Home › Blogs › SunnySingh's blog ›

The Thinking Machine 4

Submitted by SunnySingh on Tue, 2006-03-21 21:58
I know this class is about emergence, but since we're frequently dabbling in the area of artificial intelligence, I feel like I can't pass up the opportunity to post this. I came across this site last year and recently rediscovered it today. It's an interesting chess-playing AI applet called the Thinking Machine 4. When it's the computer's move, it searches the board for the best move; what makes this interesting is that while the computer is "deciding" it's move, it actually maps the possible moves and counter-moves with colored lines. It reminded me of the idea of how an agent should determine how its actions change the world around it--which is something we discussed in Intro to AI. As the gameplay progresses, it's evident that the program makes more specific moves. I don't know if I would quantify this as intelligence though. Could it be that the search tree the program is traversing is becoming smaller? The method by which a typical AI program uses to quantify a 'good' move doesn't usually change during a game--to my knowledge at least. With that said, it's hard for me to say if the moves are becoming 'intelligent' or if the 'good' moves are just a product of the current state of the board. On a more technical note, the programmers used something the quiescence search method in conjunction with alpha-beta pruning in order to cut down on the size of the search tree. It's pretty neat, so check it out. Enjoy!

Comments

2 Intelligence and Beyond

Submitted by LauraKasakoff on Wed, 2006-03-22 00:48.
That is a pretty neat applet. I wish I could quantify intelligence when it comes to chess, or at least I wish I could be intelligent when I play chess. I wonder how we are different than computers when we play chess. I know that when I am playing a game of chess I often wish that I could hold more information in my head. I feel like I can only keep track of three moves in my head, that is, I can only
  1. Choose a move to make
  2. "Think" about what counter-move my opponent would make
  3. See a possible advantageous move
I always feel the limits of my brain when I play chess because I always wish I could see further ahead. So what is my point in saying all of this other than letting you all know that I am a poor chess player? My main concern is that more than we'd like to admit, the human brain works at chess the way it does for most things in life. It surveys a situation, makes a conclusion of what to do based on how it's changing the environment and whether that change will lead to a better environment to interact with. Of course I don't want to suggest that the world is as simple as a game of chess. Although, now after studying emergence, I think that these same simple steps that we use to play chess are at the root of all of our higher thinking. If computers can master the art of playing chess, it makes me think that computers are but a few emergent step away of matching/ surpassing our own intelligence. That all said, I'm not sure if this applet is quite on its way to emulating intelligence. I gather that it figures out the best possible move for a given board setup, but does it "learn" based on how its human opponent performs? Where's the untapped emergent intelligence? Anyway, I think the colored mapping of the possible moves is quite spectacular and the collection of all possible moves in a game would probably fall under the heading of emergent art!

Search depth vs. static evaluator

Submitted by DavidRosen on Wed, 2006-03-22 18:07.
Computer chess programs (like this one) work in a fundamentally different way from human players. They play well mainly because they can look so many moves ahead, but usually have a relatively simple "static evaluator" (a function that takes in a possible state of the board and then returns a value for how promising it looks). As you say, the computer's static evaluator almost never changes over the course of the game. Its power rests almost entirely on its ability to process millions of board states per second. Humans, on the other hand, normally can only evaluate one or two board states per second. However, we have an extremely strong static evaluator that "caches" how similar patterns of moves played out in previous games, so we have a much stronger sense of what moves are worth considering. Chess playing AI is probably the most impressive achievement of GOFAI (good old fashioned AI), but I think it is fundamentally different from how human intelligence works in many ways, and is a very limited approach.

That is pretty neat, though

Submitted by Laura Cyckowski on Thu, 2006-03-23 16:44.
That is pretty neat, though having an opponent that can calculate so many possibilities does take a lot of the fun out of it :) Thinking about the search-trees and look ahead devices reminded me of the issue of short term vs long term for agents, which was mentioned in Johnson's Emergence (if I remember correctly you read it as well). Anyway, I hadn't thought of it in terms of AI, but it is a noteworthy feature of the limits at least on human intelligence.