Interesting. How does the program determine hard questions (and their answers) without qualifying as generating them itself? That is, the part about enumerating other programs seems superfluous.
[Edit added seconds later] Ok, I see perhaps it could ask something like ‘predict what’s gonna happen 10 seconds from now’, and then wait to see if the prediction is correct after the universe runs for that long. In that case, the parent program isn’t computing the answer itself.
You don’t need to be able to generate the answer to your hard problem yourself, only to check that the superintelligence’s offered answer is correct. These two abilities are equivalent if computing resources are unlimited, but you could run the superintelligence for a limited time… This line of thought seems to lead into the jungle of complexity theory and you should probably take my comments with a grain of salt :-)
Interesting. How does the program determine hard questions (and their answers) without qualifying as generating them itself? That is, the part about enumerating other programs seems superfluous.
[Edit added seconds later] Ok, I see perhaps it could ask something like ‘predict what’s gonna happen 10 seconds from now’, and then wait to see if the prediction is correct after the universe runs for that long. In that case, the parent program isn’t computing the answer itself.
You don’t need to be able to generate the answer to your hard problem yourself, only to check that the superintelligence’s offered answer is correct. These two abilities are equivalent if computing resources are unlimited, but you could run the superintelligence for a limited time… This line of thought seems to lead into the jungle of complexity theory and you should probably take my comments with a grain of salt :-)