Why the "universal algorithm" is the modern equivalent of the perpetual motion machine, and researchers don't seem willing to give up on it despite its theoretical problems.