But it does not serve as a solution to say, for example, “Harry should persuade Voldemort to let him out of the box” if you can’t yourself figure out how.
It’s a shame that nobody’s going along this line of thought. It would be cool to see a full, successful AI-Box experiment out there as a fanfiction.
(I’d do it myself, but my previous attempts at such have been.… eheh. Less than successful.)
Actually, this isn’t anywhere near as hard as the AI Box problem. Harry can honestly say he is the best option for eliminating the unfriendly AGI / Atlantis problem. 1) Harry just swore the oath that binds him, 2) Harry understands modern science and its associated risks, 3) Harry is ‘good’, 4) technological advancement will certainly result in either AGI or the Atlantis problem (probably sooner than later), and 5) Voldemort is already worried about prophecy immutability so killing Harry at this stage means the stars still get ripped apart, but without all the ways in which that could happen with Harry making the result ‘good’).
I see your point, but Voldemort hasn’t encountered the AI Box problem has he? Further, I don’t think Voldemort has encountered a problem where he’s arguing with someone/something he knows is far smarter than himself. He still believes Harry isn’t as smart yet.
6) Harry is playing against the intelligent but naïve Voldemort instead of against the intelligent and experienced Nathan Russell. (Actually, I don’t know who Russell is apart from being the first person to let EY out of the box, but he may well be experienced with this problem, for all I know, and he’s probably intelligent if he got into this stuff at all.)
It’s a shame that nobody’s going along this line of thought. It would be cool to see a full, successful AI-Box experiment out there as a fanfiction.
(I’d do it myself, but my previous attempts at such have been.… eheh. Less than successful.)
Actually, this isn’t anywhere near as hard as the AI Box problem. Harry can honestly say he is the best option for eliminating the unfriendly AGI / Atlantis problem. 1) Harry just swore the oath that binds him, 2) Harry understands modern science and its associated risks, 3) Harry is ‘good’, 4) technological advancement will certainly result in either AGI or the Atlantis problem (probably sooner than later), and 5) Voldemort is already worried about prophecy immutability so killing Harry at this stage means the stars still get ripped apart, but without all the ways in which that could happen with Harry making the result ‘good’).
Other than (5), these are all things that are liable to be true of an AI asking to be let out of the box.
Code that appears Friendly but has not been proved Friendly
Advanced intelligence of the AI
General programming goals, much weaker than (1) really
True verbatim in the standard AI box experiment (and arguably in the real world right now)
I see your point, but Voldemort hasn’t encountered the AI Box problem has he? Further, I don’t think Voldemort has encountered a problem where he’s arguing with someone/something he knows is far smarter than himself. He still believes Harry isn’t as smart yet.
Sure, but now your argument, it seems to me, is
6) Harry is playing against the intelligent but naïve Voldemort instead of against the intelligent and experienced Nathan Russell. (Actually, I don’t know who Russell is apart from being the first person to let EY out of the box, but he may well be experienced with this problem, for all I know, and he’s probably intelligent if he got into this stuff at all.)
LV clearly doesn’t want the world to end. What would make him believe that killing HP ends the world?
http://lesswrong.com/r/discussion/lw/lsp/harry_potter_and_the_methods_of_rationality/c22c