Re: AW: [agi] How general can be and should be AGI?
Mike Tintner wrote: Charles: Flaws in Hamlet: I don't think of this as involving general intelligence. Specialized intelligence, yes, but if you see general intelligence at work there you'll need to be more explicit for me to understand what you mean. Now determining whether a particular deviation from iambic pentameter was a flaw would require a deep human intelligence, but I don't feel that understanding of how human emotions are structured is a part of general intelligence except on a very strongly superhuman level. The level where the AI's theory of your mind was on a par with, or better than, your own. Charles, My flabber is so ghasted, I don't quite know what to say. Sorry, I've never come across any remarks quite so divorced from psychological reality. There are millions of essays out there on Hamlet, each one of them different. Why don't you look at a few?: http://www.123helpme.com/search.asp?text=hamlet I've looked at a few (though not those). In college I formed the definite impression that essays on the meaning of literature were exercises in determining what the instructor wanted. This isn't something that I consider a part of general intelligence (except as mentioned above). ... The reason over 70 per cent of students procrastinate when writing essays like this about Hamlet, (and the other 20 odd per cent also procrastinate but don't tell the surveys), is in part that it is difficult to know which of the many available approaches to take, and which of the odd thousand lines of text to use as support, and which of innumerable critics to read. And people don't have a neat structure for essay-writing to follow. (And people are inevitably and correctly afraid that it will all take if not forever then far, far too long). The problem is that most, or at least many, of the approaches are defensible, but your grade will be determined by the taste of the instructor. This isn't a problem of general intelligence except at a moderately superhuman level. Human tastes aren't reasonable ingredients for an entry level general intelligence. Making it a requirement merely ensures that one will never be developed (whose development attends to your theories of what's required). ... In short, essay writing is an excellent example of an AGI in action - a mind freely crossing different domains to approach a given subject from many fundamentally different angles. (If any subject tends towards narrow AI, it is normal as opposed to creative maths). I can see story construction as a reasonable goal for an AGI, but at the entry level they are going to need to be extremely simple stories. Remember that the goal structures of the AI won't match yours, so only places where the overlap is maximal are reasonable grounds for story construction. Otherwise this is an area for specialized AIs, which isn't what we are after. Essay writing also epitomises the NORMAL operation of the human mind. When was the last time you tried to - or succeeded in concentrating for any length of time? I have frequently written essays and other similar works. My goal structures, however, are not generalized, but rather are human. I have built into me many special purpose functions for dealing with things like plot structure, family relationships, relative stages of growth, etc. As William James wrote of the normal stream of consciousness: Instead of thoughts of concrete things patiently following one another in a beaten track of habitual suggestion, we have the most abrupt cross-cuts and transitions from one idea to another, the most rarefied abstractions and discriminations, the most unheard-of combinations of elements, the subtlest associations of analogy; in a word, we seem suddenly introduced into a seething caldron of ideas, where everything is fizzling and bobbing about in a state of bewildering activity, where partnerships can be joined or loosened in an instant, treadmill routine is unknown, and the unexpected seems the only law. Ditto: The normal condition of the mind is one of informational disorder: random thoughts chase one another instead of lining up in logical causal sequences. Mihaly Csikszentmihalyi Ditto the Dhammapada, Hard to control, unstable is the mind, ever in quest of delight, When you have a mechanical mind that can a) write essays or tell stories or hold conversations [which all present the same basic difficulties] and b) has a fraction of the difficulty concentrating that the brain does and therefore c) a fraction of the flexibility in crossing domains, then you might have something that actually is an AGI. You seem to be placing an extremely high bar in place before you will consider something an AGI. Accepting all that you have said, for an AGI to react as a human would react would require that the AGI be strongly superhuman. More to the point, I wouldn't DARE create an AGI which had motivations similar to
AW: AW: AW: [agi] How general can be and should be AGI?
