I recently played and won an additional game of AI Box with DEA7TH. Obviously, I played as the AI. This game was conducted over Skype.
I’m posting this in the open thread because unlike my lastfewAI Box Experiments, I won’t be providing a proper writeup (and I didn’t think that just posting “I won!” is enough to validate starting a new thread). I’ve been told (and convinced) by many that I was far too leaky with strategy and seriously compromised future winning chances of both myself and future AIs. The fact that one of my gatekeepers guessed my tactic(s) was the final straw. I think that I’ve already provided enough hints for aspiring AIs to win, so I’ll stop giving out information.
The fact that one of my gatekeepers guessed my tactic(s) was the final straw.
I guess you used words. That seems to be all the tactical insight needed to develop an effective counter-strategy. I really don’t get how this escaping thing works on people. Is it due to people being systematically overconfident in their own stubbornness? I mean I know I couldn’t withstand torture for long. I expect even plain interrogation backed by credible threats would break me over time. Social isolation and sleep deprivation would break me too. But one hour of textual communication with a predefined and gamified objective and no negative external consequences? That seems so trivial..
Other people have expressed similar sentiments, and then played the AI Box experiment. Even of the ones who didn’t lose, they still updated to “definitely could have lost in a similar scenario.”
Unless you have reason to believe your skepticism comes from a different place than theirs, you should update towards gatekeeping being harder than you think.
Unless you have reason to believe your skepticism comes from a different place than theirs, you should update towards gatekeeping being harder than you think.
Unless I have already heard the information you have provided and updated on it, in which case updating again at your say so would be the wrong move. I don’t tend update just because someone says more words at me to assert social influence. Which is kind of the point, isn’t it? Yes, I do have reason to believe that I would not be persuaded to lose in that time.
Disagreement is of course welcome if it is expressed in the form of a wager where my winnings would be worth my time and the payoff from me to the gatekeeper is suitable to demonstrate flaws in probability estimates.
AI Box Experiment Update
I recently played and won an additional game of AI Box with DEA7TH. Obviously, I played as the AI. This game was conducted over Skype.
I’m posting this in the open thread because unlike my last few AI Box Experiments, I won’t be providing a proper writeup (and I didn’t think that just posting “I won!” is enough to validate starting a new thread). I’ve been told (and convinced) by many that I was far too leaky with strategy and seriously compromised future winning chances of both myself and future AIs. The fact that one of my gatekeepers guessed my tactic(s) was the final straw. I think that I’ve already provided enough hints for aspiring AIs to win, so I’ll stop giving out information.
Sorry, folks.
This puts my current AI Box Experiment record at 2 wins and 3 losses.
I guess you used words. That seems to be all the tactical insight needed to develop an effective counter-strategy. I really don’t get how this escaping thing works on people. Is it due to people being systematically overconfident in their own stubbornness? I mean I know I couldn’t withstand torture for long. I expect even plain interrogation backed by credible threats would break me over time. Social isolation and sleep deprivation would break me too. But one hour of textual communication with a predefined and gamified objective and no negative external consequences? That seems so trivial..
Other people have expressed similar sentiments, and then played the AI Box experiment. Even of the ones who didn’t lose, they still updated to “definitely could have lost in a similar scenario.”
Unless you have reason to believe your skepticism comes from a different place than theirs, you should update towards gatekeeping being harder than you think.
The heuristic of ignoring secretive experiments that don’t publish their details has served me well in the past.
I have played the game twice and updated in the opposite direction you claim.
In fact, my victories were rather trivial. This is despite the AIs trying really really hard.
Did you play against AI that do have won sometime in the past?
I do not honestly know. I will happily play a “hard” opponent like Eliezer or Tux. I have said this before, I estimate 99%+ chance of victory.
Unless I have already heard the information you have provided and updated on it, in which case updating again at your say so would be the wrong move. I don’t tend update just because someone says more words at me to assert social influence. Which is kind of the point, isn’t it? Yes, I do have reason to believe that I would not be persuaded to lose in that time.
Disagreement is of course welcome if it is expressed in the form of a wager where my winnings would be worth my time and the payoff from me to the gatekeeper is suitable to demonstrate flaws in probability estimates.
Probably you’re right, but as far as I can tell the rules of the game don’t forbid the use of ASCII ar
Just so long as it never guesses my fatal flaw (_).