|
Post by tkorrovi on Jul 27, 2006 21:43:31 GMT -5
Yes there are constant i/o nodes. In a more developed system, there would develop a kind of structures or processes which protect the i/o node from deleting, so in practice they stay and would not be deleted. But for the beginning of development, there is provided a mechanism for substituting i/o nodes, if they really happen to be deleted.
|
|
|
Post by Abram Orion Demski on Jul 28, 2006 10:43:55 GMT -5
AOh, OK, good.
That means that I now have full understanding of the basic processes. So the question becomes: What is the logical behavior of such a system? I suppose it depends on starting state.
|
|
|
Post by tkorrovi on Jul 28, 2006 11:23:46 GMT -5
If we talk about logical behaviour, it's a kind of logic which consists of two parts, a kind of association and natural selection, of connections. Just anything theoretically supposed to be possible to implement that way. But the starting structure is mostly very simple, just to connect all i/o nodes. Sure if we would create a more complicated one, we might get the behaviour we want, but it is very difficult to manually create such structures. Just that the system is extremely dynamic, is no joke, every structure would mostly start to develop very quickly, and it is difficult to control what happens. There should be ways to create more static structures, this "trigger" is an example of these, but because of the extreme dynamism of the system it's difficult to figure them out. This is why no one really tried to create such, and the only thing done has been studying the system which starts to develop by itself from the most simple structure, and train it with the training program. Such systems are not really studied, almost not at all, so we don't even know all which might be possible to do. One thing, some systems for network analysis may be used, maybe like JUNG sourceforge.net/projects/jung which is though based on Java, or others in sf category Artificial Intelligence. Network analysis is a very universal method, and may be useful for so many things. But all that is not very easy, and so no one really do that.
|
|
|
Post by Abram Orion Demski on Jul 30, 2006 12:57:07 GMT -5
Cool. Very cool.
Darwinism is obviously applicable, so here's some of the logical behavior based on that:
--system will grow if points tend to create new points that tend to create new points-- in other words, if the property of having common neighbors with neighbors is inherited (iff inheritance is possible in the system).
--System with shrink if points tend to delete points due to not enough common neighbors (obviously)
--System will grow and then shrink if nodes are created (due to common neighbors) that have no common neigbors (and therefore delete things).
So we can classify structures by their behavior:
1. deletion nodes
2. Creation of type 1 nodes
3. Creation of type 2 nodes
4. Creation of type 3 nodes
...
If all nodes can be classified in this way, the structure will always grow for a time then shrink. (Therefore, we can say that not all nodes can be classified in this way; the situation is more complicated.)
|
|
|
Post by tkorrovi on Jul 30, 2006 14:31:21 GMT -5
Yes this is not a random mechanism, it is derived, like to be the only one such, you may look at the derivation (in the project's site adsproject.sourceforge.net/ads-ac) and find out how much it is so. One may consider the nodes only consisting of connections to be like grouping, or sets. All the arithmetic and logic supposed to be based on set theory, and that mechanism supposed to be able to do anything with sets. This mechanism anyway has different meanings, like it is interconnected, and changing at that. Also there theoretically has to be a reason, why such system survives hundrieds of times longer than structures in cellular automaton (the only other system which supposed to be unrestricted), which anyone can easily reconfirm experimentally.
|
|
|
Post by Abram Demski on Jul 31, 2006 19:13:45 GMT -5
Yes, I've read the derivation-- that's part of what is so strange. The system isn't derived to do anything in particular, but to be unpredictable; yet it seems to work.
If nodes did represent sets, things would be different. The question for me is still this: how does the system store learned information to use later?
|
|
|
Post by tkorrovi on Aug 1, 2006 2:13:39 GMT -5
The nodes would not immediately disappear, they supposed to remain while they are necessary. That way it certainly stores things, at least for a short time, and can refresh that knowledge. Just that I don't know what to explain, if it ever may interest, there is an old chatbot I made years ago, extremely simple www.geocities.com/tkorrovi/rb.html If you succeed to run it, worked only in internet explorer, and didn't find that thing it so important, so that to find out why it doesn't work in mozilla. Anyway, it combines the words which are *different* for two sentences, and that way it is able to solve simple syllogisms, because exactly that is necessary to solve these syllogism. The most simple ones "tomatoes are red fruites", "red fruites are salty", which gives us "tomatoes are salty" just because "red fruites" is common and would be omitted, and "tomatoes" and "salty" are different, and therefore would be combined. This is very primitive of course, and we can no way ever make it advanced enough if we work at such level, sentences and words, but still it some way shows some logic which the ADS-AC mechanism has. And sometimes, after many sentences, it makes some very surprising sentences, we may get them certainly just randomly, but not so very probably. Though that chatbot is certainly not more than a very stupid toy, I just tried how short I can make the javascript code, finally I could make it 60 lines :-) But of course, with such chatbots, we cannot do anything very advanced or unrestricted, that's a dead end, because we further have to code so many rules of logic, the coding never ends, new rules would always be found, and it would still be not good enough, and still kind of too dull. Such situation means always that the system has not enough generality, in the other words, is not unrestricted enough. Just that always it takes a long time and a lot of coding, to find out again that we cannot code by far as much as would be necessary for a good enough system, in the other words that the task is infinite. At least so it has happened, so many times, probably thousands of AI projects, started, developed some time, and abandoned then. The worst, no advances in AI by most of them, so just unnecessary work, even a smallest real advance in the theory would be thousands of times better result.