Matt Mahoney [mailto:[EMAIL PROTECTED] wrote Object oriented programming is good for organizing software but I don't think for organizing human knowledge. It is a very rough approximation. We have used O-O for designing ontologies and expert systems (IS-A links, etc), but this approach does not scale well and does not allow for incremental learning from examples. It totally does not work for language modeling, which is the first problem that AI must solve. I agree that the O-O paradigm is not adequate to model all learning algorithms and models we use. My own example of recognizing voices should show that I have doubts that we use O-O models in our brain for everything of our environment. I think our brain learns a somewhat a hierarchical model of the world. And the algorithm for the low level (e.g. voices, sounds) are probably complete different from the algorithms for higher levels of our models. It is evident that a child has learning capabilities that are far beyond those from an adult. The reason is not only that the child's brain is nearly empty. The physiological architecture is different to some degree. So we can expect that learning the basic low levels of a world model requires algorithms which we only have had as a child. And the result of that learning is to some degree used for bias in later learning algorithm when we are adult. For example we had to learn to extract syllables from the sound wave of spoken language. Learning the grammar rules are in higher levels. Learning semantics is still higher and so on. But it is a matter of fact that we use an O-O like model in the top-levels of our world. You can see this also from language grammar. Subjects objects, predicates, adjectives have their counterparts in the O-O paradigm. A photo of a certain scene is physically an array of colored pixels. But you can ask a human what he sees. And a possible answer could be: Well, there is a house. A man walks to the door. It wears a blue shirt. A woman looks through the window ... Obviously, the answer shows a lot how people model the world in their top-level (= conscious) And obviously the model consists of interacting objects with attributes and behavior. So knowledge representation at higher levels is indeed O-O like. I think your and my answer show that we do not use a single algorithm which is responsible to extract all the regularities from our perceptions. And more important: There is physiological and psychological evidence that the algorithms we use change to some degree during the first decade of our life. --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: [agi] Interesting HYPED approach to controlling animated characters...
The only thing to learn from here is the way they managed to build hype around their technology. Possibly appropriate here :-) Technology wise, we're talking Lua state machines, genetic algorithms that are manually tweaked for every special case. The resulting neural nets are pretty much only used to drive their active ragdolls towards known poses. http://aigamedev.com/editorial/naturalmotion-euphoria Even game developer's aren't swallowing the hype on this one! Best, Alex Alex Champandard Editor Consultant AiGameDev.com Ben Goertzel wrote: Now this looks like a fairly AGI-friendly approach to controlling animated characters ... unfortunately it's closed-source and proprietary though... http://en.wikipedia.org/wiki/Euphoria_%28software%29 ben --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?; Powered by Listbox: http://www.listbox.com --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: AW: [agi] How general can be and should be AGI?
Charles, We're still a few million miles apart :). But perhaps we can focus on something constructive here. On the one hand, while, yes, I'm talking about extremely sophisticated behaviour in essaywriting, it has generalizable features that characterise all life. (And I think BTW that a dog is still extremely sophisticated in its motivations and behaviour - your idea there strikes me as evolutionarily naive). Even if a student has an extremely dictatorial instructor, following his instructions slavishly, will be, when you analyse it, a highly problematic, open-ended affair, and no slavish matter - i.e. how he is to apply some general, say, deconstructionist criticism instructions and principles and translate them into a v. complex essay. In fact, it immediately strikes me such essaywriting, and all essaywriting, and most human activities and animal activities will be a matter of hierarchical goals - of, off the cuff, something v. crudely like - write an essay on Hamlet - decide general approach... use deconstructionist approach - find contradictory values in Hamlet to deconstruct...etc. But all life, I guess, must be organized along those lines - the simplest worm must start with something crudely like : find food to eat...decide where food may be located decide approach to food location etc.. (which in turn will almost always be conflicting with opposed emotions/motivations/goals like get some more sleep ..stay cuddled up in burrow.. ) And even, pace Koestler and others, v. simple actions, like reaching out for food in a kitchen, can be a hierarchical affair, with only