Hm, I might have confused what you are allowed to do if you had enough compute to run Solonomoff Induction with Solomonoff induction itself, so that’s maybe the issue I had here.
If I wanted to create my own argument for why Solomonoff induction could do it, it’s because that it’s essentially a halting oracle, which allows it to compute/create all the digits of Chaitin’s constants, given that it can compute in general the recursively enumerable sets, and since it can compute all the digits of Chaitin’s constants, it can basically read them like a book, and thus it has predicted the sequence of digits.
Solomonoff induction is a specific probability distribution. It isn’t making “decisions” per se. It can’t notice that it’s existence implies that there is a halting oracle, and that it therefore can predict one. This is because, in general, Solomonoff induction is not embedded.
If there was a physical process for a halting oracle, that would be pretty sick because then we could just run Solomonoff induction. As shown in my post, we don’t need to worry that there might be an even better strategy in such a universe; the hypotheses of Solomonoff induction can take advantage of the halting oracle just as well as we can!
As shown in my post, we don’t need to worry that there might be an even better strategy in such a universe; the hypotheses of Solomonoff induction can take advantage of the halting oracle just as well as we can!
You do mention that the methods of reasoning have to be computable for this to work, and there I’m quite a bit more skeptical of that condition holding.
Hm, I might have confused what you are allowed to do if you had enough compute to run Solonomoff Induction with Solomonoff induction itself, so that’s maybe the issue I had here.
If I wanted to create my own argument for why Solomonoff induction could do it, it’s because that it’s essentially a halting oracle, which allows it to compute/create all the digits of Chaitin’s constants, given that it can compute in general the recursively enumerable sets, and since it can compute all the digits of Chaitin’s constants, it can basically read them like a book, and thus it has predicted the sequence of digits.
Solomonoff induction is a specific probability distribution. It isn’t making “decisions” per se. It can’t notice that it’s existence implies that there is a halting oracle, and that it therefore can predict one. This is because, in general, Solomonoff induction is not embedded.
If there was a physical process for a halting oracle, that would be pretty sick because then we could just run Solomonoff induction. As shown in my post, we don’t need to worry that there might be an even better strategy in such a universe; the hypotheses of Solomonoff induction can take advantage of the halting oracle just as well as we can!
You do mention that the methods of reasoning have to be computable for this to work, and there I’m quite a bit more skeptical of that condition holding.