|
|
|
Post by tkorrovi on Aug 26, 2006 12:02:39 GMT -5
You say ADS is derived to be random. But ADS is not derived to be random, it is derived to be absolutely dynamic. This means that within a bigger system, any system can emerge, and disappear. This first means that any possibility can be generated, and that the multiple drafts principle works, so that only these subsystems remain, which fit into their environment. Third, based on the derivation there seems to be only one kind of system, which is dynamic enough to satisfy requirements. The system has not exactly that logic which I thought should be in such system, it is just strictly a system which the derivation gives. So I don't know exactly why such logic supposed to be the best, though I can say some things which are positive in it, and what it means, just as it is the only possible kind of systems which satisfies the requirements, then there must be reasons why it has to be the most correct logic. You may look at how it is derived, and find out whether everything has to be exactly so, though I thought it through so many times, that I have no serious reasons to doubt, also no ideas what could be different. Would be so nice if there would be anyone, who would work further with it, whatever way, because there is no other way, I have brought these things to the point where they are, at least there is all the methodology how to do such research, and in addition that there are no fast computers, I don't feel that I'm able to go further, it was big enough effort to bring it to certain point, where others can go further. You, Orion, seem to be the brightest who I have seen, at least understand some things which others don't. Though I of course cannot be sure that you would ever do that, and I don't ask either, you may either create some other thing, or not be exactly a proper person for whom such things really fit, think what you want and do that. Just I don't even know what benefit doing all that really gives, I sometimes think whether I should regret wasting so much time for such things, but then I feel that I don't regret, for a single reason that I proved to myself, that I'm able to do such thing, do something unique which no one before was able to do, and this gives so much, I know who I am, foundedly. In the world of fools, it is sometimes difficult to find yourself, have some ground to decide whether you should believe what they say, or not. But this is about me, personally. So as a conclusion, ADS was derived to satisfy certain strict requirements, to have certain properties, therefore at least it is not certain, that it should act randomly. The systems which are created to act randomly, are cellular automatas, and they are also not derived to satisfy any requirements, more than that they should be able to work. Though even such random systems, can implement Turing machine, though a very clumsy ones, and there is no reason to consider, that such Turing machine can even theoreticallt emerge by itself, during the development of the system.
|
|
|
Post by tkorrovi on Aug 27, 2006 18:39:43 GMT -5
I mean, any system can emerge in a bigger system, as a result of a development. If it was just about generating random systems, it was random. But it is a system derived to be capable to develop, to anything, development is already by itself something more organised, going smoothly from one thing to another, not just a sequence of random states. And this is about deriving a system, which can develop to anything, so supposedly capable of the most advanced development. This is almost all which I now can answer directly to the question, what was it derived to.
|
|
|
Post by NickyBlue on Mar 18, 2015 8:48:45 GMT -5
Hey Trav is their anything left in the world which you haven't skimmed through at least? I think you really a great asset to buddy with only if you don't get week in knees when times comes for it btw wht is Automatic Control engineer preciesly?
|
|
|
Post by tkorrovi on Mar 18, 2015 8:59:31 GMT -5
Well, that's just everything about controlling technical systems, with regulators, microcrocontrollers, computers.
|
|
|
Post by NickyBlue on Mar 18, 2015 9:14:31 GMT -5
Wow! that's great! So you online, eh? smile. btw I just spammed your general board completely now buddy or sort of... haha! think I enjoyed it! done it for 1st time in life. feels good Hope you don't mind its kinda already dead in some way, isn't it? And hey if you online can't you create a member account for me here. My browser giving problem on regitration form u c. Just make it with NickyBlue999@gmail.com password: would you?
|
|
|
Post by NickyBlue on Mar 18, 2015 9:20:33 GMT -5
Hey Trav gtg buddy.. its party time!
pls do create my account and reply to this post if done. I will get check it when I came bck in night. Okay?
luv u, cya!
|
|
|
Post by tkorrovi_ on Mar 18, 2015 9:51:04 GMT -5
I did, an email was sent to you, check your email.
|
|
|
Post by NickyBlue on Mar 18, 2015 22:29:50 GMT -5
Thank you sooo much!!!
|
|