To: cvkellison:::yahoo.co.uk Subject: Re: gee Extended Abstract for UMSL Undergraduate Philosophy Club Talk Suggested Title: AI, The Matrix, and The Philosophy of Computing R. P. Loui Assoc. Prof of Computer Science & Program on Legal Studies & Center for Semantic Control and Optimization Washington University in St. Louis R.P. Loui, Ph.D. in Computer Science and Philosophy, 1988, Univ. of Rochester, has been published in Analysis, Theory and Decision, Journal of Philosophy, Computational Intelligence, Cognitive Science, Artificial Intelligence, AI and Law, AI Magazine, Journal of Symbolic Logic, Computing Surveys, Minds and Machines, Communication and Cognition, Defeasible Deontic Logic, Philosophy and AI, The Robot's Dilemma, and the MIT Encyclopedia of Cognitive Science. He edited the Society for Exact Philosophy Proceedings of 1988, (Knowledge Representation and Defeasible Reasoning, Kluwer, 1990). He is hosting the 2001 Intl Conference on AI and Law in St. Louis, and finishing two books: Argument Games and Negotiation Games, and Easier CGI: A Better Language for Web Programming. He just discovered that his web page on Camille Claudel & Auguste Rodin is used in a senior art history class at Michigan State. We will use The Matrix to launch into a discussion of some claims about AI, the scope of computing, and the interpretation of rules. Most of these claims are furiously rejected by the mainstream, and have met with ferocious efforts of the scholarly masses to protect their canons. What is the Matrix? The Matrix is right: + The expressiveness of autonomous agents depends on the complexity of the virtual world they inhabit, and the virtual worlds are getting more interesting (no good AGENTS without a great MATRIX); and this is partly why AI has turned to protocols for intelligent social organization, toward social psychology instead of the psychology of the individual. + Computational agents are not omnipresent, omniscient, infallible, or instantaneous actors (there's a reason why the bullets keep missing, Mr. Ebert); and this is why it is frustrating that thirty years after Herbert Simon's Nobel Prize-winning ideas on bounded rationality, economists still don't understand process and procedural rationality. + No system of rules will ever be better than you can be (but you have to be capable of perfect rule-following, then come to understand the rationales that permit deviation, and the processes for rule revision); and this is why Oxford's Herbert Hart was right about defeasible reasoning and why Cambridge logicians have gone wrong for fifty years (since Wittgenstein's departure). + It is the life-giving force of womanhood that avails man of victory over machine (the Matrix was a love story, not just a Messianic science fiction kung fu action-rebellion film); and this is why feminist philosophy must be careful in its dissuasion of man's idolatry toward woman. Unpopular Claim 1. Computer "Science" is not science at all. Its method is exploratory design. "Design method" is unlike scientific method (empiricism based on induction) and unlike applied mathematical method (analysis based on deduction). Good design is novel in its tradeoffs, construction, function, paradigm, vernacular, or expression, and it must be evaluated in a social context, usually with an historical method. Exploratory design of evolvable phenomena (AI explores designs of intelligence) has been AI's distinction among engineering disciplines. But it will cease to be distinct with the rise of economic design, institutional design, biological design, etc. (Journal of Experimental and Theoretical AI invited, then suggested withdrawal of this paper and a related paper was rejected by Philosophy of Science.) Unpopular Claim 2. Logic has nothing to do with reasoning. Logic is a system of representation, a shorthand notation. To commit to a set of symbols (as a description, for example), is to commit to the enlarged set of symbols that is the original set's "consequents." Keynes told Russell and Whitehead that there is a process to reasoning that is not captured by deduction. Wittgenstein told Turing that there is nothing wrong with inconsistency, so long as the logical game says what to do about inconsistency. Both were right. The logic of argument is "ampliative," where logical entailment is nonampliative; the logic of argument is "nonmonotonic," while logical entailment is monotonic; the logic of argument is a social process, while logical entailment is static and antisocial. (Cognitive Science and Philosophy of Science rejected this paper which was read at a Wash U philosophy colloquium, and Quine rejected a follow-up letter on this subject; but this work has led to a dozen doctoral theses in the past decade, esp. in the Netherlands and Argentina, and the manifesto was eventually published ten years late in Canada.) Unpopular Claim 3. The mathematics of microconomics is one hundred years out of date. It is based on two major mistakes (and one minor one): (1) that value is utility (i.e., extensively measurable