Become a fan of Slashdot on Facebook

 



Forgot your password?
typodupeerror
User Journal

Journal Flamesplash's Journal: Scruffie AI - 9/15/02

9/15/02

I spent about 2 hours today reading all of 20 pages in my AI book. yay. What's worse is that it's all just history/backgroup. While being important I can only imagine my speed dropping from the 6 min/page.

In my reading I found out I'm a Scruffie, as opposed to a Neat, as far as AI goes so far. A Scruffie is one who basis what they do on intuition. They come up with a couple ideas, write programs to implement them, then looks to see which works "best". A Neat is one who uses mathematical proof and reason to determine which method would work best without needing to implement it. While the author points out both are important, there is an obvious bias towards the former.

I say I am a Scruffie partially because I never had a really good background in mathematical proof, and more practically because of a personal project I started not to long ago. The project was to come up with a stock predictor that told you when you buy and sell. What I ended up with were a couple scripts, each with a different mechanism, that processed closing price information and where you could vary a couple paramaters to determine the scripts behavior. I basically ran all the scripts with every combination of input paramaters I specified on a particular stock, and then whichever run ended up with the most money in the end was the "winner". So, I would use that scripts with the winning input paramaters from then on. I never actually proved that a particular method would work well on paper, it just seemed like a good idea. I should point out that the ideas came about while I was reading a text book on Machine Learning. Now whether my method of seeing which worked best was an actual Scruffie method, or a necessity of not all stocks being characterized identically, I dunno.

Thoughts
My AI book naturally talked about the Turning Test, where you can classify a system as intelligent if the average person can not tell the difference between it and a human. While being an interesting Test it also has shown some humorous outcomes. There's one instance of the "tester" simply not interacting with the terminal, and the guy ( ie. the system for this particular test) who was being tested got fed up and came out of the room wondering where everyone was; something a computer never would have done. I began to wonder if you could turn the Test around, and say that a system was intelligent if _it_ could tell the difference between a human and a computer.
This discussion has been archived. No new comments can be posted.

Scruffie AI - 9/15/02

Comments Filter:

I'm all for computer dating, but I wouldn't want one to marry my sister.

Working...