No, I’m not losing track. You’re getting lost in logical definitions, ungrounded by any examples.
Talking about elements of a program is a way of focussing the distinction between narrow AI and AGI. A program can be said to be a set of instructions about 1) how to take *actions* upon 2) *objects* in a given field/environment. The elements of a program then are 1) the kinds of actions and 2) the kinds of objects involved – for example, the chess pieces and the moves they can make in a chess program. Introducing a new element means introducing a new kind of action or object. You are making it mystical because you can’t think robotically. If we program a robot – (and we won’t be able to do it algorithmically) – to build rock walls, then almost every time it reaches out for a new rock, it will be incorporating a new element – a new form of rock (because rocks are all individually formed) – by complete contrast with a Lego brick building narrow AI program, which will produce a set of variations on Lego building with the same form bricks, without introducing any new elements. A human can happily introduce new elements – new pieces - into the game of chess – and devise a new board game with a new set of rules. It won’t be that much of a stretch for an AGI robot to do similar (although this won’t happen any time in the near future). It’s similarly no big deal to create new dishes by throwing new ingredients in a trial and error way into a stew. The basic mechanics of it are more or less self-evident. Creativity/AGI is actually essentially an extremely mundane business – just reach out and add some new elements – new building blocks/new ingredients/new patchwork shapes/new creatures or characters into a given story genre/new punches or kicks into a physical sport/ new moves into a given form of dancing. Creativity is at heart simply a matter of “objets (et actions) trouves”. Of course this can be done mentally (“in your head”) as well as physically – but only once a physical foundation has been established. It’s somewhat mystical to you because you can’t think a) in terms of creative problem examples and b) robotically and c) non-algorithmically. Deal with actual AGI/creative problems and you won’t find it all so hard to understand – or think AGI has anything to do with complexity. From: Jim Bromer Sent: Tuesday, November 20, 2012 9:14 PM To: AGI Subject: Re: [agi] Randomness: Mathematics as Perceptual Bias On Tue, Nov 20, 2012 at 9:33 AM, Mike Tintner <[email protected]> wrote: Mike: You still don’t understand. Jim: I do understand. Mike: The problem of AGI is for a machine to be able to deal with *new* unpredictable elements – as humans can – Jim: That is right. Mike: the endless unpredictable things that can crop up in any field from a street to a text Jim: You mean the seemingly endless unpredicted things. Some of those things could be predicted. Mike: – AND to be able to proactively produce new elements Jim: This is where you are losing track. An AGI program is not going to "produce new elements." This is so important that I will have to get technical (so to speak) here even though I don't like doing that and I know that it doesn't usually work. What does the word "element" mean? 1. A fundamental, essential, or irreducible constituent of a composite entity. Can a new fundamental essential irreducible thing be invented by mind alone? Well it can but only through the element of mind (whatever that is.) So even though the mind can, for example, create a new abstraction, it must use the 'element' of mind that it possesses to do so. Now, perhaps the mind is actually regenerating and so it is creating new elements all of the time, but they are elements that are composed of the elements of the nervous system which had existed before. In other words, you are losing track (either intentionally or unintentionally) of the fact that the brain matter (whatever it is) is different than the things it thinks about. You must be talking about dealing with the patterns of things that are sensed because you then proclaim: Mike: There is only one way,. broadly, to actively incorporate new elements – and that is for a robot to reach out and grab new objects. Jim: Yes that is one way. But we are capable of creating new abstractions in order to gain new essential insights about the nature of the world around us. It is not necessary for an AGI program to be capable of reaching out and grabbing new objects, it is necessary for an AGI program to be able to create new abstractions about the Input Output data environment. An AI program has to be capable of interacting through Input and Output. That is one of the essential abstractions of AGI. The ability to use reason is another. Jim Bromer On Tue, Nov 20, 2012 at 9:33 AM, Mike Tintner <[email protected]> wrote: You still don’t understand. The problem of AGI is for a machine to be able to deal with *new* unpredictable elements – as humans can – the endless unpredictable things that can crop up in any field from a street to a text – AND to be able to proactively produce new elements – to independently explore the world, and move down new paths and into new fields, and seek out new objects. So you don’t need a program that can produce an endless variety of *patterns.* Patterns can’t produce new elements OR altogether new patterns other than themselves. A machine that could produce an endless diversification of patchworks would however be AGI- although that’s not at all a natural entrance point for AGI. P.S. There is only one way,. broadly, to actively incorporate new elements – and that is for a robot to reach out and grab new objects. And there is only one way broadly to find new ways of reacting to new objects/obstacles, and that is for a robot to reach out and grope for new lines of movement for its body., There is no way for a boxed in recipe/algorithm by itself to incorporate new elements. Since you don’t believe in robotics – basically because you’re lazy and unwilling to incorporate any new technologies in your thinking - you will never be able to deal with AGI, and will continue to think like other AGI-ers forever inside rather than outside the box/algo/formula/pattern. From: Jim Bromer Sent: Tuesday, November 20, 2012 1:26 PM To: AGI Subject: Re: [agi] Randomness: Mathematics as Perceptual Bias Well, I can't provide an example of an AGI program -that actually works - because it is still beyond the fringes of science. However, we have seen expected milestones go by: chess programs that can beat almost all human players, Watson who can beat all human Jeopardy players (so far), cars that drive themselves, toy