, Section V.F "Supervised Learning in the Case Study", you will find an expanded example of the dynamics of learning and progressive granularity refinement as it happens in EI. Three years ago I used a different terminology. Please substitute "causal set" for MMC and canonical matrix. Charging me with "having found the engine but ignoring the rest of the car" is fun on the blog, but it distracts attention. I just said above: "The map is independent of the elements of the causal set." Therefore I MUST ignore the elements, do the EI, and only then be reminded that the elements may have structure and begin asking questions. Only then I will be able to refine my granularity. Not "I", of course, I mean the AGI. Sergio From: Jim Bromer [mailto:[email protected]] Sent: Saturday, September 01, 2012 5:47 PM To: AGI Subject: Re: [agi] Granularity Sergio, Well I tried. Just to reply to this message. I don't understand where facts come from you said. You don't really think that is a likely and effective explanation for my scepticism do you? What kind of remark is that? You honestly thought that I don't understand the relationships of a logical statement? That kind of conclusion on your part does not make any sense. "The sticky part here, is how to draw the conclusions. This, is not rational. It can not be explained, or understood, or reasoned about. It is not expressible. But it is a fact, it happens." Ok, that is, if we are to believe you, your point of view. Again, it does not really hold water. You just finished explaining how it was rational and explainable, understood and reasoned about. I am really interested in seeing what happens to this free-energy thing but I am not waiting for the problem to solve itself or for it to be the solution that we agiers just don't get. I talk about problems because I need a word to use to denote and bind what I am talking about. I could have used reasoning or something like that instead. It is just a working term that I use sometimes because a better word doesn't pop into my head when I am writing stuff like this. Yes toddlers do not think (or at least do not always think) in terms of problems that have to be solved, but I am talking about the problem of creating a computer that can think, not whether or not toddlers consider their activities from an self aware plain of problem solving. (It's a different kind of thingy.) I wish getting you guys to acknowledge some humility about this was easier. On the other hand, I have tried numerous times and that is that. Jim Bromer On Sat, Sep 1, 2012 at 3:15 PM, Sergio Pissanetzky <[email protected]> wrote: Jim, I can understand your exasperation, but you are exasperated because you are wrong. You missed so many things. You missed the point that conclusions are drawn from what we know, not from what we don't. So, if I know B and C, why wouldn't you let me draw my conclusion? Any reasonable person would, you would too. I could have said a snake did it, or lightning killed him at 7:30. It doesn't matter what the butler is. Only the causal relationships matter. And if you were to let me draw the conclusion D, then I would have already a new fact. D is a new fact about the world. That's why you are asking so many questions, because you dont' understand where facts come from. But there is another, much more fundamental point that you missed in full. If I start asking questions, where do I stop? 10 questions? 1000? I can't go to the big bang. So, where do I stop? The stop point is arbitrary. So you don't ask anything, you first draw whatever conclusions you can from what you know. Then you know what questions to ask, ask them, and draw more conclusions again. Your view is static. You think of putting "everything" in a big heap before you even consider it. In my language, that would be accumulating uncertainty, or entropy. The view I describe is completely dynamic. Your senses feed information to your brain, your brain constantly draws conclusions from whatever it has at the time, and this is an ongoing process. To be more precise, it is two processes happening at the same time, one learns and stores (guest), the other draws conclusions (the host). I did explain host/guest in my posts. The sticky part here, is how to draw the conclusions. This, is not rational. It can not be explained, or understood, or reasoned about. It is not expressible. But it is a fact, it happens. It is observed! This is why I always insist, to the dismay of some, that the action functional was observed, not invented. Much of Physics is like that. It is made of hypothesis, every natural science is falsifiable. JIM> the logic of a real world problem (or even a theatrical hack like this) becomes inexpressibly complicated. SERGIO> Sure, for you. Because you accumulate uncertainty instead of getting rid of it by drawing conclusions from what you know, and only then learn some more. And you are doing this because you fear the "draw conclusions" part, so you just leave it for later. I bet that if anyone knew the draw conclusions part, then they would immediately realize that sequence draw conclusions/learn works, while learn/draw conclusions doesn't. JIM> a lot of us were aware that in trying to find the answer to a problem one often has to collect more information. SERGIO> You are influenced by the notion of *problem*. You are "solving a problem", so you require "all" the information before you even start