Here's a notion I've been kicking around for a while, sort of the opposite of Deep Blue. In the IBM initiative a single team of computer scientists (with the help of a grandmaster) built an extraordinarily complex machine to master a specific task. In what I have in mind (I'm tempted to call it "shallow pink" but I'm afraid the name might stick) a number of teams will write simple programs (at least simple by today's standard) that will do something extremely general.
The task is to write a program to play a game without knowing exactly what the game is.
This is how it would work:
The game will be played on a rectangularly or hexagonally tiled board of dimensions no more than 20x20 or 15x15x15. It would involve placing and/or moving pieces that may or may not be differentiated. The rules should be simple enough for a human player to get up to speed relatively quickly (no humans will actually be playing; this is just a rule of thumb to keep things manageable). The rest of the rules won't be announced until the programs have been submitted for that year's competition.
I'll leave the details to those better qualified to supply them but here are the basics. The programs will be size constrained, held to a format, set up so that game rules can be entered as a separate set of parameters, and configured for automated play.
Of course, the quality of play won't be that impressive but that's not the point. The idea is to get lots of bright people trying to come up with innovative, elegant and flexible approaches to problem solving.
I suspect most OE reader are better programmers than I am so I'm opening the floor for suggestions.
* Yeah, I mean half-baked, but this way sounds better.
No comments:
Post a Comment