I agree. I feel like the post is poking a bit of fun at hokey religion, and in so doing falls into an error. The Norse would do quite badly in life if they switched to a prior based on description lengths in Turing machines rather than a description length in their own language, because their language embodies useful bias concerning their environment. Similarly, English description lengths contain useful bias for our environment. The formalism of Solomonoff induction does not tell us which universal language to use, and English is a fine choice. The “thunder god” theory is not bad because of Occam’s razor, but because it doesn’t hold up when we investigate empirically! Similarly, if the Norse believed that earthquakes were caused by giant animals moving under the earth, it would not be such a bad theory given what evidence they had (even though animals are complex from a Turing-machine perspective); animals caused many things in their environment. We just know it is wrong today, based on what we know now.
I agree. I feel like the post is poking a bit of fun at hokey religion, and in so doing falls into an error. The Norse would do quite badly in life if they switched to a prior based on description lengths in Turing machines rather than a description length in their own language, because their language embodies useful bias concerning their environment. Similarly, English description lengths contain useful bias for our environment. The formalism of Solomonoff induction does not tell us which universal language to use, and English is a fine choice. The “thunder god” theory is not bad because of Occam’s razor, but because it doesn’t hold up when we investigate empirically! Similarly, if the Norse believed that earthquakes were caused by giant animals moving under the earth, it would not be such a bad theory given what evidence they had (even though animals are complex from a Turing-machine perspective); animals caused many things in their environment. We just know it is wrong today, based on what we know now.