The latest fragment of an idea which has got me interested: a wrapper for the MacMAME video game emulation system that will capture frames of a number of video game “sessions” and stash the resulting dataset somewhere. Then, I can play Ms. Pac Man endlessly, and in so doing, build a “map” of the evolution of a typical game. There’s a lot of image processing and math (stochastics, Markov chains?) which I’ll need to dig into in order to figure out ways to extract meaningful information from a huge 2×2xn array of pixels.
A natural extension of this idea might be a system which could parse all this data and determine, given the current state of a particular game, what the best next move for Ms P might be: imagine a glowing arrow just in front of Her Roundness, pointing the way to victory. It’s very pie-in-the-sky, but seems at least within the realm of possibility.
At any rate, just having data mined from a large number of video game runs would be nice. It’s like having a bunch of semi-random but fully deterministic experimental runs in a tightly-constrained system. I can imagine transposing the pac-data into music, or into alternate visual systems, in interesting ways.
The first major task will have to be capturing screen images and extracting data from them. I was considering Perl or Java for this task, though it seems like hacking the MacMAME source might be an ideal way to do this, and that’s likely written in C++. So I’ll grab that source (and CodeWarrior, I guess, to compile it!) and see what’s going on.
PS - my goal is to post at least once a day to this blog in order to keep things moving, my mind working, myself honest. We’ll see how long I can stick to the plan.