erich@eecs.cs.pdx.edu (Erich Boleyn) (09/13/90)
As strange as this posting my be, if you're interested, bear with me folks... brat@c3pe.C3.COM (John Whitten) writes: >I have been interested in Robotics and Artificial Intelligence for many years >and have followed discussions concerning Neural Networking and Parallelism >rather closely... Yes, I have done the same thing, with a lot of reading on the side included. I wonder if the following opinions presented by you are also very prevalent (with individual differences, of course ;-), since they seem to be similar to mine. > ...and that we assert ourselves to BE intelligent and not >merely a clever trick of nature... indeed, I believe it could be argued quite >sucsessfully that we are NOT intelligent... (end of tangent) That would be interesting, but I believe that you have a point on the whole, our society does seem bound up in its illusion of reasonability. >'raw hardware' is an important part, it is the underlying STRUCTURE that forms >the basic framework upon which an intelligence hangs... This belief has been presented by many people interested in AI... (especially the more progressive ones). >Experience is also a part. One of the most fundamental experiences is the >discovery that 'I' (an intelligent entity) am not alone and that 'you' exist. >I firmly believe that intelligence cannot form in isolation. Something that is >totally isolated has no use for intelligence, and that intelligence is a tool >used to successfully interact with others in 'our' universe. I would argue with this conclusion of experience. Intelligence is useful in any rapidly changing and complex environment if that has an effect on survival. I think that the "I" of separation comes from social interaction among others... higher intelligence is more *useful* when combined with social interaction and communication since we can then learn directly from others (sort of like long-range experience, or handing your experience in compressed form to someone else), and this increases survival dramatically (ignoring everything else ;-). We also live in an environment where there are other animals that exhibit recognizable (semi-)intelligent behavior, which adds to the usefulness of our intelligence. >The third major condition involves the ability to perceive and affect the >environment. Without this ability, the boundaries of where 'I' end and 'you' >begin cannot be established. This is necessary to produce any concept of the world at all, disregarding any boundaries... but like anything else, they would have to be experienced, I'll grant you that! > ...One of my first beliefs is that a successful >system will not be constructed that comprises of only one type of network. >And that in fact it takes the concerted effort of many different types of >networks working together as well as independently to process all the >different type of information that must sorted and evaluated. The "toss-everything-in-and-hope-it-works" philosophy, eh? ;-) Seriously, though, it very well may be that a single underlying structure (yet to be discovered, I think) would be a sufficient basis for all of the functions (thats what *we* use, and it works quite well), but that specialized components are simply more efficient... but isn't that always the case? >Another question concerns the initial 'seed' (or grain of sand) that kicks >the whole process off... I belive that every intelligent system begins life >with a sort of 'Random number' (for lack of a better word) which becomes the >essence of that entity... and that given identical hardware and environments, >different 'seed points' will develop into radically different personalities. For us you could say this 'seed' was the genetic code (plus the environment in the ova cell, of course). I don't see the point for introducing this 'Random number', though... a close look at developmental neurobiology would show you that even with an *identical* genetic code, there are likely differences in the development of two nervous systems, since the development relies heavily upon stimulation, i.e. the hardware isn't identical in *any* case, though it may be similar. When talking about hypothetical Artificial systems, though, they would probably require experince galore to develop as well, so you would not likely have identical systems even if they started out the same. >This leads to both 'Network Programming' and 'Network Topographies'. Massively ... >designs worked out and a system built,how would you program the damned thing?' >Neural Networks are funny beasts... seems you hafta TRAIN them to do what you ... >are closer to the optimal values... There seems to be a catch-22 forming >here... how can you initialize them with numbers that approximate performance >to problems yet to be encountered?This seems to be where the real intelligence >part comes in. >Fractals. This is where I am now. Dunno if this is a blind alley or what. From >what I understand of fractals, it seems the perfect programming language for >Neural Networks... the idea of describing a topography using fractal-based >means is intriguing to me... this would allow networks to 'pass on' the >benefits (and setbacks) of their experience(s) to future generations... this >would in turn tend to 'tweak' the values over time/generations to those that >would more closely approximate the actual values without hard-coding them into >the system. In this manner, fractals could be used to simulate the built-in >routines (or instincts) that we humans (and other similar life forms) posess. >The rest of course is up to the network. I have hit upon a similar idea, and believe that it is essentially what is going on in mammilian central nervous systems (a bit different in some cases). What I have studied of neurobiology has helped lead to this conclusion... When the central nervous system is developing it uses enzyme gradients to control which cells differentiate into which (sort of like having a central broadcaster where all the receivers determine what they do by the distance that they are away from the broadcasters)... this is not the whole story, of course, but it *is* an important part. Anyway, you have a bunch of cells that can amplify or respond to this signal, or even send out their own in response, and on the whole, this begins to take on a fractal-like branching activity, i.e. a *lot* of cells (acting according to local rules only) then differentiate according to the correct function that the cell they should be, by a set of complex feedback loops and such. By altering one of these rules slightly, it can change the overall pattern dramatically (as anyone who has experimented with doping metals or some such activity could attest). Even after the initial growth of the brain and spinal cord, later rules based on the same concept come into effect, managing pathways and neural connections by the level of stimulation involved, also *entirely* by local rules. Very fascinating stuff... I have been very curious about this material for a long time, but still have a couple of years of intense study before I'll feel ready to make a serious crack at it (more Mathematics and neurobiology, at least). I have come to think that the "seed" idea mentioned above (well, creating a fractal code, say, for the development of an AI system, and working on the analog of the "DNA" material... I still have to learn more before I will know enough to determine its usefulness) is more-or-less the way to produce a very robust and powerful learning and adaptive AI system. And though direct creation by enough people working on it might do the trick, I have a sneaking suspicion that such a system will be missing something that we just won't think of, and that the best method is to work on the generalized learning methods, and how to get them to come out on their own (since such a system would also be easier to alter for testing, related to the Genetic Algorithms studies). I suspect that trying for the construction of an AI directly (of sufficient capability) is nearly (if not there already) too complex for us to handle (although again, this is only a tentative feeling). I'm ready and willing to hear opinions, counters (even flames, frighteningly enough!), etc. Regards, Erich P.S.: One of the things that I really approve of is the multidisciplinary approach that has been taken up so widely in AI education and research... but I don't think that it has gone far enough. To tackle AI sufficiently we need to take on at least a little bit of any subject that crosses its path: neurobiology, mathematics, computer science (hardware and software, of course ;-), psychology (cleaned up a bit, I've had many doubts about modern psychology ;-), linguistics, (even god forbit, philosophy ;-), etc. just to name the major subjects involved, and work with it all to come up with solutions. I'm working along this path now, and besides kibitzing in AI discussions am still working on the education side of it, I'm *very* curious about others working with it (or having similar feelings about it) as well (perhaps still in that education process? ;-). P.P.S.: I'm also interested in working into a project (possibly of the "seed" type mentioned earlier) of massive scope, and realize that it would require large numbers of talented people to pull it off, and am curious about others similarly interested (I never have been good at setting my sights for easily realizeable goals, and besides, how else will I get into the Mad Genius club? *evil smilie* ;-). / Erich Stefan Boleyn Internet: erich@eecs.ee.pdx.edu \ >--={ Portland State University Honorary Graduate Student, }=--< \ Mathematics Department Mad Genius wanna-be / "Some say I'm a visionary, others just say I'm seeing things"