This parallels some of the work I'm doing with fun-theoretic utility, at least in terms of using information theory. One big concern is what measure of complexity to use, as you certainly don't want to use a classical information measure - otherwise Kolmogorov random outcomes will be preferred to all others.