Xref: utzoo comp.society.futures:486 comp.ai:1586 Path: utzoo!mnetor!uunet!mcvax!ukc!strath-cs!glasgow!gilbert From: gilbert@cs.glasgow.ac.uk (Gilbert Cockton) Newsgroups: comp.society.futures,comp.ai Subject: Re: The future of AI [was Re: Time Magazine -- Computers of the Future] Message-ID: <1053@crete.cs.glasgow.ac.uk> Date: 28 Apr 88 10:53:57 GMT References: <8803270154.AA08607@bu-cs.bu.edu> <962@daisy.UUCP> <5789@swan.ulowell.edu> <978@crete.cs.glasgow.ac.uk> <445@novavax.UUCP> Reply-To: gilbert@cs.glasgow.ac.uk (Gilbert Cockton) Organization: Comp Sci, Glasgow Univ, Scotland Lines: 81 In article <445@novavax.UUCP> maddoxt@novavax.UUCP (Thomas Maddox) writes: >By comparison, sociologists produce a great deal of nonsense, and indeed the >social "sciences" in toto are afflicted by conceptual confusion at every >level. Ideologues, special interest groups, purveyors of outworn >dogma (Marxists, Freudians, et alia) continue to plague the social >sciences in a way that would be almost unimaginable in the sciences, >even in a field as slippery, ill-defined, and protean as AI. There are more of them :-) But if you looked at the work of U.K. sociologists like Townsend and Halsey on Age, Poverty, Health and social mobility, you might find something less concerned with theory and more with rigorous investigation. I find the conflict in the humanities and behvioural "sciences" far more healthy than the uncritical following of fashions of paradigms in science. Whilst the former areas encourage an understanding of methodology and epistemology, the sciences assume their core methods are correct and get on with it. A lot boils down to personality (Liam Hudson, Contrary Imaginations). The reason that ideology and methodological pluralism would be unimaginable in the sciences may have something to do with the nature (and please, not the LACK) of the scientific imagination compared to the humanist imagination. Note that materialism, determinism, statistical inference and positivism are no less outworn dogmas and ideologies than are Marxism, Freudianism, etc. My experience is that someone from a humanist critical tradition will have a better understanding of the assumptions behind methodologies than will scientists and even more so, engineers. Out of such understandings came the rejection of first Medieval Catholicism, then Seventeeth Century materialism, Twentieth Century Behaviourism and Systems Theory, and now the "pure" AI position. Assumptions behind AI are similar to many which have been around since the warm humility of Renaissance Humanism cooled into the mechanical fascination of the Baroque. >So talk about "philistine technical vacuums" if you wish, but >remember that by and large people know which emperor has no clothes. So who is it who is deciding strategy for most Western social programmes? Clothes or no clothes, social administrators have an empire which extends beyond academia and many of them draw on sociological concepts and results in their work. It is in their complete ignorance of socialisation that AI workers fall down in their study of machine learning. Most human learning always takes place in a social context, with only the private interests of marginal adolescents and adults taking place in isolation - but here they draw on problem solving capabilities which were nutured in a social context. The starkest examples of the nature and role of primary socialisation come from those few unfortunate children who had been isolated from birth. They are savage animals. If parents had to interact with their children in FOPC or connectionist inputs, the same would be true, until the children were taken into care. >Also, if you want to say "one dead end after another," you might adduce actual >dead ends pursued by AI research and contrast them with non-dead ends. DEAD ENDS Computational Lingusitics, continuous speech understanding, intelligent vision, reliable expert systems which do not require endless maintenance, human problem solving, the physical symbol system hypothesis, knowledge representation formalisms using computable models. Largely areas where some other paradigm within another discipline can make progress as the lead weight of computability is not suffocating research. Generally due to knowledge representation problems - even the Novel has problems here :-) If you can't write it in a text-book (e.g. clinical diagnosis, teaching techniques, advocacy), you'll never get it on a machine - impossible in superset (NL) => inpossible in subset (FOPC, computationally denotable/constructable). A problem in AI is trying to solve other people's problems, where those other people know more about the problem than you ever will - they live it day in day out. NON-DEAD ENDS Much work done under the name of AI is good - low-to-medium level vision, restricted natural language, knowledge-based programming formalisms, theorem-proving and highly-constrained technical planning problems. Indeed, most technical knowledge, being artificial and symbolic from the outset, is an obvious candidate for AI modelling and there is nothing in the humanist tradition which would doubt the viability of this work. Here knowledge representation is easy, because the domain will generally be so boring (but economically/environmentally/security critical) that no-one wants to argue about it. Much technical expertise executed by humans is best suited to machines. In HCI research, sensible work on intelligent (=supportive) user interfaces is getting somewhere, but then coming up with a computer model of a computer system is hardly a major challenge in knowledge representation techniques. Coming up with a computer model of a user is also possible, as long as we don't try to model anything controversial, but stick to observable behaviour and user-negotiated input. The main objection to AI is when it claims to approach our humanity. It cannot.