Stephen, Sounds like some interesting stuff you are working on with Cyc. I will preface all my comments here with the disclaimer that I have no AI training and have just begun to write some AI programs for my own amusement and education. I had checked out Cyc and their Cycl language, and frankly, was very disappointed that after a decade of work that I couldn't enter a simple sentence into Cyc to teach it something. So any efforts on your part to enable a more practical way of inputting information\concepts into Cyc should greatly expand its viability\usability IMO.
Regarding sensory inputs and their pertinence to NLP and AGI, it occured to me while reading the dialog here that language for humans is a somehwat visual experience. For instance, if you write to me that you saw a video of president Bush shooting Saddam Hussein(sorry for the poor example), I would immediately get an image in my mind of president Bush shooting Hussein. This would occur without me trying, without me ever actually seeing Bush shoot Hussein, or even knowing the setting. Because my mind knows what Bush looks like, and what Hussein looks like, and what a gun looks like, I am able to construct a somewhat realistic, if somewhat inaccurate picture of this event. It is thru this picture that I can understand what happened and what you are saying. This led me to the feeling that for true NLP to occur, a computer would need to have visual references to draw upon that it has categorized and learned from, and the ability to "imagine" or construct images composed from these learned experiences. In a related item somewhat tangental, I once sat down and calculated, based upon the storage of my Digital Video recorder, how much storage would be required to capture the entire visual\audio record of a person from birth to death. The number was obviously high, but not that high. In the next few years we can reasonable expect that the amount of storage required will be available in a small form factor. This essentially means that ( i am not condoning this!) when a child is born, a video camera\microphone could be attached to a child and kept there for 80 or so years and capture every audio\visual experience the child had. So what follows is given this type of information, could it be fed into a computer that could "learn" from it about people, shapes, colors, gravity, etc, just as the child did. And thru this experience, derive a better understanding of the words used in speech. I know this will be a long and arduous path, but I can't see why it will not be possible. I also considered what you said about a blind person. Helen Keller could communicate without being able to see or hear. So it would seem to represent that these are not necessary for language processing. But from another point of view, we cannot say for certain that she was not forming some type of image(I use this term loosely as her way of understanding something would be beyond us and may not correlate directly to an "image") when she received a sentence. Her "image" may have been instead a "feeling" image instead of visual image. She new what a car felt like, so when we say car, she might have thought of the feeling of a car. I apologize for my poor understanding of these topics and hope my comments aren't entirely too elementary... On last note on Cyc, i really wish you guys would be working on beneficial interests that were non-defense related. I have deep fears about the direction of our country and military. The last thing we need is even greater military superiority right now. Peace, Kevin ----- Original Message ----- From: "Stephen Reed" <[EMAIL PROTECTED]> To: <[EMAIL PROTECTED]> Sent: Monday, November 18, 2002 2:59 PM Subject: Re: [agi] Language translation by computer > Alan, > > Thanks for speaking frankly. Of course Cyc is very far from _perfectly_ > implementing human-like abstract logic, but I understand your viewpoint of > the Cyc project's emphasis on hand-entered symbolic assertions, as opposed > to knowledge directly sensed, and perceived by the system. On a tip from > Ben, I have been developing a hierarchical control system for Cyc that I > have dubbed "Cognitive Cyc", that could bottom out in robot style > actuators and sensors, but won't for now. Instead I will target English > dialog (chat) and structured knowledge sources as the Real World, allowing > Cognitive Cyc to interact with and learn from a symbolic-friendly real > world. The immediate goal is automatically mapping/adding terms from > non-Cyc ontologies into Cyc. > > My view is that your approach is bottom up, whereas I hope that Cognitive > Cyc can achieve AGI from the top down. I do not believe that rich sensory > input and actuation are required, although it is a worthy approach other > such as yourself to pursue. Rather I draw inspiration from > deaf-blind-from-birth humans who communicate via email and construct web > sites indistinguishable from the rest of us having full sight and hearing. > Many deaf-blind-from-birth humans acquire the great majority of their > abstract knowledge completely by symbolic means. I believe that Cognitive > Cyc can do the same. > > I am working hard to flesh out the hierarchical control structure (a.k.a > NIST/RCS, a.k.a. Albus Reference Model Architecture) application that > implements the first iteration of Cognitive Cyc. If I have something > working by mid December then I can present it to Ron Brachman and his > Darpa deputies when they survey Cycorp ideas for Darpa programs in an > expected one-day visit. > > Good luck with your computer acquisition. > -Steve > > On Mon, 18 Nov 2002, Alan Grimes wrote: > > > Stephen Reed wrote: > > > >From the viewpoint of a developer working to overcome the perceived > > > deficiencies of Cyc with regard to AGI, I ask what problems do you see > > > with Cyc technology as you understand it? I promise not to debate nor > > > to attempt to change your mind. > > > > Cyc is really strange. > > > > It takes the one *optional* component of human resaoning (the high level > > abstract logic) and implements it _perfectly_ but ignores the stuff that > > really needs to be there. (the stuff that looks for signals in the input > > stream and the stuf that dictates top-level goals). > > > > At this stage in the game it is far more urgent to write a sim-redneck > > who "don't think so good". But will be able to learn to dance the polka > > just like anyone else. Once we reach that point, the abstract reasoning > > becomes relevant. The system might be powerful enough to learn how to > > reason just as humans do but it would probably be more efficient to > > design reasoning software that will accelerate the process. At that > > point you would be firmly in SI teritory... > > > > So if I ever get my new computer functional (the post-office lost a $140 > > ram module; can't afford another), I'll be trying to implement something > > that will try to use a GUI program just as a human would... (by looking > > at the color patterns and moving a mouse and typing.) > > > > > > -- > =========================================================== > Stephen L. Reed phone: 512.342.4036 > Cycorp, Suite 100 fax: 512.342.4040 > 3721 Executive Center Drive email: [EMAIL PROTECTED] > Austin, TX 78731 web: http://www.cyc.com > download OpenCyc at http://www.opencyc.org > =========================================================== > > ------- > To unsubscribe, change your address, or temporarily deactivate your subscription, > please go to http://v2.listbox.com/member/ > ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/
