There is a relative notion of Kolmogorov complexity. Roughly K(X| Y) is the size of the smallest program that outputs X given Y as input. I agree that K(Thor | Human) << K(Thor), but I believe that this is still much larger than K(Maxwell|Human). This is because K(Maxwell|Human) ⇐ K(Maxwell), which is really small. On the other hand to specify Thor given a description of a human, you get huge savings on describing how a humanoid works, but still have the task of describing what lightning actually is (as well as the task of describing how Thor’s thoughts translate into lightning).
To reply years after the original post...
There is a relative notion of Kolmogorov complexity. Roughly K(X| Y) is the size of the smallest program that outputs X given Y as input. I agree that K(Thor | Human) << K(Thor), but I believe that this is still much larger than K(Maxwell|Human). This is because K(Maxwell|Human) ⇐ K(Maxwell), which is really small. On the other hand to specify Thor given a description of a human, you get huge savings on describing how a humanoid works, but still have the task of describing what lightning actually is (as well as the task of describing how Thor’s thoughts translate into lightning).