Well, I agree that the current field of AI is way too broadly defined. It's a heck of a grab-bag.
But I think there is more meaning to "AI in general" than "A little AGI plus a lot of computer science" or "What AI researchers do." I'd break up the AI field into 4 categories as follows: 1) AGI 2) Domain-specialized learning software. These are software programs that carry out learning or goal-oriented self-organization in some particular domain 3) Emulations of human intelligent behavior. These are software programs that emulate activities that for humans involve significant learning or complex cognition; but the programs themselves don't do any learning, they're just ordinary, rigidly-behaving computer programs 4) Emulations of human perception and action in the absence of cognition, or software carrying out perception or action in the absence of cognition -- e.g. computer vision, robotics, etc. "Narrow AI" consists of categories 2 and 3, which are quite different animals, however. Note that a program gets into categories 1 or 2 because its internal activities are viewed as intelligent. A program gets in category 3 because its external behavior is viewed as intelligent. In category 2 I'd put: a program that learns how to drive a car, or a program that predicts the stock market by building adaptive models.... These programs have some intelligence but it's very narrowly focused. There are indeed a lot of commonalities among various narrow AI programs in this category. There is a real science of narrowly-specialized learning; it's not all domain-dependent tricks. Statistical learning theory is part of this science, but not all of it. On the other hand, programs in category 3 don't have many commonalities amongst each other. The problem is that there really are few commonalities among different programs that do things that, for humans, require intelligence. It seems very arbitrary which programs in category 3 are categorized as AI and which are not. Why is a calculator not AI? It carries out activities that require a lot of intelligence in humans? Why are computer algebra systems like Maple less "AI-ish" than GraphPlan, which is arguably less sophisticated, less "intelligent"...? Category 3 is where AI basically merges freely into "advanced algorithmics." Category 4, on the other hand, really doesn't have that much to do with "intelligence." It's advanced algorithmics, modeling or emulating interesting biological behaviors. To the extent that these biological behaviors relate to cognition, it may be related to AI; but so are a lot of other things related to AI... I think there's a decent argument for the existence of an AI discipline inclusive of my categories 1 and 2: AGI plus domain-specialized learning software. On the other hand, throwing in bits and pieces of 3 and 4 makes things very loosely-defined: one could really throw in about half of the computer science curriculum, and which items in 3 and 4 are considered AI versus non-AI is basically a matter of historical accident... -- Ben G > -----Original Message----- > From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On > Behalf Of Pei Wang > Sent: Wednesday, January 08, 2003 6:59 PM > To: [EMAIL PROTECTED] > Subject: Re: [agi] Q: Who coined "AGI"? > > > > ----- Original Message ----- > From: "Ben Goertzel" <[EMAIL PROTECTED]> > To: <[EMAIL PROTECTED]> > Sent: Wednesday, January 08, 2003 6:38 PM > Subject: RE: [agi] Q: Who coined "AGI"? > > > I guess most AI researchers consider AI to be inclusive of AGI and ASI. > > That's Ok with me ... ASI is interesting too, though quite > different, and > > currently overemphasized... > > I'd rather use more concrete (and more informative) terms for the "ASIs", > such as search, reason, vision, planning, pattern recognition, > and so on --- > if there is no common factor identified among them, to call them "AI" > collectively doesn't add much information. Actually this is what has > happened in the history of AI, that is, whenever a subfield becomes mature > enough, most people stop using the label "AI" in it. It will > continue to be > the case until a "general factor" is identified. It is only at that time > then AI can be claimed to be a branch of science for its own sake, not a > fancy label to be attached to a group of things that have little > in common, > except being "that which AI researchers do." > > Pei > > > My (least) favorite definition of AI is in Luger's AI text. He > defines AI > > as "that which AI researchers do." > > > > Hmmm... > > > > -- 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] > ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
