Mike, On 9/18/08, Mike Tintner <[EMAIL PROTECTED]> wrote:
> Steve:View #2 (mine, stated from your approximate viewpoint) is that > simple programs (like Dr. Eliza) have in the past and will in the future do > things that people aren't good at. This includes tasks that encroach on > "intelligence", e.g. modeling complex phonema and refining designs. > > Steve, > > In principle, I'm all for the idea that I think you (and perhaps Bryan) > have expressed of a "GI Assistant" - some program that could be of general > assistance to humans dealing with similar problems across many domains. A > diagnostics expert, perhaps, that could help analyse breakdowns in say, the > human body, a car or any of many other machines, a building or civil > structure, etc. etc. And it's certainly an idea worth exploring. > > But I have yet to see any evidence that it is any more viable than a proper > AGI - because, I suspect, it will run up against the same problems of > generalizing - e.g. though "breakdowns" may be v. similar in many different > kinds of machines, technological and natural, they will also each have their > own special character. > Certainly true. That is why it must incorporate lots of domain-specific knowledge rather than being a completed work at the get-go. Every domain has its own, as you put it, special character. > > If you are serious about any such project, it might be better to develop it > first as an intellectual discipline.rather than a program to test its > viability - perhaps what it really comes down to is a form of systems > thinking or science. > This has been done over and over again by many people in various disciplines (e.g. *Zen and the Art of Motorcycle Maintenance*). Common rules/heuristics have emerged, e.g.: 1. Fixing your biggest problem will fix 80% of its manifestations. Then, to work on the remaining 20%, loop back to the beginning of this rule... 2. Complex systems usually only suffer from dozens, not thousands, of potential problems. The knowledge base needed to fix the vast majority of problems in any particular domain is surprisingly short. 3. Symptoms are usually expressed simply, e.g. shallow parsing would recognize most of them. 4. Chronic problems are evidence of a lack of knowledge/understanding. 5. Repair is a process and not an act. We must design that process to lead to a successful repair. 6. Often the best repair process is to simply presume that the failure is the cheapest thing that could possibly fail, and proceed on that assumption. This often leads to the real problem, and with a minimum of "wasted" effort. 7. Etc. I could go on like this for quite a while. I have considered writing a book, something like "Introduction to Repair Theory" that outlines how to successfully tackle hypercomplex systems like our own bodies, even where millions of dollars in failed research has preceded us. The same general methods can be applied to repairing large (e.g. VME) circuit boards with no documentation, addressing social and political problems, etc. My question: Why bother writing a book, when a program is a comparable effort that is worth MUCH more? >From what I have seen, some disciplines like auto mechanics are open to (and indeed are the source of much of) this sort of technology, Other disciplines like medicine are completely closed-minded and are actively disinterested. Hence, neither of these disciplines would benefit much if any at all. Only disciplines that are somewhere in between could benefit, and I don't at the moment know of any such disciplines. Do you? However, a COMPUTER removes the human ego from the equation, so that people would simply presume that it runs on PFM (Pure Frigging Magic) and accept advice that they would summarily reject if it came from a human. Anyway, those are my thoughts for your continuing comment. Steve Richfield ------------------------------------------- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244&id_secret=114414975-3c8e69 Powered by Listbox: http://www.listbox.com