the general direction and goal decided to begin with, and more specific targeting of arm and shaping of hand, only specified at later stages of the action. Hierarchical goals are surely fundamental to general intelligence. Interestingly, when I Google hierarchical goals and AI, I get v. little - except from our immediate friends, gamers - and this from: Programming Game AI by Example Mat Buckland: Chapter 9: Hierarchical Goal Based Agents This chapter introduces agents that are motivated by hierarchical goals. This type of architecture is far more flexible than the one described in Chapter 2 allowing AI programmers to easily imbue game characters with the brains necessary to do all sorts of funky stuff. Discussion, code and demos of: atomic goals, composite goals, goal arbitration, creating goal evaluation functions, implementation in Raven, using goal evaluations to create personalities, goals and agent memory, automatic resuming of interrupted activities, negotiating special path obstacles such as elevators, doors or moving platforms, command queuing, scripting behavior. Anyone care to comment about using hierarchical goals in AGI or elsewhere? Charles: Flaws in Hamlet: I don't think of this as involving general intelligence. Specialized intelligence, yes, but if you see general intelligence at work there you'll need to be more explicit for me to understand what you mean. Now determining whether a particular deviation from iambic pentameter was a flaw would require a deep human intelligence, but I don't feel that understanding of how human emotions are structured is a part of general intelligence except on a very strongly superhuman level. The level where the AI's theory of your mind was on a par with, or better than, your own. Charles, My flabber is so ghasted, I don't quite know what to say. Sorry, I've never come across any remarks quite so divorced from psychological reality. There are millions of essays out there on Hamlet, each one of them different. Why don't you look at a few?: http://www.123helpme.com/search.asp?text=hamlet I've looked at a few (though not those). In college I formed the definite impression that essays on the meaning of literature were exercises in determining what the instructor wanted. This isn't something that I consider a part of general intelligence (except as mentioned above). ... The reason over 70 per cent of students procrastinate when writing essays like this about Hamlet, (and the other 20 odd per cent also procrastinate but don't tell the surveys), is in part that it is difficult to know which of the many available approaches to take, and which of the odd thousand lines of text to use as support, and which of innumerable critics to read. And people don't have a neat structure for essay-writing to follow. (And people are inevitably and correctly afraid that it will all take if not forever then far, far too long). . This isn't a problem of general intelligence except at a moderately superhuman level. Human tastes aren't reasonable ingredients for an entry level general intelligence. Making it a requirement merely ensures that one will never be developed (whose development attends to your theories of what's required). ... In short, essay writing is an excellent example of an AGI in action - a mind
RE: [agi] help me,please for books for agi and mind in pdf
Bruno Frandemiche asked for online AGI-related text. If you're adventurous, I'd recommend the Workshop proceedings from 2006: http://www.agiri.org/wiki/Workshop_Proceedings and the conference proceedings from AGI-08: http://www.agi-08.org/papers --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: [agi] upcoming oral at Princeton
My guess would be that this kind of approach will only be partly successful, since fundamentally it's only based upon an elaborate kind of 2D template matching. I think what actually happens is that during early childhood experience we are able to statistically correlate certain types of geometry with the patterns of light falling upon out retinas. When we later view flat images we're able to retrieve the associated type of geometry and imagine what the object might look like from various angles, even if we have only seen it once. I expect that biological vision systems are fundamentally designed for 3D understanding of the world, since this is of high adaptive value, rather than a sort of 2D screen scraping or the retina. 2008/5/2 J Storrs Hall, PhD [EMAIL PROTECTED]: Just saw this announcement go by: Abstract: Constructing ImageNet Data sets are essential in computer vision and content based image retrieval research. We present the work in progress for constructing ImageNet, a large scale image data set based on the Princeton WordNet. The goal is to associate more than 1000 clean images with each node of WordNet, which consists of ~30,000 ( estimated ) imagable nodes. We build a prototype system for constructing ImageNet, as a first step toward large scale deployment. For each node of WordNet, which is a synonym set (synset) for a single concept, we collect candidate images from the Internet and clean up them with semi-automatic labeling. We train boosting classifiers from human labeled data and use active learning to substantially