Josh, On 4/11/08, J Storrs Hall, PhD <[EMAIL PROTECTED]> wrote: > > Actually, the medical expert systems of the 80s I had any conection with, > such > as the glaucoma expert from Rutgers, beat out human doctors in diagnoses > within their field of expertise. (And still weren't adopted...)
Yes. There was a presentation at the last WORLDCOMP AI conference where a computer was routinely beating people at identifying details in jaw X-rays, that were needed to make key measurements to fit prosthetics. This sort of an approach works well in closed systems that are well understood, but poorly in "wide open systems" where the goal is to get some handle on a bad situation, from which to leverage some sort of cure - and possibly even something that has never before been proposed for that particular ailment. How can a (simple) computer possibly do this? Because new (to the computer) cause and effect chains usually have some known (to the computer) links. If just one link in the lead up to the self-sustaining loop in the cause and effect chain is known and has a cure, and if just one link in the self-sustaining loop has any sort of treatment at all, then the condition can easily be cured for once and for all, even though the overall condition and diagnosis is completely unknown, and remains unknown to the computer. BTW, the attached paper included some remarks about Jay Forrester & System > Dynamics. Forrester came out of exactly the same background as Cybernetics > -- > working on automatic radar-directed fire-control systems, at MIT, during > WWII. And both his stuff and Cybernetics consists basically of applying > feedback and control theory (and general differential analysis) to things > ranging from neuroscience to economics. Like Joe Weizenbaum, Jay Forrester had a REALLY good idea that he put into his writings, but only as a side note and not any sort of main theme. However, I'll take a good idea from anywhere that I can find it. That idea was that when you analyze many similar systems, like corporations, that you typically find a few common modes of failures from among the countless possible modes of failure. Once you have gotten to that point, you can quickly correct most of the failures by simply screening for the common modes of failure, with no simulation needed. Jay was fixing corporations that are each unique, though they do operate in a "standard" world. However, people (and circuit boards and most of the other things that people might like computers to fix) are MUCH more alike than corporations, so that most experts already know most of the known modes of failure - with no computerized analysis needed to identify those potential modes of failure. Even military science has remained much as it was millennia ago, as exemplified by writings like *The Art of War*, despite the emergence of modern weapons. For example, in Roman times it took ~2% of a population to occupy it, and that percentage remains true today. That we are attempting to do this with ~1% of the population of Iraq underlies our failures there. This was pointed out early in the Iraq conflict, but those generals were promptly discharged for having a bad attitude. Steve: If you're saying that your system builds a model of its world of > discourse as a set of non-linear ODEs (which is what Systems Dynamics is > bout) then I (and presumably Richard) are much more likely to be > interested... No it doesn't. Instead, my program is designed to work on systems that are not nearly enough known to model. THAT is the state of the interesting (at least to me) part of the real world. In short, I am apparently going where no one has gone before - applying new methods to solving difficult problems in poorly understood systems. I'll gladly leave the easy stuff (modeling well-understood systems) to others. BTW, there is a generally unrecognized principle (except to some experienced System Dynamics types), that the cause and effect chains are LONG and usually involve some lack of understanding among those who designed the systems that we must now deal with. Only the most arrogant would presume their own perfection in comparison with those who designed the world in which we live. Correcting that arrogance is THE primary benefit of System Dynamics, which forces people to code how the systems REALLY work (to make the simulations play like reality) and not just how they THINK that those systems work. Hence, simulational System Dynamics must be confined to systems whose operation can be observed or instrumented. Unfortunately, this lets out most of the REALLY important real-world problems, especially medicine, from simulated solution. That reasoning new cures for medical conditions that are unknown to the computer at once appears to be SO difficult, yet is relatively easy given the right approach, is why I/we chose chronic illness, the hardest part of medicine, as our demo. > ps -- of course, you know that if you're using Excel to integrate > dynamical > systems, you are in a state of sin. (Sigh of relief for not being in a state of sin) Access is VERY different from Excel. Access is really just Visual Basic with an integrated interface into relational databases and a forms engine. When writing large/complex AI programs that will never be fully tested, it is crucial that you generate as few subtle/hidden bugs as possible, and I can think of no better platform than VB to accomplish that. It can never be a "VB Puzzle Book" because the language is SO simple and direct, but the very presence of the "C Puzzle Book" is grounds for dismissing C as being a serious AI programming language. BTW, I have my own thoughts that are VERY different from the mainstream on what the perfect AI programming language would be like, but that is a different subject... Steve Richfield ------------------------------------------- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244&id_secret=98558129-0bdb63 Powered by Listbox: http://www.listbox.com