IGS Discussion Forums: Learning GS Topics: GS can make safe our definition of intelligence
Author: Ralph E. Kenyon, Jr. (diogenes) Tuesday, October 9, 2007 - 02:32 pm Link to this messageView profile or send e-mail

I don't know what "our definition of intelligence" "is".

As far as Jeff Hawkins commercial venture is concerned, virtually every definition of intelligence and development has adherents who have tried to go forth and "preach" their "latest" model, often trying to use the interest to sell some product. For decades we have had a model of intelligence based no rational reasoning, and liteally trillions of dollars have been spent tying to devise a machine that could rival human performance - as Jeff Hawkins noted - all resulting in a dismal failure or using resources so vast as to make them impractical. Consider Deep Thought, the highly specialized chess playing program. Language tranlation and artificial intelligence efforts were dubbed by Wendy Lehnert as a "degenerating research project" at the AAAI conferences I attended back in the '80's. But I don't think "On Intelligence" was initially written for the purpose of furthing the subsequent commercial venture. I think he genuinely wanted to share his exciting new model. The venture, "Can we use this model to try to build artificial intelligence again?" represents the perseverence factor that results in getting something with utility.

I see "intelligence" as a bit like special effects. It has been said, "It's science if you know how to do it, magic if you do not.", and the practicing magicians have their science of trade secrets. Several examples of human behavior have been heralded as hallmarks of "intelligence" until such time as someone demonstrated an effective procedure for accomplishing the previously heralded ability, at which time the "is that all it is" syndrome resulted in the discounting of the behavior as exhibiting "intelligence"; thereafter it was "merely" a "learned ability" - a "skill" that could be taught.

"IQ" tests, to commit an objectivism, measure what they measure. Many instruments of measurement produce correlated results. The correlation itself, when there are anomalies, is used as a diagnostic tool to "identify" various learning disabilities.

I repeat. I have have no idea what "our" (who?) "definition" of "intelligence" "is". Since it is "obviously" a "concept by intuition", and a very abstract one at that, it seems to me to be a little outside the more extensional approach of general semantics. Instead of asking what is one's intelligence, ask what can one do, what can one not do, how can one learn to do what one cannot do, and how easy is it for one to learn that to do it. Trying to even formulate a "definition" for "intelligence", let alone measure it, seems to me like a "red herring" distracting us from the business of learning to sucessfully navigate our environments (physical, symbolic, cultural, political, etc.).

Personally, I'm still excited by Jeff Hawkins model of intelligence in terms of auto-responding memory and the memory predictive framework together with massive amounts of experience.

How can we use this in general semantics?

1. Experience: Get some (LOTS) with extensional orientation.
2. Prediction: Learn valid scientific inference techniques.
3. Synthesis: Abstract from experience and build a model that predicts what to expect as a function of action based on past experiences and insights.
4. Testing: Use the model for daily navigation.
4a. Remain aware of and alert to prediction failures.
1a. Add results to experience.
3a. Analysis: Use valid scientific inference techniques to analyze the difference between the model's prediction and the experience that was actually seen.
3b. Revision: Update the model continually.