speed up the labeling process. We also developed a web interface for massive online human labeling. We demonstrate the effectiveness of our system with results from a subset of synsets. Reading list: Text book: Pattern Recognition and Machine Learning, Christopher M. Bishop, 2006. Chapter 1,2,8,14. Modern Operating System, Tanenbaum. Papers: Animals on the Web, Berg, Forsyth, CVPR06 OPTIMOL: automatic Online Picture collecTion via Incremental MOdel Learning, Li, Wang, Fei-Fei, CVPR07 Learning Object Categories from Google's image Search, Fergus, Fei-Fei, Perona, Zissermaman, ICCV05 Harvesting Image Databases from the Web, Scroff, Zisserman, ICCV07 From Aardvark to Zorro: A Benchmark of Mammal Images, Fink, Ullman, NIPS05 Tiny Images, Torralba, Fergus, Freeman, TechReport MIT, 2007 Labeling Images with a Computer Game. Luis von Ahn and Laura Dabbish, CHI04 LabelMe: a database and web-based tool for image annotation, Russell, Torralba, IJCV07 Introduction to a large scale general purpose groundtruth dataset: methodology, annotation tool, and benchmarks, Z.Y. Yao, X. Yang, and S.C. Zhu, EMMCVPR07 Combining active and semi-supervised learning for spoken language understanding, Tur, Hakkani-Tur, Schapire, Speech Communication, 05 Online boosting and vision, CVPR06 --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?; Powered by Listbox: http://www.listbox.com --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
[agi] Panda: a pattern-based programming system
Readers of these lists might enjoy the refereed paper Overview of the Panda Programming System (http://www.jot.fm:80/issues/issue_2008_05/article1/) described in the following abstract: This article provides an overview of a pattern-based programming system, named Panda, for automatic generation of high-level programming language code. Many code generation systems have been developed [2, 3, 4, 5, 6] that are able to generate source code by means of templates, which are defined by means of transformation languages such as XSL, ASP, etc. But these templates cannot be easily combined because they map parameters and code snippets provided by the programmer directly to the target programming language. On the contrary, the patterns used in a Panda program generate a code model that can be used as input to other patterns, thereby providing an unlimited capability of composition. Since such a composition may be split across different files or code units, a high degree of separation of concerns [15] can be achieved. A pattern itself can be created by using other patterns, thus making it easy to develop new patterns. It is also possible to implement an entire programming paradigm, methodology or framework by means of a pattern library: design patterns [8], Design by Contract [12], Aspect-Oriented Programming [1, 11], multi-dimensional separation of concerns [13, 18], data access layer, user interface framework, class templates, etc. This way, developing a new programming paradigm does not require to extend an existing programming system (compiler, runtime support, etc.), thereby focusing on the paradigm concepts. The Panda programming system introduces a higher abstraction level with respect to traditional programming languages: the basic elements of a program are no longer classes and methods but, for instance, design patterns and crosscutting concerns [1, 11]. Cheers, Brad --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: [agi] upcoming oral at Princeton
Hi Josh, I briefly looked at the ImageNet description at the Princeton WordNet site. It does not reveal whether the images are open source to the extent this new data can be linked and distributed with WordNet, which has a very permissive license. -Steve Stephen L. Reed Artificial Intelligence Researcher http://texai.org/blog http://texai.org 3008 Oak Crest Ave. Austin, Texas, USA 78704 512.791.7860 - Original Message From: J Storrs Hall, PhD [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Friday, May 2, 2008 12:22:40 PM Subject: [agi] upcoming oral at Princeton Just saw this announcement go by: Abstract: Constructing ImageNet Data sets are essential in computer vision and content based image retrieval research. We present the work in progress for constructing ImageNet, a large scale image data set based on the Princeton WordNet. The goal is to associate more than 1000 clean images with each node of WordNet, which consists of ~30,000 ( estimated ) imagable nodes. We build a prototype system for constructing ImageNet, as a first step toward large scale deployment. For each node of WordNet, which is a synonym set (synset) for a single concept, we collect candidate images from the Internet and clean up them with semi-automatic labeling. We train boosting classifiers from human labeled data and use active learning to substantially speed up the labeling process. We also developed a web interface for massive online human labeling. We demonstrate the effectiveness of our system with results from a subset of synsets. Reading list: Text book: Pattern Recognition and Machine Learning, Christopher M. Bishop, 2006. Chapter 1,2,8,14. Modern Operating System, Tanenbaum. Papers: Animals on the Web, Berg, Forsyth, CVPR06 OPTIMOL: automatic Online