considering. But "all" you know only after you try to solve the problem, unless somebody solved it before and told you. Brains don't work with problems. The notion of problem itself is a conclusion from a considerable amount of previous experience. You learn "problem" in elementary school. Toddlers don't know about problems, yet they do AGI much better than any machine. This fixation on problems is causing havoc in AGI. Yes, I did too, I exagerated a little to make my points. And yes, I knew this would be difficult to grasp. Because it is so very much different from what AGI'ers believe. Sergio From: Jim Bromer [mailto:[email protected]] Sent: Saturday, September 01, 2012 11:33 AM To: AGI Subject: Re: [agi] Granularity But you forgot to explain how an AGI program would know: What is a butler? What is an assassin? How are those two related? Why would anyone assume that a butler has something to do with an assassin? What is a the? What is a is? What is a the again? What is a he? (And knowing what a he is how does it become bound the the butler?) What is a... (Do I actually have to spell it all out for you?). The logic of the problem that you gave is EXTREMELY simple and yes a lot of us were aware that in trying to find the answer to a problem one often has to collect more information. But the problem that currently confronts the development of AGI is that when you actually try to assign ALL the details to your story into a fine grained explanation of how a person is able to figure it out, it suddenly becomes MUCH much more complicated. When you add to that the recognition that there is a potential for lot of sources of ambiguity in the derivation of a conclusion for a real world problem and that an AGI program is going to be going on automatic without people spoon feeding it with explicit directions, the logic of a real world problem (or even a theatrical hack like this) becomes inexpressibly complicated. I certainly don't want to belabor what most of know is obvious, but I have noticed that you aren't a stickler for detail either. The skepticism that I have towards your claim (implicit or explicit) that you have it figured out is not that the theory is completely lame but that you seem to have no idea what it is that I have been trying to say. Do you get it now? Of course not, because, in your mind, you don't have to: You already have it all figured out. Sorry if I seem unpleasant, but this is my honest opinion. I am not exaggerating to make my point seem stronger than it is. Jim Bromer On Sat, Sep 1, 2012 at 10:56 AM, Sergio Pissanetzky <[email protected]> wrote: AGI, I don't recall having explained granularity on this blog before. My neglect has caused misunderstandings. So here it is. When we describe our world, we arbitrarily select the *granularity* of our description. (A) "The buttler is the assasin. He was at the scene, and he has a motive." That's a very coarse granularity, but it is already a causal set. Still, we can already take action. We can arrest the buttler and find more facts. (B) "The crime took place at 7:30." (C) "The buttler left the scene at 7:29." And we can draw a conclusion: (D) "The buttler was not at the scene at the time of the crime. We have now a better description, a finer granularity. But what does it mean to "draw a conclusion?" It means to *bind* or *associate* B and C to generate D. This is a logical step, and requires a mathematical logic operating on the causal set. The logic removes uncertainty from the causal set and creates structure. The set {B, C} is uncertain. We know B and C, but we ask "so what?" To reach D we have to remove the uncertainty. Removing uncertainty is easy for our brains. We don't even notice it happened because we are not aware of it happening. It is unconscious, and our brains do it all the time. You the reader can continue the excercise. Every bit of information we acquire is followed by a "remove uncertainty" step. Our brains are constantly removing uncertainty. AGI will never work unless we learn how to "remove uncertainty" on a computer. No adaptive system will work unless we learn how to "remove uncertainty" on a computer. Have courage, don't leave the uncertainty there. I must have tired everyone by saying entropy all the time. But entropy is just a measure of uncertainty. If someone is 6 feet tall, I must use "feet" to say that. Sergio AGI | <https://www.listbox.com/member/archive/303/=now> Archives <https://www.listbox.com/member/archive/rss/303/10561250-164650b2> | Modify Your Subscription <https://www.listbox.com/member/archive/rss/303/10561250-164650b2> <https://www.listbox.com/member/archive/rss/303/10561250-164650b2> AGI | Archives | Modify Your Subscription <https://www.listbox.com/member/archive/rss/303/10561250-164650b2> <https://www.listbox.com/member/archive/rss/303/10561250-164650b2> <https://www.listbox.com/member/archive/rss/303/10561250-164650b2> AGI | Archives | Modify Your Subscription <https://www.listbox.com/member/archive/rss/303/10561250-164650b2> <https://www.listbox.com/member/archive/rss/303/10561250-164650b2> <https://www.listbox.com/member/archive/rss/303/10561250-164650b2> AGI | Archives | Modify Your Subscription <https://www.listbox.com/member/archive/rss/303/10561250-164650b2> <https://www.listbox.com/member/archive/rss/303/10561250-164650b2> ------------------------------------------- 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
