Welcome Guest    Thursday, September 09, 2010

Ai Site > Ai Forums > The Artificial Intelligence Forum > How about a little discussion, eh?

Last PostsLoginRegisterWhy Register

	Topic: How about a little discussion, eh?

kimjo625 posted 7/29/2010  14:19 Hi, I am a college student triple majoring in NeuroSci/CompSci/Mathematics. Why haven't people come up with a true AI so far? It seems to me that making true AI is not that of a hard problem. tkorrovi posted 7/29/2010  15:04 I guess this discussion was almost about the same topic http://www.ai-forum.org/topic.asp?forum_id=1&topic_id=69434 The biggest problem has always been that people don't understand these things, and it is very difficult to explain. In short, i have been in this forum for years, and many who got to know about Absolutely Dynamic System have tried to find an "intermediate solution" which is more convenient and less extreme. But none of them got anywhere, thus there may well be no intermediate solution. From that come the reasons why it is a hard problem. Such things needs a fundamental research, it is not an "applied science", "engineering science" or any such rubbish. They give huge money to fundamental research, like recently 1.5 billion eur to iter, who already said that there would be no results in 30, and maybe 100 years. But they have only ever given a few hundred thousand for Artificial Consciousness research. And the people who develop AI here, not only that they have never been given a single cent, they also have no real possibilities to publish any results of their work. And this is not just a misunderstanding.  Artificial Consciousness ADS-AC project Last edited by tkorrovi @ 7/29/2010 3:21:00 PM kimjo625 posted 7/29/2010  15:12 Hello, Tkorrovi, From your previous conversation with Will I Dream, I assume that you have made a program which handles information in somewhat unrestricted manner. Tell me more about your project. I assume the creator can also summarize their project (what it can do; what is its problem; how did you approach) Sorry, however, I do not have time to play with your program, (which may take days to extensively test its full functionality). tkorrovi posted 7/29/2010  15:35 In short, well, it may be inexact but i'll try. It is a system with changing topology of connections. All it does is creating and deleting knots, every new knot is what is the connections equivalent of symmetric difference in sets, of two old knots. This mechanism may seem "extreme", but it is the result of deriving a system which is as dynamic (changing) as possible. This system can be trained and there is evidence that it can learn. Last edited by tkorrovi @ 7/29/2010 3:37:00 PM kimjo625 posted 7/29/2010  15:51   tkorrovi wrote @ 7/29/2010 3:35:00 PM: This system can be trained and there is evidence that it can learn.   Is there a judgment sector in this program? And some instances of the evidence that it can learn would be appreciated. Thanks. tkorrovi posted 7/29/2010  16:24 I don't know what do you mean by judgement sector. It can in theory do every action. But it is nothing more than the mechanism which i described, implemented in computer. About evidence. Download the package from the project summary page and extract it to some directory. Then download all files from the download section in the end of the project main page, to the same directory (or compile yourself). Run it, load di.str and dt.dll and check the work checkbox. It would repeat "hi ", this is the first thing which i trained it to do. But there is another example, nim30.str and dtnim.str. You would not see anything very interesting on screen but the training program writes statistics to text file and calculates z score. I don't know whether you find z score 3.64 to be high enough, this is just the highest i got until i had no time any more to play with it. Last edited by tkorrovi @ 7/29/2010 4:30:00 PM kimjo625 posted 7/29/2010  16:41 Interesting. But no cigar yet, I see. So what you've made is computerized "mold" of actions. It just tries to mimic your inputs and you train it with suppressing unnecessary actions. But other than reflective nature there is no primary judgment sector to motivate its learning and there is no improving of the sector, so everything is in a stalemate after certain point. Interesting idea and interesting "impression" technique. But the fact that true AI depends a lot more on judgment sectors, this program is just a small fractional simulation of human brain capacity. Also, this program requires too much training effort. And strictly speaking, it cannot learn because your training only factor that "shaped" its functions. If you stop, training it, there is no improvement after that. Your AI does not progress "on its own". This might have worked to be a true AI if someone trains right for a long time and gain all necessary complex functions from the trainer, but trainer himself does not know the functions needed to achieve this. So it's also a dead end until we know much better. (DNA has trained humans/predecessors for millions years of evolution. But we simply don't have time for that kind of trial-and-error method.) Last edited by kimjo625 @ 7/29/2010 4:51:00 PM tkorrovi posted 7/29/2010  16:56 No, unrestricted system has no fixed goal and thus no motivation inside. The training is based on finding out what is motivating for it (cause further actions) and then using it like award. Then it develops some goal which it then would always try to achieve. This is a pure unrestricted system, like that which supposed to be in a neuron. In bigger systems there can be some "motivation", but this is likely some simple inflexible system influencing the unrestricted system. The training program is really an example of such inflexible system, it is simple and stupid, but influences almost everything in the system. kimjo625 posted 7/29/2010  17:09   tkorrovi wrote @ 7/29/2010 4:56:00 PM: No, unrestricted system has no fixed goal and thus no motivation inside. The training is based on finding out what is motivating for it (cause further actions) and then using it like award. Then it develops some goal which it then would always try to achieve. This is a pure unrestricted system, like that which supposed to be in a neuron. In bigger systems there can be some "motivation", but this is likely some simple inflexible system influencing the unrestricted system. The training program is really an example of such inflexible system, it is simple and stupid, but influences almost everything in the system.   