Seems like a lot of procrastination going on. How about one architectural diagram and some code? ~PM
Date: Thu, 7 May 2015 08:45:13 -0400 Subject: Re: [agi] Re: Starting to Define Algorithms that are More Powerfulthan Narrow AI From: [email protected] To: [email protected] On Wed, May 6, 2015 at 8:42 PM, Benjamin Kapp <[email protected]> wrote: In my experience when ever I come up with a "novel" idea in a field whose scholarly journals I haven't read, then almost always (~99 out of 100) someone else has thought of it already. But when I read all of the scholarly journals I find that I can almost always come up with novel approaches. One reason for this is that some of the journal articles will explicitly say what directions future experiments might prove fruitful :) Of course people should study other people's experiments and they will often find that ideas that they thought were novel had already been tried. But, there is another side of the story. Perhaps if you had been a little more determined and actually tried some of the 'novel' ideas you might have been able to make them work better than the scholarly journalists had been able to. To some extent it can depend on how you apply the idea. If you have had some novel approaches that the journal articles had explicitly described might be fruitful then where is your experimental results supporting your intuition about it? My thesis, in this thread, is that we can use reason-based reasoning to try to explain why some methods should go beyond narrow ai but then we have to design experiments to verify these contentions. Since we cannot jump into a full-fledged AGI program these experiments will typically be preliminary tests. But the point is that we can design them to test certain hypotheses in a step wise manner.Jim Bromer On Wed, May 6, 2015 at 8:42 PM, Benjamin Kapp <[email protected]> wrote: One lesson you can learn is that if you wish to advance the state of the art it is important to know what the state of the art is. The Wright Brothers worked with someone who literally wrote the book on the state of the art at the time. So perhaps you yourself need not know the state of the art, perhaps you need only collaborate with those who do. In my experience when ever I come up with a "novel" idea in a field whose scholarly journals I haven't read, then almost always (~99 out of 100) someone else has thought of it already. But when I read all of the scholarly journals I find that I can almost always come up with novel approaches. One reason for this is that some of the journal articles will explicitly say what directions future experiments might prove fruitful :) Of course the Wright Brothers also benefited from an obsession with flight, but if you are reading this then (for the field of AI) you probably already have that aspect covered :) Comparisons are a problem if you use them to express your meaning, because in general those things being compared often have many properties, but the author/speaker usually only intends a subset of those properties to be compared, however he often doesn't explicitly denote which of these properties he is referring to as such the audience may erroneously believe the author/speaker intends a different subset of those properties and thus the intended meaning fails to be expressed, hence why many thinkers through out time have complained of the heavy usage of metaphors and similes by a great many people whose works stand at the foundation of western thought. For example see the complete works of Aristotle. ;) If what I said about knowing the current state of knowledge is true, then it would seem one ought to be asking questions such asWhat are the most important journals to read in the field of AGI?Who knows the current state of knowledge in the domain of AGI?What books have they written?What problems are they working on?What problems do they think we need to solve to bring about AGI? Does this make sense? On Wed, May 6, 2015 at 6:18 PM, Jim Bromer <[email protected]> wrote: I guess Ben Kapp was saying that because of the advances made in relevant technology at the time that the Wright brothers knew that a plane was feasible and one could argue that success in AGI might just be based on being there at the right time. I could suggest that because the Wright's father was a bishop of the United Church of the Brethren in Christ and they had been brought up as believers that they spiritually knew that could succeed and that is what is needed. The point is still that they succeeded by combining insightful theory with actual design and experimentation and redesigns. So I have thought about abandoning my attempt to create a stable data-management system (which always seems to need a few more months to complete) and start doing some experiments with a simpler framework. But I don't have to abandon anything, I can work with both. But now I have to simplify my theories somehow to make them light enough to be feasible but at the same time try to keep them different from the normal AI narroware. And I think I finally see it. I need to design the basics around acting more like a human being in a few ways. I don't mean that I am trying to make a human-level AGI program, I mean that I can try to find some elementary features that are different than what is typically found in an AI program and which might be useful in artificial thought. So this is the opposite of waiting for something that I can't quite figure out to emerge in the data, instead I am going to try to make a simple AI program with some features that other simple AI programs seem to lack. So it won't be a human-like AGI program but it will have a few primitive characteristics that you see (in more elaborate forms) in human thinking.) But some people say that is the problem with AI research. Someone makes something that looks cool at first but then no one can build better AI programs on it just because it is narrow AI. I am saying that if I do succeed at the first step then I can continue using the refine your design - test it - refine the experiment - try the new test on it - and then refine your design - loop. Jim Bromer On Wed, May 6, 2015 at 5:38 PM, Jim Bromer <[email protected]> wrote: > Nevertheless, despite their brilliant successes, the Wrights never > completely understood quantitatively the problem of stability of > rotational motions. They shared that deficiency with all their > contemporary inventors, for the same reason: they never wrote or > considered equations for rotational motions.Without the benefit of > that formalism, they could not understand the true essence of > stability of rotations. As a practical matter, they could not identify > the physical contributions to stability, a failure that had > significant consequences for their work: > > But the argument that the Wright brothers knew enough aerodynamic > equations so that the plane was just an engineering problem misses the > point. I don't agree with the statement but it is not relevant to what > you were originally saying. The fact is that they did a lot of > experiments, changed their designs and did a lot more experiments. > Right now there are enough mathematical and pseudo-code theories about > AI to begin experimenting with actual AGI functions even though it > does seem obvious that there are crucial theories that are still > missing. So ignoring the fact that there have been great advances in > AI it is -as if- we are analogously at the point the Wrights were. So > the next step is to create more insight about the problem as you begin > some kind of actual experimentation. These experiments would probably > be basic feasibility and design studies but waiting for the > mathematics to catch up would not be smart unless you didn't want to > get involved in the effort until it became a single semester course. > Jim Bromer > > > On Wed, May 6, 2015 at 1:44 PM, Benjamin Kapp <[email protected]> wrote: >> So the Wright Brothers flew ~1900.. but "Working from at least as early as >> 1796, when he constructed a model helicopter,[18] until his death in 1857, >> Sir George Cayley is credited as the first person to identify the four >> aerodynamic forces of flight—weight, lift, drag, and thrust—and the >> relationships between them" >> [http://en.wikipedia.org/wiki/History_of_aerodynamics].. "Cayley is also >> credited as the first person to develop the modern fixed-wing aircraft >> concept" As such the fundamental principles of aerodynamics where known >> more than a hundred years earlier.. >> >> "In 1889, Charles Renard, a French aeronautical engineer, became the first >> person to reasonably predict the power needed for sustained flight." So the >> math had already been worked out for sustained flight, before the Wright >> Brothers flew.. >> >> Otto Lilienthal, following the work of Sir George Cayley, was the first >> person to become highly successful with glider flights. Lilienthal believed >> that thin, curved airfoils would produce high lift and low drag. So the >> understanding that one could produce lift from curved airfoils was known >> prior to the Wright brothers.. >> >> Octave Chanute's 1893 book, Progress in Flying Machines, outlined all of the >> known research conducted around the world up to that point.[24] Chanute's >> book provided a great service to those interested in aerodynamics and flying >> machines. >> >> "With the information contained in Chanute's book, the personal assistance >> of Chanute himself, and research carried out in their own wind tunnel, the >> Wright brothers gained enough knowledge of aerodynamics to fly the first >> powered aircraft on December 17, 1903"... So basically we already had the >> fundamentals of aerodynamics, and we knew how to create lift and what was >> needed for sustained flight. What was left for the Wright brothers to >> discover? It seems to me that given what the Wright brothers knew human >> flight was more of an engineering problem rather than a research problem. >> Does this make sense? >> >> >> >> On Wed, May 6, 2015 at 1:05 PM, Jim Bromer <[email protected]> wrote: >>> >>> Here is an interesting article about the Wrights that sounds similar to >>> what I have read before. >>> >>> http://wrightflyer.org/wp-content/uploads/2012/10/The-First-Aeronautical-Engineers-and-Test-Pilots.pdf >>> >>> The belief that the Wrights, who invented the wind tunnel, "just saw an >>> algorithm and then all they needed to do was to plug the right variables >>> into the mathematical equation" in order to create the first successful >>> powered airplane capable of carrying a person is an not an acceptable >>> hypothesis to explain how they went about creating the airplane. >>> >>> Jim Bromer >>> >>> On Wed, May 6, 2015 at 12:22 PM, Piaget Modeler >>> <[email protected]> wrote: >>>> >>>> The only way to test your hypothesis, like the Wright brothers did, is to >>>> build working prototypes and then refine them. >>>> >>>> No way 'round it. >>>> >>>> Just Do it. >>>> >>>> (Oh, that's Nike's slogan). >>>> >>>> ~PM >>>> >>>> ________________________________ >>>> Date: Wed, 6 May 2015 10:59:23 -0400 >>>> Subject: Re: [agi] Re: Starting to Define Algorithms that are More >>>> Powerfulthan Narrow AI >>>> From: [email protected] >>>> To: [email protected] >>>> >>>> >>>> I think that the Wright Brothers approach is appropriate for AI / >>>> Stronger AI / AGI as well. However, I also think it is obvious that there >>>> is >>>> ample evidence that digital programming has made numerous advances in AGI >>>> even though the successes seem to lack many human-like methods of thought. >>>> >>>> I have often wondered why the Wrights got so involved in control surfaces >>>> before they had a successful powered flight. Was it just common sense to >>>> realize that you needed to 'steer' the plane once it got off the ground, or >>>> was it just ego - since they 'knew' they would succeed they designed it for >>>> their flights of imagination. Or was it a common meme amongst aeronautical >>>> enthusiasts at the time? Or, did they realize, based on their experiments >>>> with gliders, that they would be able to extend their flights with >>>> mechanisms to control the attack of the plane in the air even though the >>>> plane would be heavier. (They decided to use wing warping to control the >>>> turns. NASA just tested a jet that is capable of changing the shape of its >>>> wings by the way.) Because this last possible reason might be related to >>>> the >>>> design-experiment-modify the design experiment method as it can be applied >>>> to AI / Stronger AI research. >>>> >>>> I want to find some evidence that my design principles would work to >>>> produce Stronger AI. So, by including some control mechanisms in my designs >>>> I might be able to stretch the distance it can get with the designs I have >>>> in mind. But, if I design for the some-day-in-the-future my control >>>> mechanisms would get so heavy that they could become a hindrance to any >>>> feasible programs that I might try now. But, by designing for a test I >>>> could >>>> run in the near future I might find some essential control features that >>>> could be lightweight and effective to stretch the capabilities of the >>>> program. >>>> >>>> But you have to have some feasible plan in mind to do that. If you want >>>> to try to do something with AGI right now your program (or device) has to >>>> be >>>> simple but effective - in some way. Even though you might not be able to >>>> convince other people based on primitive experiments, you have to be able >>>> to >>>> find some evidence that your ideas are going to do something different than >>>> most contemporary AI programs. >>>> AGI | Archives | Modify Your Subscription >>>> 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/303/26973278-698fd9ee Modify Your Subscription: https://www.listbox.com/member/?&; Powered by Listbox: http://www.listbox.com 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-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