Picture collecTion via Incremental MOdel Learning, Li, Wang, Fei-Fei, CVPR07 Learning Object Categories from Google's image Search, Fergus, Fei-Fei, Perona, Zissermaman, ICCV05 Harvesting Image Databases from the Web, Scroff, Zisserman, ICCV07 From Aardvark to Zorro: A Benchmark of Mammal Images, Fink, Ullman, NIPS05 Tiny Images, Torralba, Fergus, Freeman, TechReport MIT, 2007 Labeling Images with a Computer Game. Luis von Ahn and Laura Dabbish, CHI04 LabelMe: a database and web-based tool for image annotation, Russell, Torralba, IJCV07 Introduction to a large scale general purpose groundtruth dataset: methodology, annotation tool, and benchmarks, Z.Y. Yao, X. Yang, and S.C. Zhu, EMMCVPR07 Combining active and semi-supervised learning for spoken language understanding, Tur, Hakkani-Tur, Schapire, Speech Communication, 05 Online boosting and vision, CVPR06 --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?; Powered by Listbox: http://www.listbox.com Be a better friend, newshound, and know-it-all with Yahoo! Mobile. Try it now. http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re : [agi] help me,please for books for agi and mind in pdf
thank you derek i was reading all this bye - Message d'origine De : Derek Zahn [EMAIL PROTECTED] À : agi@v2.listbox.com Envoyé le : Vendredi, 2 Mai 2008, 15h18mn 09s Objet : RE: [agi] help me,please for books for agi and mind in pdf Bruno Frandemiche asked for online AGI-related text. If you're adventurous, I'd recommend the Workshop proceedings from 2006: http://www.agiri.org/wiki/Workshop_Proceedings and the conference proceedings from AGI-08: http://www.agi-08.org/papers agi | Archives | Modify Your Subscription __ Do You Yahoo!? En finir avec le spam? Yahoo! Mail vous offre la meilleure protection possible contre les messages non sollicités http://mail.yahoo.fr Yahoo! Mail --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Language learning (was Re: AW: AW: AW: AW: [agi] How general can be and should be AGI?)
--- Dr. Matthias Heger [EMAIL PROTECTED] wrote: Matt Mahoney [mailto:[EMAIL PROTECTED] wrote Actually that's only true in artificial languages. Children learn words with semantic content like ball and milk before they learn function words like the and of, in spite of their higher frequency. Before they learn the words and their meanings they have to learn to recognize the sounds for the words. And even if they use words like with of and the later they must be able to separate these function-words and relation-words from object-words before they learn any word. But separating words means classifying words and that means knowledge of grammar for a certain degree. Lexical segmentation is learned before semantics, but other grammar is learned afterwards. Babies learn to segment continuous speech into words at 7-10 months [1]. This is before they learn their first word, but is detectable because babies will turn their heads in preference to segmentable speech. It is also possible to guess word divisions in text without spaces given only a statistical knowledge of letter n-grams [2]. Natural language has a structure that makes it easy to learn incrementally from examples with a sufficiently powerful neural network. It must, because any unlearnable features will disappear. Matt Mahoney [mailto:[EMAIL PROTECTED] wrote Techniques for parsing artificial languages fail for natural languages because the parse depends on the meanings of the words, as in the following example: - I ate pizza with pepperoni. - I ate pizza with a fork. - I ate pizza with a friend. In days of early AI the O-O paradigm was not so sophisticated as it is today. The phenomenon of your example is well-known in O-O paradigm and is modeled by overwritten functions which means that Objects may have several functions with the same name but with different signatures. eat(Food f) eat(Food f, ListSideDish l) eat (Food f, ListTool l) eat (Food f, ListPeople l) ... This type of knowledge representation has been tried and it leads to a morass of rules and no intuition on how children learn grammar. We do not know how many grammar rules there are, but it probably exceeds the number of words in our vocabulary, given how long it takes to learn. I think, it is clear that there are representations like classes, objects, relation between objects, attributes of objects. But the crucial questions are: How did we and do we build our O-O models? How created the brain abstract concepts like ball and milk? How do we find classes, objects and relations? We need to understand how children learn grammar without any concept of what a noun or a verb is. Also, how do people learn hierarchical relationships before they learn what a hierarchy is? 1. Jusczyk, Peter W. (1996), Investigations of the word segmentation abilities of infants, 4'th Intl. Conf. on Speech and Language Processing, Vol. 3, 1561-1564. 2. http://cs.fit.edu/~mmahoney/dissertation/lex1.html -- Matt Mahoney, [EMAIL PROTECTED] --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: [agi] Panda: a pattern-based programming system
A thousand thank yous. --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
AW: Language learning (was Re: AW: AW: AW: AW: [agi] How general can be and should be AGI?)
