On Wednesday, June 09, 1999 5:10 PM, Philidor11 [SMTP:Philidor11@aol.com] wrote: > > Some time ago I read that physicians in emergency rooms now have > available to them a computer program which is able to develop a > prognosis for patients which is at least as good as the one developed > by the doctors themselves. The use of expert systems in medicine is well known for litterally decades. In fact, the Mycin (sp?) system was one of the very first such systems ever. Together with geology (locating likely locations of minerals and oil) medicin was the field in which expert systems started. > Apparently it is possible to develop invariable rules which have predictive value. It definitely is. It's not an easy process, but in principle the knowledge that human experts use (whether they explicitly work by rules, work by rules of thumb or they don't even know exactly themselves how they do it) can be laid down in an expert system that then can ask the user questions, builds hypotheses with probabilities and asks just those further questions to zoom in onto the solution. If the so called knowledge engineering has been done and worked out correctly in the sense that it really captures the "knowledge" of the human expert(s) conculted during the process, such an expert system should be able to replace, to a very large extent or even entirely, the expert(s) themselves. In fact, for a course in expert systems I had to build a small expert system myself for a practical test. As the subject I chose dinosaur identification. I originally wanted to use myself as "the expert", but after asking around (also here on the list) I found out the Black Hills Institute was in fact working on such a system, so that groups in the field could do some preliminary identification of found remains without having to have one of the experts present on site. I was given permission to use some of their information for my expert system project and even though it was a very simple system, I personally think that DiKnowSaur (as I called it ;-)) did a reasonable job of recognising dinosaur remains down to a certain level. In the publication section on my website (http://www.jarno.demon.nl/publish.htm) you can find a reference to the report on DiKnowSaur. I never did convert it to HTML so it's not available to be read there. Besides, it is in Dutch, and my "real" knowledge of dinosaurs has increased tremendously these last four years because of all the books I bought and read (before that I mainly some of the well know popular books, not much on actual physical details). I could however try to post the example sessions for DiKnowSaur from that document (the system itself is in English) and/or send interested people a copy of the system (MS-DOS) as soon as I locate it among my things at home. It could be an example of possible use of expert systems in dinosaur paleontology. I'm sure that it would be possible (if not difficult and a lot of work!) to create such a system to identify even scrappy remains down to all the couple of hundred dinosaur species now known instead of the one or two dozen genera as DiKnowSaur does. It just needs a lot more rules than the hunderd or so (if I remember correctly) that DiKnowSaur contains. > To me, paleontology, like chess, has an element of art to it no > matter what the computers do. This art, or arbitrariness, is part > of my interest. This is of course true, but to me the systematics of anatomical details and the rules for naming genera and species, the fact that cladograms (which are trees, one of the basic computer science datastructures, used for all kinds of things) are constructed using matrices of data and computer programs and of course that the Tree of Life itself is by its nature a tree are (I suspect...) some of the main reasons why I, as a computer scientist, am interested in paleontology. Or rather, due to my interested I became a computer scientist and a dinosaur nut, and I find it fascinating to find out that the reasons for this might actually in part be the same. Met vriendelijke groeten, Jarno Peschier
Attachment:
WINMAIL.DAT
Description: Binary data