Partager via

Finally, a computer I can beat at chess

  • Article

The last time I could beat a computer at chess was probably about 15-20 years ago. I've gotten marginally worse, and computers have gotten just a tad* better.

Avid readers may recall I got a Nintendo DS earlier. I was given Clubhouse Games (2 thumbs up) as a present, and that has a large variety of traditional games including a Chess program. It was easy on the easy level, but I was floored when I could actually beat it at the harder levels.  What I really noticed was that it took a very long time to move on the hardest level to move.

That got me to check out the official hardware specs at https://www.nintendo.com/techspecds just say One ARM9 and one ARM7. Other less-official sources give more details and say the CPUs are 67 mhz and 33 mhz respectively, with 4 MB ram. 

So it's no Deep Blue. That's not a lot of computational power, compared to basic desktop machines today with 1+ Gigabyte Ram and processor speeds in ghz.

Hardware  aside, I noticed an algorithm deficiency. The DS chess game seriously undervalues the queen. A simple algorithm adjustment here would make it more formidable. So it's not all a hardware issue.

The bottom line is that if my opponent is bad enough, I can win.

On a related note, it would be an interesting tool to evaluate a human player and then list the approximate chess computer strength they correspond to. It would be like those cheap magazine quizes: "You are a  486 50mhz 4 MB ram chess player". I bet it already exists.  Anyone got a link?

 

(* = sarcasm.)

Comments

  • Anonymous
    April 03, 2007
    Unfortunately, computers never play remotely like humans, whether in total information games like chess, or in more computer-oriented games like first-person shooters. The mapping function is non-linear and multidimensional. For example, I've never seen an FPS AI opponent aim like a human - that is, reflecting the difficulty of adjusting the aim over time through an arc - they're usually too poor at aiming at long-distance still targets, and far too good with up-close fast-moving targets. Aiming difficulty isn't a function of distance; it's a function of arc adjustment velocity, while factoring in the issue of fine motor control for tiny arc adjustments. It's similar with chess. Humans build hierarchical abstractions above the simple layout of pieces on the board, to the point that the current layout of the board is a terrain, and each move both navigates and modifies the terrain. Most computer chess algorithms don't do this; they typically use simple alpha-beta with pruning and a fixed number of iterations. With a bit of insight, it should be easy for an experienced player to manipulate the computer into playing a certain way, taking advantage of its limited foresight.

  • Anonymous
    April 03, 2007
    Unfortunately, computers never play remotely like humans, whether in total information games like chess, or in more computer-oriented games like first-person shooters. The mapping function is non-linear and multidimensional. For example, I've never seen an FPS AI opponent aim like a human - that is, reflecting the difficulty of adjusting the aim over time through an arc - they're usually too poor at aiming at long-distance still targets, and far too good with up-close fast-moving targets. Aiming difficulty isn't a function of distance; it's a function of arc adjustment velocity, while factoring in the issue of fine motor control for tiny arc adjustments. It's similar with chess. Humans build hierarchical abstractions above the simple layout of pieces on the board, to the point that the current layout of the board is a terrain, and each move both navigates and modifies the terrain. Most computer chess algorithms don't do this; they typically use simple alpha-beta with pruning and a fixed number of iterations. With a bit of insight, it should be easy for an experienced player to manipulate the computer into playing a certain way, taking advantage of its limited foresight.

  • Anonymous
    April 04, 2007
    I believe Fritz 8 has the rating feature (although I have not enabled it in my copy).  It regularly kicks my but at Sparing-Hard but I win about 1/3 of the games at Sparing-Medium.