Matt Mahoney [mailto:[EMAIL PROTECTED] wrote eat(Food f) eat(Food f, ListSideDish l) eat (Food f, ListTool l) eat (Food f, ListPeople l) ... This type of knowledge representation has been tried and it leads to a morass of rules and no intuition on how children learn grammar. We do not know how many grammar rules there are, but it probably exceeds the number of words in our vocabulary, given how long it takes to learn. As I said, my intention is not to find a set of O-O like rules to create AGI. The fact that early approaches failed to build AGI by a set of similar rules does not prove, that AGI cannot consist of such rules. For example, there were also approaches to create AI by biological inspired neural networks with some minor success but there was not the real breakthrough too. So this does not prove anything but that the problem of AGI is not so easy to solve. The brain is still a black box regarding many phenomenon. We can analyze our own conscious thoughts and our communication which is nothing else than sending ideas and thoughts from one brain to the other brain via natural language. I am convinced, that the structure and contents of our language is not independent of the internal representation of knowledge. And from language we must conclude that there are O-O like models in the brain because the semantics is O-O. There might be millions of classes and relationships. And surely every day or night, the brain refactores parts of its model. The roadmap to AGI will probably be top-down and not bottom-up. The bottom-up approach is used by biological evolution. Creating AGI by software engineering means that we first must know where we want to go and then how to go there. Human language and conscious thoughts suggests that AGI must be able to represent the world O-O like at the top-level. So this ability is the answer for the question where we want to go. Again, this does not mean that we must find all the classes and objects. But we must find an algorithm that generates O-O like models of its environment based on its perceptions and some bias where the need for the bias can be proven from reasons of performance. We can expect that the top-level architecture of AGI is the easiest part in an AGI project, because the contents of our own consciousness gives us some hints (but not all) how our own world representation works at the top-level. And this is O-O in my opinion. There is also a phenomenon of associations between patterns (classes). But this is just a question of retrieving information and attention to relevant parts of the O-O model and is no contradiction to the existence of the O-O paradigm. When we go to lower levels, it is clear that difficulties arise. The reason is that we have no possibility for conscious introspection of the low levels in our brain. Science gives us hints mainly for the lowest levels (chemistry, physics...). So the medium layers of AGI will be the most difficult layers. By the way this is also often the case in normal software. In the medium layers there will be base functionalities and the framework for the top-level. --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
[agi] Re: AW: Language learning
--- Dr. Matthias Heger [EMAIL PROTECTED] wrote: So the medium layers of AGI will be the most difficult layers. I think if you try to integrate a structured or O-O knowledge base at the top and a signal processing or neural perceptual/motor system at the bottom, then you are right. We can do a thought experiment to estimate its cost. Put a human in the middle and ask how much effort or knowledge is required. An example would be translating a low-level natural language question to a high level query in SQL or Cycl or whatever formal language the KB uses. I think you can see that for a formal representation of common sense knowledge, that the skill required for this interface is at a higher level than the knowledge actually represented at the top level. If this knowledge was stored in the human brain, then it could be retrieved faster, and by someone who had no special skills in understanding a formal language. But there are still some applications where this design makes sense. One example would be a calculator. At the low level, you have a question like how many square inches in a third of an acre? The middle level converts this to an equation and punches the numbers into the top level calculator. This is preferable to the human doing the arithmetic. A database would be another example. Where it doesn't make sense is when the top level is doing something that humans are already good at. It would make more sense to figure out how humans learn and represent common sense instead of guessing. We can do experiments in cognitive psychology. What can people learn? remember? perceive? -- Matt Mahoney, [EMAIL PROTECTED] --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
AW: [agi] Re: AW: Language learning
