Visual Knowledge Structure Language

By writing in a "visual knowledge structure" style, instead of normal text, we hope to eventually be able to transfer bodies of knowledge an order of magnitude faster and more accurately. This is a very experimental high risk probe.

Last updated 8/1/00 - Jack Harich - Go Back

Status

0. See structural document for an example of our current progress..
1. A clear direction, as shown by the structural document and Nutshell Vision.
2. Slight confirmation (26% faster) that the Nutshell Vision is possible.
3. Need to run 2nd Experiment.
4. Exploring 3D tools which may allow better prototypes and faster evolution.
5. After some experience, created an overview of VKSL.

Mini Process 2

Nutshell Vision - A way to transfer knowledge an order of magnitude faster and more accurately.

Narrative - See Initial Concept.

Iteration One

Goal - Proof that the Nutshell Vision can be achieved by at least a factor of two.
(This is the Acceptance Test. The various experiments are the subtests.)

Risk Mgt - Some risks apply to many iterations.

Bogging down in technical fascitation or time sinks. - Resolved by using off-the-shelf tools like Visio to generate successively better prototypes.

Much work but little real progress - Resolved by small experiments that test progress towards goal.

Use Case List - These are the elements and uses of VKSL.

Elements - Must be sufficiently expressive for:

1. Concepts, aka nodes.Use nouns, sentences, images, etc.
2. Relations. Use verbs.
3. Relation direction. Use arrows.
4. Body of Knowledge border. Dotted line.
5. Relation order. Use numbers on arrowheads.
6. Importance. Use line thickness for now.
7. Certainty. (Not yet solved)
8. Custom properties. (Not yet solved)

Note that 5 and up are really just custom properties, which allow properties such as Universal ID, I know, Author, Date, Body of Knowledge, etc.

Possible future elements - Flow of logic, suggested reading sequence, more expected.

Uses - VKSL must allow various user types to:

Author - Create, edit, publish.

Reader - View, navigate, zoom, edit their own custom properties, etc. In particulare a Reader needs to find a particular piece of knowledge quickly.

Analyst - Merge or compare Bodies of Knowledge, inference, etc.

Trait List - Traits apply to several or many Use Cases.

1. A Reader must be able to learn how to "read" VKSL nearly instantly.

2. Can express nearly anything prose can.

3. Unambiguous. High accuracy.

4. Pleasurable to Author and Reader.

5. Very fast for authoring, reading, infomation finding.

Chronological Snips and Snails

6/10/00 - Concept - This gives a long look at the Initial Concept in general. We have left this unaltered over time, so that you can see how far the current direction has shifted.

6/22/00 - 1st Experiment - The JSL Study Group jumped in the water with Visual Knowledge Structure Language, or VKSL. First they did an experiment to see if VKSL was better than normal text. Then we discussed VKSL. Then they broke into groups of 2 to develop their own approach. At the end some of the groups reported on their observations, and the tabulated results of the experiment were reviewed. VKSL was 26% faster than text and about the same accuracy. I had expected a larger difference, but at least we got some.

To examine the experiment, see the Easy Quiz on VKSL, the structural document and the text document. The quiz was identical for both documents and had 20 True or False statements. The full results were:

Structural 1st 441 sec, 13.5 right, 9 subjects Structural 2nd 384 sec, 14.9 right, 9 subjects Textural 1st 597 sec, 14.7 right, 10 subjects Textural 2nd 406 sec, 14.7 right, 8 subjects Structural 1st is 156/597 = 26% faster than Textural, significant .8/14.7 = 5% more accurate, insignificant

(soon) - 2nd Experiment - Do the experiment again, but with a better quiz using checkoffs rather than True or False. There appeared to be an over emphasis on correctness, greatly slowing down those using the VKSL document for the quiz.

Status - New Quiz ready, not yet tested on a few people before large group again.

(Will take awhile) - 1st Milestone - This is to prove that something like VKSL actually is much better than text. To reach this milestone, try several things:

Start writing lots in VKSL to get deep into it and start evolution. For example do the website section about VKSL in VKSL, just as we did in the 1st experiment.
Keep tight, little, easy but useful experiments going.
Eventually get to an experiment that proves the milestone conclusively.
Consider 3D drawing package for better structural prototypes.
Avoid squandering time on Declarative Knowledge, renderer, etc now. Do that later when the investment is justified by having passed milestone one.
Postpone research until we have a strong initial solution path and deep problem understanding. This avoids making the same mistakes other have made.
Assume that the #1 use for VKSL is human knowledge transfer, but that #2 is human or machine inference. Eventually #2 will become #1, with incredible ramifications. Actually, inference is a type of knowledge transfer but it involves creating new knowledge out of old.

7/3/00 - The Nature of Abstraction - A short essay pointing to the extreme usefulness of explicit knowledge structures like those VKSL makes possible.