Date: Fri, 27 Oct 2000 12:58:09 -0400
|Poetry and Knowledge Management
Paul Fernhout wrote:
On Poetry vs. Fine-grained Meaning in Knowledge Management
The more I reflect on this, the more I think the issue of understanding the differences and similarities of Poetry and Knowledge Management is key to seeing the effective limits of hyperlinking and maybe working through that into ideas for better KM tools.
I have been reviewing some of Rod Welch's site, especially pages related to Knowledge Management, especially the comments related to trying to define what KM means and who will use it or pay for it or change their daily practices to get its benefits (if any). Or, in other words, the OHS purpose and vision.
I have also been thinking about the previous message I sent discussing the distinction between referencing text and referencing concepts and mentioning how one could not hyperlink poetry in a meaningful way (because to fix the meaning of words defeats much of what the poet attempts to convey with purposeful ambiguity).
In this sense Poetry represents the Knowledge Management problem in a very bright light. Poems are often intentionally ambiguous, with interpretation expected oftentimes to depend on the reader. To an extent, poetry describes all communications, even though the intent may be to convey more precise meaning.
When we talk about "unique IDs" and "global identifiers" we are very much talking about sharing meaning through communications. Linking is an attempt by the author to force (or make convenient the movement of) the reader to a certain metaphorical understanding of the linked item. Yet, the reader may prefer other links (either metaphorically or to other resources) depending on the reader's needs or intents or interests. Or the reader may interpret a reference, phrase, or link in a way other than as the author intended.
On reflection, I would say pointers to knowledge or concepts cannot be called "fine grained", as opposed to the way that we might call pointers to lines within a web page more fine grained than a pointer to the web page itself.
The "finest grained" thing we have is words, but they are usually defined in context. Example: we are lost in the woods and you point to a tree (making a signal somewhat equivalent to saying a word). That signal could mean any of:
I explained the meaning of the signal "tree" in terms of words. But, as you think about those words, you will realize they too are just signals -- just pointers. And so, I haven't completely resolved the problem. What does it mean to "cut down the tree"? What does "cut down" mean? Pull off a branch? Chainsaw through the tree? Saw through it? Cut it into logs? Make it into boards? So again, vagueness. The desired outcome depends on the context -- the intent behind the signal. The intender might not even be sure of exactly which is desired -- focusing more on the end goal (burning wood vs. building shelter) than on the exact cutting pattern discovered by trial and error or limited by available tools.
If one considers communication and related knowledge that inspires it (especially verbal communication) as metaphorical, then we can't say knowledge is ever fine grained. Augment's numbers are locations of paragraphs, like Rod Welch's communication metrics numbers indicate lines on his web site pages. (Both are somewhat more than that because they are hierarchical, so fragments indicate larger textual units, and in Rod's case the date is also encoded.) I would say though that what is being pointed to in a "knowledge" sense is not so much a word or sentence or line or paragraph, as much as a pointer into an ongoing presentation of metaphors in a certain larger context. To understand the intended meaning of the word "tree" at a location on a web site, one must understand the context around it. (Infinite regress up to understanding the universe can be avoided by at some point us thinking we understand the context of the conversation as a conventional one we are used to workign with.)
My point for going on at length is to say that I am realizing (or re-realizing or remembering?) that there is to an extent no way to do "fine grained" knowledge representation. You can point at a word, but since the word loses its meaning by itself, you are pointing at a paragraph or essay -- which is a context. It is true that pointing in one place in an essay may conjure up a different meaning than pointing to another part. Anyway, my point is that while it may be easy to think about pointing to textual artifacts (messages, documents) it is hard to point to specific "meanings". At best we can say, I think that section of text is intended to mean "X" where "X" is another set of signs. So, to reiterate, even when we point to the words, we are not pointing to the meanings. The sign is not the signified. The words are not the wisdom. This is common knowledge in sociology, communications studies, and a bunch of other fields -- I'm just hammering on this point in this context of designing knowledge management tools.
People's minds consist of words and images (and impressions, thoughts and memories, etc.) in action. That is, people process information and have motivations, have perceptions, and take actions. The "knowledge" or "wisdom" of a person (inside a person?) consist of information in that *active* context, and alongside other information also in that context. That processing is quite complex -- involving multiple simultaneous ways of representation (e.g. Marvin Minsky's latest work) and very complex perceptions related to combinations of visualization, verbalization, and other sense impressions acting in a sort of mental world simulator full of various thinking tools (i.e. simplified ways of predicting the future or the past or making a choice). This is one reason that conversational AI-type systems that just process textual symbols fail to do a very good job of duplicating human thought; they can't for example handle simple 3D geometry problems like imagining using an umbrella to knock down a banana which any chimpanzee could easily solve. So -- thought is more than language, although language is used in much human thought.
You can't easily point to a bit of knowledge in a document, any more than you can point to one dot in a painting by Georges Seurat (a painter who created Pointilism) and say that is a picture by itself.
So, we must distinguish between creating memory aids and document management systems, and creating artificial intelligences. Obviously, to the extent people are using memory aids, they are "augmenting" themselves into being a sort of artificial intelligence. This is not an argument against AI; it is just to distinguish "AI" from "Knowledge Management".
Anyway, I am trying to get at the issue that our understanding of a knowledge management system has to rise above the notion that the "knowledge" or "wisdom" being managed is in the computer system. It is in the intelligences (typically based around people) of which the knowledge management system may form a part (an aid for memory, communication, and calculation).
The designs for Knowledge Management tools must soar above the mundanity (but necessity) of managing chunks of texts, images, sounds and so on. This is in line some with Doug's point about how the user (or user community) must co-evolve with the tool and information in it. In effect, the knwoledge is distributed throughout the entire system. But the system itself must still reflect the special needs of doing KM which may require interfaces and processes different from more conventional tools. What these interfaces and architectures should be is a subject of debate -- obviously Augment or Memex or Xanadu set the stage as archetypes.
So, what I am saying is knowledge is in the system including the people. When we talk about knowledge management systems we are talking about systems that help people or communities to manage their knowledge -- that help people organize knowledge, communicate it, revise it, and so forth. But that does not to mean we ever have to say the "knowledge" is in the system, any more than we need to say that "knowledge" is in a book.
A book may have words, and page numbers, and may inspire you, and tell you things you didn't know -- but the knowledge in the book resides more in the system of author and reader sharing certain metaphorical backgrounds and thus being able to understand a certain communication made in print.
And one must admit, since the author and reader may never share exactly the identical metaphorical background, the meaning of any communication to the reader may not be what the author intended. Perhaps one can call this meaning shift "concept drift?" Most non-routine communications probably contain some element of "concept drift", as I'm sure does this communication.
But with enough communciations, generally I would think the parties begin to understand the other's metaphorical system, even if they may decide not to share it in the sense of relate values or assignment of "truth". Thus, they may come to better understand the intent of the communications by the sender, even as they may still also interpret the communication as poetry using their own metaphorical system.
However, there must be some commonality in metaphor, otherwise the reader could get little out of a book or message at all. Think about the StarTrek:TNG episode "Darmok" where the aliens talked in terms of mythological figures and storylines which the Enterprise crew has no knowledge of. The words were spoken and understood -- but there was no meaningful communication.
The bottom line: We'll never be able to point to the "Knowledge" in a "Knowledge Management" system. But, that doesn't mean pointers into text aren't useful, or that one can't construct tools to help manage knowledge as it is communicated by text, images, sounds, and so on. Or in other words, think of an Augment-ish library as communications system (as opposed to an AI). Which brings us back to email as a good vehicle...