Theoretically it should be possible to construct a generic species-level brain emulation from the genome alone and then populate it with the individual’s experiential data, thus effectively recreating said individual, yet in my opinion it would be much more economical to look for efficient representations of the functional and structural connectome, acquired directly by some appropriate means from the actual brain in question.

Of course, you could also just directly compress the connectome plus whatever extra quantitative detail is needed (even using a very crude compression scheme), and this would probably also be “much more compact,” the above argument just suggests something about how much more simple it should get.

Basically: if you believe that your environment is computationally simple, then you can look out at the environment and see humans, deduce that they are computationally simple, observe they have basically the correct I/O behavior, and then start to predict the features of U by observing and interacting with humans (or experimenting on). Conversely, if you believe that U is likely to reflect something like human volition, then you can quickly conclude that the instrumentally correct thing to do is to care about versions of yourself with lots of measure under the universal prior. But if you neither know anything about U, nor think your environment is computationally simple, the situation seems a little more complicated.