I was probably too confrontational with my last message. I know that there are no solid reasons to believe that some kind of embodiment is absolutely necessary for the advancement of agi. However, there is nothing wrong with making that effort if someone wants to, just as there is nothing wrong with using robotics to study agi. And, while it may seem that I am contradicting myself, I do believe that there are some necessary methods for agi that can be discovered by working with virtual embodiment or with a virtual world. It may sound like a conceit, but I honestly think that I have already considered the key elements to agi that a researcher might find by pursuing this kind of course. At least I have considered them in a general way. An IO data environment has to have some characteristics that would make learning possible and some that could enhance the progression of learning. One of those is the use of different multiple references to reference objects that could be used as construction blocks in further learning. These multiple references have to, at least in some cases, vary in nature. To give an example, if one reference to a data object was simple and easy to detect against the background of the IO data environment, then it could be used to enhance the subject matter that would be more difficult to 'comprehend' from the use of a more complicated reference. I am not saying that this insight is a big deal, I am saying that I derived this idea from considering the advantage of using multiple IO data types, and it is one of the strongest reasons for using different IO types with an agi program. But it can be constructed within a single IO type as long as the type is complex enough and contains the variations that would be necessary to produce this kind of effect and the program was able to make the connection. So researchers pursuing different methods can all contribute to the knowledge that would be required to get these kinds of programs working. But I would be very surprised if it turned out that embodiment was absolutely necessary as had been suggested in this thread. What is necessary is that certain kinds of variations of reference within the IO data environment are possible and the program is able to attend to them and make some of the desired associations that it would need to make.
Jim Bromer On Mon, Aug 11, 2008 at 8:53 AM, Jim Bromer <[EMAIL PROTECTED]> wrote: > On Sun, Aug 10, 2008 at 6:43 PM, Ben Goertzel <[EMAIL PROTECTED]> wrote: >> You'll be convinced in time, for instance if an OpenCogPrime instance starts >> refuting your arguments on this mailing list! > > That is not very likely to occur if the 'magic' of agi is absolutely > dependent on embodiment, and the only embodiment that your program is > going to have is virtual. > > A processor only interacts with the one kind of data, a computer > program does not have some other special kind of transcendent > processing that is able to interact with the real or virtually > simulated world. How could the processing of data that comes from a > fictional database of relations, or even the data that comes from real > world sensors, provide some kind of 'magic' that would suddenly make > higher general intelligence work when it has not worked very well > before? > > If you can't get an agi program to work based on carefully constructed > IO data environment, why do you think you could get it to work with an > advanced game-like data environment? This fantasy flies in the face > of most scientific development which sees some progression from > simplistic to more sophisticated. Game environments are simplified to > make the game as interesting to the user as it can be given the > demands of efficiency, they are not designed specifically to advance > agi. > > I have no doubt that virtual embodiment and sensors will make agi much > more interesting and powerful once some poorly understood problems of > artificial intelligence are solved. And I feel that the IO data > environment has to have certain features and complexities (in the > general sense) in order to facilitate learning and these can be > derived from experience with virtual worlds, but the idea that > embodiment (with a real or virtual world) is going to be the > independent key to unlocking less well understood problems is not a > well reasoned theory. > > There is one strong reason for embodiment and that is that a data > relation may sometimes be simple in one representation of the world > (IO media) than another. But this desirable feature can be easily > encoded in a single data environment (like a text based IO data > environment.) > > Jim Bromer > ------------------------------------------- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244&id_secret=108809214-a0d121 Powered by Listbox: http://www.listbox.com
