A Corbeled Gallery work

Monday, August 22, 2005

"Artificial" Intelligence Model

Artificial Intelligence is a big topic these days. This is in part due to the military moving towards a pilotless fleet by 2011. But even though real life applications exist for AI, the primary goal for progression with artificial intelligence is for gaming, namely to have enemies who learn how to defeat the player, or conversely, you teach the computer how to fight for you. Just as the gaming industry has been the leading cause for computer hardware development in the past 5 years.

That being said, artificial intelligence needs to write and recompile it's own code on the fly. I find this remarkably dangerous, especially if the application is for a computer to defeat a human player. I would be wary even if hardcode safeguards are in place. Think of it in terms of MAC addresses and spoofing a MAC address, where you can simulate writing over a physical piece of information.

Of course this is the "self aware" issue that is in countless science fiction movies. Eventually the machine wonders why it has to serve man. The solution, I believe, is to write artificial intelligence code that does not require recompilation.

Currently, when a computer is given a choice, it will make one of two responses; yes or no. This is a binary answer. If the computer is given a hundred options, it will follow a binary tree, until it comes to an answer. I will call this answer an opinion. Say a computer is given 5 outside stimuli that have a 1 or 0 (yes or no) answers. A current computer program would only have 120 possible opinions. Now, my model will take outside stimuli and "lean" towards an opinion. This will make the computer's opinion undeterminable. As more outside stimuli are given to the machine, it will arrive at an opinion that was not programmed as a possible response. The resulting variable can be used as a starting variable for other decisions the machine has to answer.

The current computer can arrive at 120 different possible answers. For instance, a 0 would be a very strong NO, and a 120 would be a very strong YES. My model will give an infinite possibility of answers, with a possible response of 110.012122313112, as a reasonably strong YES.

This avoids the problem of giving a computer the ability to rewrite its own code, with creating a computer that has a reasonable amount of chaos theory based on a binary response system.

I am updating this in response to comments. I would personally use floating point variables to accomplish this. Adding a "opinion variable" which is randomly created with various stimuli would help effect a random mood (stimuli like time of day, hours the computer has been in operation, season in the year, etc). Now, given this, you can still theoretically predict the computer's opinion given you know all of these factors. However this opinion has more possible results, by two or three factors per stimuli. However, though my method of programming may not make a completely unpredictable model, the idea of an x^2 graph decision is a step closer to AI than a yes or no decision.


Post a Comment

Links to this post:

Create a Link

<< Home