Hmm, if you start with 'uniform' prior, akin to feeding random strings to an universal Turing machine, or running random Turing machines on random data, it seems to me that if you run the machine for a while then look at it, you'll see either a halted machine or a machine doing something really dull, and of the interesting behaviours it seems to me that it'd be predominantly very simple algorithms that will be munching the data in the way that's not unlike simulation of an universe with very complicated state and very simple laws of physics. A sequence of transitions is exponentially unlikely over the length not to halt or degenerate .

The easy-to-visualize example of how simple algorithms arise from randomness is Conway's game of life with random start state. The complicated structures will be very rare. You'll end up with predominantly very simple 'machines' doing very simple things; even a glider gun will be very rare.