On Dec 26 Ben Goertzel said: >> One basic problem is what's known as "symbol grounding". this >> means that an Ai system can't handle semantics, language-based >> cognition or even advanced syntax if it doesn't understand the >> relationships between its linguistic tokens and patterns in the >> nonlinguistic world.
I guess I'm still having trouble with the concept of grounding. If I teach/encode a bot with 99% of the knowledge about hydrogen using facts and information available in books and on the web. It is now an idiot savant in that it knows all about hydrogen and nothing about anything else and it is not grounded. But if I then examine the knowledge learned about hydrogen for other mentioned topics like gases, elements, water, atoms, etc... And teach/encode 99% of of the knowledge on these topics to the bot. Then the bot is still an idiot savant but less so isn't it better grounded? A certain amount of grounding I think has occurred by providing knowledge of related concepts. If we repeat this process again, we may say the program is an idiot savant in chemistry. Each time we repeat the process are we not grounding the previous knowledge further because the bot can now reason and respond to questions not just about hydrogen, it now has an English representation of the relationship between hydrogen and other related concepts in the physical world.. If we were to teach someone such as Helen Keller with very limited sensory inputs would we not be attempting to do the same thing? Humans of course do not learn in this exhaustive manner. We get a shotgun bombardment of knowledge from all types of media on all manner of subjects. The things that interest us we pursue additional knowledge about. The more detailed our knowledge in any given area the greater we say our expertise is. Initially we will be better grounded than a bot, because as children we learn a little bit about a whole lot of things. So anything new we learn we attempt to tie into our semantic network. When I think. I think in English. Yes, at some level below my conscious awareness these English thoughts are electrochemically encoded, but consciously I reason and remember in my native tongue or I retrieve a sensory image, multimedia if you will. If someone tells me that "A kinipsa is terrible plorid". I attempt to determine what a kinipsa and a plorid are so that I may ground this concept and interconnect it correctly within my existing semantic network. If A bot is taught to pursue new knowledge and ground the unknown terms with it's existing semantic net by putting the goals "Find out what a plorid is" and "Find out what a kinipsa is" on it's list of short term goals then it will ask questions and seek to ground itself as a human would! I will agree that today's bots are not grounded because they are idiot savants and lack the broad based high level knowledge with which to ground any given fact or concept. But if I am correct in my thinking this is the same problem that Helen Keller's teacher was faced with in teaching Helen one concept at a time until she had enough simple information or knowledge to build more complex knowledge and concepts upon. When a child learns to speak he does not have a large dictionary to draw on to tell him that "mice" is the plural of "mouse". No rule will tell him that. He has to learn it. He will say mouses and someone will correct him. It gets added to his NLP database as an exception to the rule. A human has limited storage so a rule learned by generalizing from experience is a shortcut to learning and remembering all the plural forms for nouns. In a AGI we can give the intelligence certain learning advantages such as these dictionaries and lists of synonym sets which do not take that much storage in the computer. I also think that children do not deal with syntax. They have heard a statement similar to what they want to express and have this stored as a template in their minds. I think we cut and paste what we are trying to say into what we think is the correct template and then read it back to ourselves to see if it sounds like other things we have heard and seems to make sense. For people who have to learn a foreign language as an adult this is difficult because they tend to think in their first language and commingle the templates from their original and the new language. But because we do not parse what we here and read strictly by the laws of syntax we have little trouble understanding many of these ungrammatical utterances. -----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]] On Behalf Of [EMAIL PROTECTED] Sent: Thursday, December 26, 2002 11:03 PM To: [EMAIL PROTECTED] Subject: RE: [agi] Early Apps. On 26 Dec 2002 at 10:32, Gary Miller wrote: > On Dec. 26 Alan Grimes said: > > >> According to my rule of thumb, > >> "If it has a natural language database it is wrong", > > Alan I can see based on the current generation of bot technology why > one would feel this way. > > I can also see people having the view that biological systems learn > from scratch so that AI systems should be able to also. > > Neither of these arguments are particularly persuasive though based on > what I've developed to date. > > Do you have other arguments against a NLP knowledge based approach > that you could share with me. One basic problem is what's known as "symbol grounding". this means that an Ai system can't handle semantics, language-based cognition or even advanced syntax if it doesn't understand the relationships between its linguistic tokens and patterns in the nonlinguistic world. However, this problem doesn't totally rule out use of a linguistic DB. One could imagine supplying a system with a linguistic DB and having it learn groundings for the words and structures in the DB... Another problem is what I call the "knowledge richness" problem. The basic idea here is that if a system learns something through experience, it then is likely to know that something in an adaptable, adjustable way. Because it knows not only the thing itself, but a bunch of other things in the neighborhood of that thing, various useful components and superstructures of the thing, etc. it knows these other related things as side-effects of the learning process. On the other hand, if a system learns something through reading out of a DB, it doesn't have this surround of related things to draw on, so it will be far less able to adapt and build on that thing it's learned... My view is that a linguistic DB is not necessarily the kiss of death for an AGI system -- but I don't think you can build an AGI system that has a DB as its *primary source* of linguistic knowledge. If an AGI system uses a linguistic DB as one among many sources of linguistic information -- and the others are mostly experience-based -- then it may still work, and the linguistic DB may potentially accelerate aspects of its learning.. Ben G ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED] ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