Of course, no one would argue that human brains are COMPLETELY unrestricted. We have primary judgment sector which is often called "instinct". What you have created is a blank page/blank language/formless clay. It is true it can be used for anything, but primary judgment sector should be innately there to guide the AI's intellectual growth. What you mean by "some inflexible system" is the innate primary judgment sector (AKA instinct). If human beings were born blank like your AI, it would mean they have a homogeneous brain, meaning, there is no specialization of sectors, existing purely as a blank network of same-throughout neurons. In reality, everything is compartmentalized and specialized. Thus, it means that we do need "inflexible system" that is already present from the formation of AI. The question left is: how do we create those "inflexible system" (instinct) to successfully simulate the entire development of human brain. Last edited by kimjo625 @ 7/29/2010 5:14:00 PM tkorrovi posted 7/29/2010  20:18 Yes there are questions and there would be more questions but there is something to work with. Thank you for ralking about these things anyway, even this talking here brings things ahead. Well, considering the reality in these forums, i'm used to reply immediately and say whatever i think. It is difficult to talk that way about the biggest problems there are, but this is the only option in these forums, as even after a week all people who were interested in something, may be gone. ts_ posted 7/29/2010  23:08 C compiler is "unrestricted system", but it is very far away from "true AI". there needs to come AI programmer who will work hard and write the actual true AI program. tkorrovi posted 7/29/2010  23:25 C compiler is strictly speaking not unrestricted system but unrestrictedly extendable system. I meant unrestricted self-developing system, a fully unrestricted system. lrh9 posted 7/30/2010  17:58 The basic ideas about a.i. have been around for decades. I think one of the barriers to success is that abstraction is not the same as implementation. It is easy to come up with an idea of how a computer 'might' be intelligent, but actually programming the computer to accomplish your idea is a different task. These numbers aren't statistically accurate, but think of it this way. Out of the people who think about a.i., only 10% know how to program. Out of those, only 0.01% could be successful because of the difficulty of the task. If you are interested in a.i. you need to be an a.i. theorist as well as a software engineer. Last edited by lrh9 @ 7/30/2010 6:00:00 PM tkorrovi posted 7/30/2010  21:06 I am software engineer, but another obstacle is the lack of time. kimjo625 posted 7/31/2010  00:59   tkorrovi wrote @ 7/30/2010 9:06:00 PM: I am software engineer, but another obstacle is the lack of time.   Come on, You can learn C++ in a month or two (although you may be very inefficient in programming, but nonetheless you can make the program that can get the task done). Using that language to create "true AI" also not hard, in my opinion. Human minds are just a sophisticated function system where the functions are created and modified by its external environment and its innate "instinct". You just need a knowledge in human brain development and psychology. Just start emulating the brain processes one by one. However, you must also have a keen intuition which process is the "base-level" process and which process is "nth-level" process. Since human brains have multiple layers of "intelligence" that have resulted from millions of years of trial-error method by evolution. IT IS NOT HARD PEOPLE. ITS MUCH EASIER THAN GETTING BREAKTHROUGHS IN PHYSICS OR SOMETHING. JUST DON'T THINK IT'S IMPOSSIBLE AND FAR AWAY. PROCESS OF EVOLUTION IS NOT THAT SOPHISTICATED. HUMAN BRAIN HAS ONLY COUPLE OF BASIC PROCESSES AND OTHERS ARE JUST CHILD PROCESSES PRODUCED BY CONTROLLED DYNAMICS AND CONTROLLED FLEXIBILITY. I thought people would have had "true AI" years ago. What is the problem, people? Last edited by kimjo625 @ 7/31/2010 1:04:00 AM tkorrovi posted 7/31/2010  03:04 No one can learn c++ in a month or two. I would rather say, no one can learn c++ in a lifetime, c++ is much too large for anyone to know it completely. One can learn c completely, though. But this is really irrelevant here, sure there are people who can do programming. But i see that you have your own idea about what a True AI should be like. And you seem to reject everything which is different from that your idea. At that, the problem is that you have not even explained that idea, more than calling it an "instinct system" or something. When you would say what your idea is, only then can others find out whether it is something important or not. In fact, you should be able to simulate your solution at least in some way in the computer, and make that code available. If it is really worth, i and many others here would be very interested about it. But just saying that you know what the solution should be, is by far not enough. At least some parts of the brain have to be