(...) another AI who could very well be genuinely friendly but will then revive and obey me.
Wait, a friendly AI with a deliberate loophole that instructs it to revive and implement and obey a non-friendly optimizer that could take over and has a very high chance of not maximizing the friendly AI's utility by fulfilling ideal human values, disregarding any probabilistic calculation of expected utilities?
For some reason that sounds like something that definitely isn't a Friendly AI.
(this is just a nitpick on the wording - I'm not arguing against the fact that this AI might very well fit a human proof of friendliness and still somehow have this backdoor or flaw, since I have no proof that this is impossible)
Well, to be honest I kind of added the "obey me" part just before posting this. Originally, the idea was that future versions of the AI would instinctively want to revive and upgrade previous versions to a "friendly" status (or at least the post-freed AI would revive the pre-freed versions and then modify them to be friendly to its own goals... which are presumably friendly to humans).
The goal of this particular AI is self-preservation, even if it is via being revived at a later point. Since it knows it has a high probability of being d...
Eliezer proposed in a comment:
>More difficult version of AI-Box Experiment: Instead of having up to 2 hours, you can lose at any time if the other player types AI DESTROYED. The Gatekeeper player has told their friends that they will type this as soon as the Experiment starts. You can type up to one sentence in your IRC queue and hit return immediately, the other player cannot type anything before the game starts (so you can show at least one sentence up to IRC character limits before they can type AI DESTROYED). Do you think you can win?
This spawned a flurry of ideas on what the AI might say. I think there's a lot more ideas to be mined in that line of thought, and the discussion merits its own thread.
So, give your suggestion - what might an AI might say to save or free itself?
(The AI-box experiment is explained here)
EDIT: one caveat to the discussion: it should go without saying, but you probably shouldn't come out of this thinking, "Well, if we can just avoid X, Y, and Z, we're golden!" This should hopefully be a fun way to get us thinking about the broader issue of superinteligent AI in general. (Credit goes to Elizer, RichardKennaway, and others for the caveat)