I’m not sure I get your analogy, but my intuition is that FactorialCode is right regarding Turing completeness / just a different minimum description length constant. Like, I think I can turn a collection ‘programatically generated’ but not self-similar sub-DAG into a structure which has internalised the generative power. For example, perhaps it now consumes and switches on some extra state.
If I can do that, I now have a self-similar building block that counts for our symmetries. There’s some overhead in the translation, hence the different MDL constant.
I’m not sure I get your analogy, but my intuition is that FactorialCode is right regarding Turing completeness / just a different minimum description length constant. Like, I think I can turn a collection ‘programatically generated’ but not self-similar sub-DAG into a structure which has internalised the generative power. For example, perhaps it now consumes and switches on some extra state.
If I can do that, I now have a self-similar building block that counts for our symmetries. There’s some overhead in the translation, hence the different MDL constant.