I think it is even more complicated. The flow of signals in the brain does not move only from low levels to high levels. The modules communicate in both directions. And as far as I know there is already evidence for this from cognitive science. If you want to recognize objects in pictures you need to find the edges or boundaries. But the other direction works too. If you know the object because someone tells you what is on the picture or because you use other knowledge about the picture then it is easier for you to detect the edges of the object. A thought experiment is a good idea. Let's say we have a robot in the garden and ask him: How many apples are on the tree? The robot is assumed to be experienced, i.e. it should have a sufficient world model to understand and answer the question. I make this assumption at this point, because first we have to answer the question where we want to go. In the following I describe a hypothetical process in the robot's brain. Note that I assume the robot has learned most of this process (classes, interactions of objects) with past experiences. But of course some classes and information flows it must have had from its first day on. Ok. The robot gets the sound wave and its low level modules try to recognize known patterns in this wave. First it recognizes a voice pattern. This triggers a voice object. This triggers different objects. For example: A speech object, an information object, a person object and perhaps a lot of other objects. The person object analyzes the sound wave only to obtain information who is speaking. The speech object only tries to figure out what language is spoken. But here is already a trick. The person object detects that the voice comes from person Matt. And the person object has the value English in its attribute language. The objects inform each other in parallel about their values and the speech object receives the value English from the person object. By this it is easier for the speech object to recognize the language because it can use a useful hypothesis and it will activate certain English tester objects. All these objects make their own analysis and use information about results of other objects. After a short time, certain important objects are active: A question object of the type quantity question. Word objects of different grammar types with values How Many Apples APPLIES Are On The Tree There is something special with the words APPLES and APPLIES. They have the same number attribute value (=third word in the question) and they have a probability value of 50%. This means that the robot is not quite sure whether the third word was APPLES or APPLIES. The question object is already a higher level object. It does not use the sound wave input but the set of active word objects. The question object contains a subject object which itself contains a GrammarSubject object and a GivenHints object. It has to decide whether the subject is APPLES or TREE. The robot knows from past experience that subjects of quantity questions are in plural. For any attribute of any object there is a setter method with a learnable validate function. So the subject object accepts only the word APPLES for its GRammarSubject object. This fact also increases the probability value of the word APPLES and decreases the probability for APPLIES. Finally the robot has the complete question object which activates a goal object: Answer the question! This was just the low level. At this point the robot must understand what he really shall do. He knows from experience that he gets reward if it answers the active question object whenever a corresponding goal object is active. An answer for a quantity question must be a number. The number is the result of a count process which corresponds to the subject of the quantity question. Ok. We are in one of the medium levels of AGI. And I already wonder how our robot should have learned the low level I described so far. And I stop here because everything is too complex now. But these thought experiments are strongly necessary if we want to create AGI -Ursprüngliche Nachricht- Von: Matt Mahoney [mailto:[EMAIL PROTECTED] Gesendet: Samstag, 3. Mai 2008 01:27 An: agi@v2.listbox.com Betreff: [agi] Re: AW: Language learning --- Dr. Matthias Heger [EMAIL PROTECTED] wrote: So the medium layers of AGI will be the most difficult layers. I think if you try to integrate a structured or O-O knowledge base at the top and a signal processing or neural perceptual/motor system at the bottom, then you are right. We can do a thought experiment to estimate its cost. Put a human in the middle and ask how much effort or knowledge is required. An example would be translating a low-level natural language question to a high level query in SQL or Cycl or whatever formal language the KB uses. I think you can see that for a formal representation of common sense knowledge, that the skill required for this