like heat or distance), especially utility that is process-independent and unrevisable by deliberation; (2) that game theory properly models rational social interaction, especially game theory that seeks "solution" at "equilibrium," with alternatives and payoffs that are unrevisable by computation, and with no serious modelling of language and social norms. (The third minor mistake is Bayesian subjective probability.) (1) has already been largely addressed by heuristics in AI. (2) is being addressed by models of interaction as an interplay of joint problem solving and self-interest. We can model negotiation with more expressive protocols, using off-the-shelf concepts from AI: planning, plan-recognition, speech acts, heuristic evaluation. But the result would not be recognizable to economists until they realize that their work includes the design of economic interactions, not just the description and "explanation" of them. (Negotiation Journal invited, then suggested withdrawal of this paper, some of which was reported at Carnegie-Mellon in a School of Computing colloquium and a Wash U economics talk; and the National Science Foundation rejected a research proposal on these ideas.) Unpopular Claim 4. Most of the time, adherence to the letter of the law is actually the promulgation of a misrepresentation of the law (i.e., qua the intent of the law). The intended process for legal intepretation is an argumentative process, not a deductive one. Even when there are no "defeasible" rules, there are argument and counterargument over the applicability of "open-textured" terms, analogical argument, case-based argument, arguments over principles and rationales, and meta-arguments over fairness, process, and suitability as precedent. Case-based projection is a technically superior form of system specification, (compared) to rule-based direction. It is also the intended logicolinguistic manner for using legislation. Any bureaucrat, minister, or functionary, who pretends otherwise, is violating the law. (Ratio Juris and Law and Philosophy rejected versions of this paper, which was read at a U. Rochester computer science and philosophy colloquium, Carnegie-Mellon philosophy colloquium, and a U. Georgia philosophy club colloquium.) Unpopular Claim 5. Social game-playing (which includes most norm-governed social phenomena) is a form of computation. Tradition holds that something is computation only to the extent that it can be modeled by "recursive functions", grammars, or Turing machines. That is, a computation is the following of an algorithm. The naive view further insists that computation be deterministic, independent of situation, nonelective, serial, purposive, autonomous, and finite (there was a time when it was assumed to be numerical, electromechanical, and expensive, too). But most of what computers do is not algorithmic, and satisfies none of these naive senses of computation. Computers are programmed to interact, not just to unpack; in fact, object-oriented programming is more like gossip regulation than pyramid building. A better test for computation is the existence of a program, not the existence of an algorithm. Computation is a following of the program, an intentional rule-following instructed (animated?) by the (symbolic?) expression of the rules. Social programming (protocol design), such as specifying a hearing or an election, constraining the trajectory of a market, or agreeing on the implicatures of speech acts, gives rise to social computations. It is as much a computation as abacus calculations are computations. If you want to say that an operating system's progress is computation, or that the distributed search algorithm computes the decision, or that the modems compute their transmission speed, then you must also say that chess players compute their outcomes, and, hence, that electorates compute their representatives and courts compute their decisions. Distributed, situated, nondeterministic, unbounded, non-progressive, interactive and multi-agent computing is the new computing. AI isn't scary because it is coming soon; AI is frightening because it is already here in the form of bureaucracy and social constraint. (Journal of Philosophy rejected this paper, which was read at the last meeting of the Soc. for Exact Philosophy.) Unpopular Claim 6. Art that is worth calling "art" is deeply invested personal inspired expression, as opposed to the identification of a name with a style through repetition and marketing. It is this first kind of art which stands in opposition with what Lewis Mumford called "technics" (he was appalled by automation and bureaucratization, well before computing arrived in force). What art is man capable of? What are the major themes of man's deeply invested personal inspired expression? What are the three things that Trinity is supposed to represent to Neo? What is it that man produces when he is not under the influence of woman's trinity? Hmm... (I actually have no philosophical stake in this discussion, but I thought I'd open the door.)