Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!wuarchive!uunet!mcsun!ukc!edcastle!helium!aifh!kk From: kk@aifh.ed.ac.uk (Kathleen King) Newsgroups: comp.society.development Subject: Re: Expert Systems in Developing Countries Message-ID: <1991Jun10.114353.17103@aifh.ed.ac.uk> Date: 10 Jun 91 11:43:53 GMT References: <1991Jun10.110537.16188@aifh.ed.ac.uk> Sender: news@aifh.ed.ac.uk (Network News Administrator) Reply-To: kk@aifh.ed.ac.uk (Kathleen King) Organization: Dept AI, Edinburgh University, Scotland Lines: 57 In article <1991Jun10.110537.16188@aifh.ed.ac.uk> e.reiter@ed.ac.uk writes: >[I posted this to comp.ai a few months ago, and got very little in the way of >responses. I thought I'd try again in this group -- Ehud] You might also try the AI in Medicine mailing list >I was recently looking over a Ph.D thesis(*) that discussed (among other things) >an evaluation of a medical expert system designed for rural Third-World areas. >One claim the author made was that the expert system she looked at was so >brittle that it could only be used by people with substantial medical intuition >(so they could tell when the system's diagnosis was wildly incorrect) and >also with substantial medical self-confidence (so they would be able to >overrule the computer when they thought it was wrong). This meant, in >particular, that the system could *not* be sent out to paramedics and nurses >in remote villages, as some of these people were not very well trained. I think one reason the system Dayo tried out was not a huge success was that it was not tailored specifically for the area it was trialed in, but rather was based on a series of diagnostic flowcharts (Ben Essex' `Diagnostic Pathways in Clinical Medicine' I think?) which was designed for general use. It might have been more successful if the contents of the knowledge base corresponded with the training that the health workers had already had (so that it knows what they know but is maybe better at remembering it all). The flowchart/decision tree method, simple and often effective though it is, is very inflexible. Questions must be answered in the prescribed order, explanation cannot be so easily built in as with e.g. goal-driven systems and the `pathways' are always the same. I think that flowcharts are much more useful in applications like therapeutic protocols where there IS a standard ordering of questions (history taking is another possibility). The issue of whether users have the knowledge/confidence to recognise and/or overrule the machine when they think it's wrong is partially a sociological one, is it not? More to do with the ROLE of the machine in the clinic and the appearance of its interface than with the particular inference mechanisms or representations implemented. This implementation was very much in the traditional role of `consultant' expert system rather than `co-operative'. Much more attention is now being paid to the role of the system in the diagnostic consultation process and there's a general move towards the system being less of a guru and more of an assistant (who happens to know a lot). This applies just as much in a GP's surgery in Walthamstow as it does in a rural clinic in Botswana. >I have always thought that it would be really nice if we could >use expert system technology to provide badly-needed expertise (medical and >otherwise) in poor Third-World areas, Yes yes! >but perhaps this dream is fundamentally unachievable. Well I still believe in it.... -- Kathleen King, Dept of AI, JANET: kk@uk.ac.ed.aipna University of Edinburgh, ARPA:kk%uk.ac.ed.aipna@nsfnet-relay.ac.uk 80 South Bridge, Edinburgh UUCP: ...!ukc!aipna.ed.ac.uk!kk