unrestricted to avoid some inevitable inconsistencies which would make more advanced development impossible. This part was cleary missing, and there is a lot of knowledge about many kind of less flexible systems. At least it seems logical that to understand these things, it's necessary to research the least known part. And you said it should be simple. From what i have done i can say that it is not simple, it is possible but not simple. Like the self-developing systems, they develop, the things what are there start to develop very easily, and at that they all depend on each other. This is just because these systems are self-developing. Now try to keep them in a jar, try to find out when an error started when everything would be quickly changed to something else. Not to talk about the deep philosophy involved, not to talk about the slowness of computers compared to natural systems, etc. No, it's not easy. Last edited by tkorrovi @ 7/31/2010 3:08:00 AM ts_ posted 7/31/2010  03:07   kimjo625 wrote @ 7/31/2010 12:59:00 AM: IT IS NOT HARD PEOPLE. ITS MUCH EASIER THAN GETTING BREAKTHROUGHS IN PHYSICS OR SOMETHING. JUST DON'T THINK IT'S IMPOSSIBLE AND FAR AWAY. PROCESS OF EVOLUTION IS NOT THAT SOPHISTICATED. HUMAN BRAIN HAS ONLY COUPLE OF BASIC PROCESSES AND OTHERS ARE JUST CHILD PROCESSES PRODUCED BY CONTROLLED DYNAMICS AND CONTROLLED FLEXIBILITY.   i hope u r being ironic, cos if u arent u have no clue about AI. the truth is the exact opposite of what u wrote. will i dream posted 7/31/2010  05:16 kimjo625 - go easy man c++ in a month - jesus man software is an artform c++ is an artform c++ in a month - is just plain stupid kimjo625 - listen - there are people here (That I know very well) who have persued AI for 20/30 years I dont know you - but I do know that if you have spent just 1 year in the persuit if AI you would begin to realise the sheer scale of the complexity of this subject and would not post silly things There are people here that I personally have total respect for - for their efforts and willingness to share their work openly to all as you said yourself "I thought people would have had "true AI" years ago. What is the problem, people?" Yeh - thousands thought the same - and they didnt find it - military research, massive funding - and zero does that not give you a hint but if you have the answer - fire away man - were all ears go easy man Last edited by will i dream @ 7/31/2010 5:20:00 AM kimjo625 posted 7/31/2010  06:47 tkorrovi wrote @ 7/31/2010 3:04:00 AM: No one can learn c++ in a month or two. I would rather say, no one can learn c++ in a lifetime, c++ is much too large for anyone to know it completely. One can learn c completely, though. But this is really irrelevant here, sure there are people who can do programming. [Grammer of C++ you can learn in a month or two (only about 60 grammar). Making an essay with C++ you can also do that in a month or two. Making sophisticated book/symphony with C++, you may take lifetime] But i see that you have your own idea about what a True AI should be like. And you seem to reject everything which is different from that your idea. [I tend to reject things that do not make sense to me. (but I do ponder for certain time period before rejecting) After thinking about the topic about 10 logic steps into, IF I find unjustifiable flaws, I reject.] At that, the problem is that you have not even explained that idea, more than calling it an "instinct system" or something. [instinct system is your "inflexible system" or your "training program". It'd be stupid to argue that one can make true AI without implementing "inflexible" or "innate" systems that will guide AI's growth.] When you would say what your idea is, only then can others find out whether it is something important or not. In fact, you should be able to simulate your solution at least in some way in the computer, and make that code available. If it is really worth, i and many others here would be very interested about it. But just saying that you know what the solution should be, is by far not enough. [No offense, but, no thanks, I'm not going to share my gold mine. Only some of my ideas will be shared here.] At least some parts of the brain have to be unrestricted to avoid some inevitable inconsistencies which would make more advanced development impossible. This part was cleary missing, and there is a lot of knowledge about many kind of less flexible systems. At least it seems logical that to understand these things, it's necessary to research the least known part. [Agreed.] And you said it should be simple. From what i have done i can say that it is not simple, it is possible but not simple. Like the self-developing systems, they develop, the things what are there start to develop very easily, and at that they all depend on each other. This is just because these systems are self-developing. Now try to keep them in a jar, try to find out when an error started when everything would be quickly changed to something else. Not to talk about the deep philosophy involved, not to talk about the slowness of computers compared to natural systems, etc. No, it's not easy. [Agreed on slowness compared to neuron brain form. However, I do not wish to talk about moral/ethical implication involved in creating AI that exceeds human mental/intellectual capacity.] kimjo625 posted 7/31/2010  06:50 ts_ wrote @ 7/31/2010 3:07:00 AM: i hope u r being ironic, cos if u arent u have no clue about AI. the truth is the exact opposite of what u wrote. You seem to know something, tell me why it is the opposite of what I've written. Let's have a fun discussion.

1  2    Send email to user      Reply with quote      Edit message Forums Home     Hal and other child machines     Alan and other chatbots     Language Mind and Consciousness

	Contact Us		Terms of Use