More on crowd feedback

Everyone has cellphones now, right? If you had a few highly directional antennae you might be able to use the amount of RF activity in a few cellphone bands as an approximation to crowd activity. You could maybe also look for Bluetooth phones and maybe remember individual Bluetooth ID's, although I'm not sure if most phones will respond to Bluetooth probes in normal operation.

Another approach would be suitable for a conference or other event where participants have badges. Simply put an RFID tag in each badge and have a short-range transceiver near each display. Now the art not only responds to aggregate preferences, but it also personalizes itself for each viewer. Effects which have previously held a participant's attention will appear more often when that participant is nearby. This will probably result in overall better evolutionary art -- instead of trying to impress the whole crowd, which is noisy and fickle, the algorithm tries to impress every individual person. While it's impressing one group, other people may be attracted in, and this constitutes more upvotes for that gene.

I think one important feature for this to work effectively is a degree of temporal coherence for a given display. If they're each showing short, unrelated clips (like Electric Sheep does), then people will not move around fast enough for the votes to be meaningful. Rather, each display should slowly meander through the parameter space, displaying similar types of fractal for periods on the order of 10 minutes (though of course there may be rapidly-varying animation parameters as well; these would not be parameters in the GA, though their rates of change, functions used, etc. might be).