helicopters that can learn to fly upside down and do other stunts while going through small windows and so on. The problem of finding "an example" of a generalization of all generalizations is a philosophical conundrum. (It cannot be done.) And finding examples that produce an endless variety of transformations is really not the problem we are interested in. We want to find a computer program that is capable of learning -like human beings-. Since this is well beyond our abilities we are looking for ways to get them to display some human-like capabilities of thought. But, while I don't have an example of an AGI program that actually works, I can think about things like a program that can draw a massive variety of types or a program that can output a massive variety of patterns. While the theoretical question was interesting to me (you need to use variable length data fields for the characteristics that you want to use) it is not AI because these "examples" would not be responding to Input Output. So I would not expect anyone to be very interested in an actual program to output a massive variety of visual patterns. I am not going to spend the time to make an example of something which no one would be interested in. Once you change the challenge so that it is becomes: write a program that will create an uncountable variety of patterns based on Input, you have taken the program nearer to AGI because now, part of the challenge is to get it to react to Input-Output in a rational way. The problem is not whether a computer program can potentially create (or model) more patterns then anyone could imagine count or keep track of, but whether or not such a potential could be harnessed to demonstrate reasoned intelligence. The challenge is interesting. For instance, could I write a program that could create an uncountable variety of patterns based only on images of garbage. Well, if the input images were constrained to images of garbage then yes I could. If the input was unconstrained then no I couldn't because it requires intelligence to identify images of garbage. So I am in an odd situation. I believe I could write a program that produced an uncountable variety of patterns but so what? No one would be interested. It is an interesting challenge because there are some theoretical problems that few people are even aware of, but I am pretty sure that I could do it. Maybe I should try it sometime just to make sure I did figure it out. Jim Bromer On Tue, Nov 20, 2012 at 4:44 AM, Mike Tintner <[email protected]> wrote: Jim: The idea is that we are arguing is that computers have a potential to 'represent' an astounding variety of things and relations. You seem quite incapable of distinguishing between “computers” and “programs” *Computers* currently have billions of apps and programs - potentially if not actually indeed representing all things. Programs – algorithmic programs – are just complex recipes with an extremely limited stock of a few ingredients, and zero new elements. You seem equally incapable of understanding the problem of AGI, which can be expressed as the incapacity of programs to incorporate any new elements. Patterns- algorithms – formulae – recipes – all only apply to a limited stock of elements. But in your wild imagination they are already all-knowing and all-encompassing and have already solved the problem of AGI. You are confusing the recipe with the chef - AGI is about creating a chef that can create any dish, and mix any new elements into dishes – – unlike algorithms and patterns which self-evidently have only a very limited set of elements – and even then run into problems of combinatorial complexity – the purely narrow AI problem that obsesses you. Examples, specifics, Jim, and not just waffle – of these all-growing, all-new-element-producing patterns, and these all-knowing, all-encompassing algorithms. You guys are really maintaining absolute nonsense – and you can’t produce a single example to back up your arguments. From: Jim Bromer Sent: Tuesday, November 20, 2012 8:28 AM To: AGI Subject: Re: [agi] Randomness: Mathematics as Perceptual Bias I guess Mike was saying that the variety of patterns that he provided a link to were, "unlike natural evolution, in which new elements, like eyes, limbs, opposable thumbs, voiceboxes, etc emerge..." But again, he is declaring that complex arrangements of cells are "elements" so even if I misunderstood how he was using the references to the images of patterns, I am still left with sense that he is not basing his argument on the idea that there are an infinite number of 'elements' but on the idea that an infinite number of complexes can emerge. No one is saying that a computer can truly represent an infinite variety of anything, no more than we are we saying that a single computer can know everything. No human being knows everything that human beings have ever known. The idea is that we are arguing is that computers have a potential to 'represent' an astounding variety of things and relations. Jim Bromer On Tue, Nov 20, 2012 at 3:13 AM, Jim Bromer <[email protected]> wrote: On Mon, Nov 19, 2012 at 8:31 PM, Mike Tintner <[email protected]> wrote: A specific pattern – Mandelbrot, Fourier transform, cellular automaton no. 30 etc.. specific. Examples are specific, Aaron, not general. And the “pattern of neuronal firings in your brain” is a fictional product of your imagination. Specifics. Evidence. e.g. https://www.google.com/search?num=10&hl=en&safe=off&site=imghp&tbm=isch&source=hp&biw=1340&bih=690&q=patterns&oq=patterns&gs_l=img.3..0l10.1679.2746.0.3037.8.7.0.1.1.0.159.516.5j2.7.0...0.0...1ac.1.hHfKyrSHq4M And here are “evolving patterns”, except that they don’t truly evolve with new elements – they aren’t truly “emergent” Mike's idea of a "patchwork" is just a fictional product of his imagination, if that is the attitude that he wants to take. And Mike's reference of the patterns in the images he provides the links to are truly formed from non changing elements of pixels of computer imagery. If his argument can be exemplified by the seemingly infinite variety of different patterns that can be found by searching on Google then that proves that the unchanging elements of pixlation can indeed represent the variety of different kinds of patterns that Mike is talking about. This seems like a contradiction of his basic argument since the question seems to hinge on whether computers can potentially -represent- a massive variety of patterns. Jim Bromer AGI | Archives | Modify Your Subscription AGI | Archives | Modify Your Subscription 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
