Eric Armstrong
eric.armstrong@sun.com



Date: Thu, 18 Jul 2002 14:04:55 -0700



From:   Eric Armstrong
eric.armstrong@sun.com
Reply-To: ba-ohs-talk@bootstrap.org


To:     ba-ohs-talk@bootstrap.org




..
Subject:   Manhattan Project to establish the Knowledge Sciences



Rod, thanks again for your inspired summary of the record.

Thoughts on two of the items below.
..
Rod Welch wrote on July 10, 2002...

On 000407 Doug's SRI group was reminded about the need to define knowledge, as previously suggested on 000120.... .... Since that time there has been a lot of discussion about ontology, Wiki, SOAP, dialog maps, IBIS, C++, Java, collaboration, semiotics, topic maps, et al, for two (2) years or more.
..
Personally, I'm inclined to think of knowledge as "categorized how-to information". At least, that's a working definition that lets me think in terms of answers to the question "What X do we need help people do useful work?"
..
I think "categorized how-to information" fits the definition of X, and I don't mind using "knowledge" for the terminology -- at least until someone gives me a better definition of knowledge and I have to call X something else.
..
That way, when I want to search for information on making my Sony stereo component work, I can search for Sony by name, but search for categories of like "troubleshooting", "installation", "wiring diagrams", or things of that nature.
..
Right now, I'm fascinated by the idea of using network analysis to identify categories and sub-categories. Once presented with a cluster of related sites, any ontologist worth their salt could hang a label on it. A group of ontologists doing that would rapidly build up a useful ontology.
..
Note: Some categories, like "Sony components" are obvious. The site's links do the clustering. But it is intriguing to think that there might be some "how to fix it" pages out there that link to a collection of pages on Sony's site, and some on other sites, that would create a "troubleshooting" category.
..
Now, using the kind of technique Doug has been favoring, where you add metadata to existing pages without modifying the pages themselves, categories could be added wholesale.
..
Once an ontologist creates a name for a cluster of closely-related items in the network, all of those items would be immediately and automatically tagged.
..
Further, since variations in the network occur incrementally, the metadata tagging could be automatically updated each night so that new pages get the appropriate tags by virtue of their links.

Of course.... Some hair ball is going to come up with idea of making links just to get themselves categorized, but since no one will be linking to *them*, they lie far down the category-chain.
..
Rod Welch wrote on July 10, 2002...

People resort to calling things "Manhattan" in hopes of....
..
I had pretty much the same emotional reaction, although I would characterize things a little differently.
..
In my mind, the idea of a "Manhattan" effort is that:
  1. You get the best and the brightest minds.
    ..
  2. You put them in an environment that is free from distractions (like the pesky annoyance of making a living)
    ..
  3. You give them a difficult, challenging, and important problem to solve.

  4. You put them in close quarters, so ideas can rub elbows, jostle each other, and spur a cycle of innovations.
..
Such mechanisms can and do work. However, failing to meet all of the requirements leads to nice cooperative efforts that fail to meet the "Manhattan" ideal.
..
In particular, #2 and #4 are key. We *want* distributed systems that will allow #4, but do not yet have them, so physical proximity is a requirement for a "Manhattan" effort, at this point in time.
..
More important, though, is #2. Combined with #4, these requirements are another way of saying "isolatation". Isolation serves a number of important purposes:
..
It would be a great idea, imho. But calling an effort a "Manhattan Project" doesn't make one -- even if it is a noble, valuable, and desirable project!

Sincerely,



Eric Armstrong
eric.armstrong@sun.com