Jack Park jackpark@thinkalong.com Street address Palo Alto, CA Zip


Memorandum

Date: Wed, 3 May 2000 18:44:41 -0700

From:   Jack Park
jackpark@verticalnet.com
Reply-To: unrev-II@egroups.com

To:     unrev-II@egroups.com

Subject:   Towards an atomic data structure

From:   Eric Armstrong

But it also has a deeper meaning based in AI, carrying concepts like deep semantic modeling and machine reasoning. In this mailing list we are carrying on discussions of both kinds, using the same term at times.

I don't recall ever using it in any other way than the AI variant of a knowledge base, one which represents human knowledge in a form suitable for learning, modeling, and reasoning.

From:   Eric Armstrong

As a theoretical pursuit, I am glad to see AI-centric investigations continuing. And this is a reasonable forum for those discussions. But there are a lot of problems we don't know how to define in that arena, much less solve. So I am focused on what we can build today that will help us carry on those conversations.

To try to achieve a better "separation of concerns", I am going to try to avoid use of the term "knowledge" in this forum. I use it only as a markety buzzword, to distinguish this effort from data base systems and the like. But it tends to flood me with issues I'm not prepared to deal with at the moment.

Now, in general, I'm a fan of designing large, general systems before narrowing down the scope to something achievable. But just as we need to draw a line between what is achievable and what isn't at this point in time, I think it makes sense to make a distinction between what we are capable of designing (given finite resources) and what lies outside our scope.

I know that an AI-style "knowledge" repository is outside my present scope.

But, it is not outside mine. In fact, it is the sole reason I am involved. I am far less interested in doing the doable. There's already tons of stuff out on the web doing that. I'm vastly more interested in doing the hard, so hard, in fact, that Doug has only scratched the surface of it. Eudora would do just fine if email is all we need. A relational database with a good sql interface would do fine if all we want is to warehouse factoids. I don't know about you, Eric, but I want more. Lots more. What I want is hard, nothing less. And, settling for what's doable is just liable to make it impossible to append what's hard on as an afterthought.

Doug Lenat's Eurisko once discovered, by playing games it invented by itself (with programmatic biases built in by Doug), the heuristic that it is frequently useful to go to both extremes in any design situation, then back up just a little. With that heuristic, Eurisko was given the Traveller's catalog of warfare stuff and it built a naval fleet to go forth and conquer. On its first outing, everyone surrendered. They took one look at the fleet and gave up. Why? Because the main rule was "last boat in the water wins." Eurisko built one enormous boat, heavily armored, very expensive. And it built one tiny, cheap, speed boat that nobody thought of. All the heavy boats could never catch it and they would run out of fuel before it did. I want that.

I want the ability to roam around in knowledge space, looking for interesting ideas, things others didn't think of. Can't do it with an email system. Sorry.

Sincerely,



Jack Park jackpark@thinkalong.com