Wayne@OZ.AI.MIT.EDU (Wayne McGuire) (09/02/87)
> Date: 24 August 1987, 23:09:52 EDT > From: john Sowa <SOWA@ibm.com> > Subject: Wittgenstein and natural kinds > > Every schema in a cluster represents one valid use of the concept > type. The meaning is determined not by any definition, but by the > collection of all the permissible uses, which can grow and change with > time. > > Does that solve the problem? Maybe, but we still need criteria > for determining what kinds of uses can legitimately be added to a > cluster. Could I say "To add something means to eat it with garlic > and onions"? What are the criteria for accepting or rejecting a > proposed extension to a concept's meaning? Under the assumption that language (and all human semiotic systems), and the concepts they label, are in great part a social contract, a collection of arbitrary conventions momentarily accepted in a ceaseless process of interaction by a particular group of people in a particular space and time and culture, perhaps what is permissible is anything that any human group, through _actual usage_, indicates they find useful as a tool of communication. Human societies are much like Humpty Dumpty in _Alice in Wonderland_: "When _I_ use a word," Humpty Dumpty said in rather a scornful tone, "it means just what I choose it to mean--neither more nor less." "The question is," said Alice, "whether you _can_ make words mean so many different things." "The question is," said Humpty Dumpty, "which is to be master-- that's all." An ideal program with general intelligence would closely monitor the actual language usage and semiotic behavior of its target domain, and assimilate as new schemata in its world model those new words, signs, and concepts which reach a user-settable level of usage or which are assigned by social fiat roles as fixed conventions (fixed for the time being, of course, for the life of this particular cultural phase). One can easily imagine wanting one's world model to be more comprehensive, however, and to include highly idiosyncratic language uses that are not social conventions. Integrating the detailed mental models and private languages of all the world's leading imaginative writers, from Homer to Norman Mailer, could be valuable for the purposes of some people and with the object of constructing a humanist superintelligence. A speculation: how many new schemata enter the set of all human discourse in a year? Probably professional dictionary-makers and terminology-compilers have given some thought to this question. I recently came across an apt passage in Wittgenstein's _Tractatus Logico-Philosophicus_: I can place over the world a unified descriptive net through which I bring everything to a unitary form. According to the kind of net that I choose there results a kind of world description. If I take various nets then I produce various world descriptions. Wittgenstein then chooses mechanics as a sample world description net, and notes: "Mechanics determines one form or description of the world by saying that all propositions used in the description of the world must be obtained in a given way from a given set of propositions--the axioms of mechanics." But there are at least as many distinct world description nets as there are persons, living and dead, and probably many more, taking into account the changing mental model of an individual over time, and the sets of unique mental models of an individual in different roles and social domains. There is much redundancy in all these models, but there is also a subset of each net that is special and which includes schemata that are the only ones of their kind. Clearly artificial intelligence researchers need to pay much more attention to pragmatics and sociolinguistics: the notion that intelligence is reducible to a set of universal principles or context-free grammar rules is misguided. Ultimately AI might seek to model and integrate, in a Supreme World Net, as many Wittgensteinian world description nets as possible, to map their literal and metaphoric relations, and to track their evolution and devolution in real time. (And while we are at it, it would be nice to build a working perpetual motion machine.) Wayne McGuire