I was not aware of Tuxedage’s ruleset. However any ruleset that allows for the AI to win without being explicitly released by the gatekeeper is problematic.
If asd had won due to the gatekeeper leaving it would only have demonstrated that being unpleasant can cause people to disengage from conversation, which is different from demonstrating that it is possible to convince a person to release a potentially dangerous AI.
I kind of agree upon reflection. Tuxedage’s ruleset seems tailored for games where there is money on the line, and in that case it feels very unfair to say GK can leave right away. GK would be heavily incentivized to leave immediately, since that would get GK’s charity a guaranteed donation.
The more natural option seems to be to treat that as a draw. The AI’s not getting out if you leave the conversation, but there’s not much point in going to the trouble of building an AI if you’re not going to talk to it.
I’ve always thought the gatekeeper should have a ‘shutdown’ option that results in both the gatekeeper and the AI losing money (but less loss for the gatekeeper than releasing). That should make verbal abuse strategies a good deal harder.
In Tuxedage’s rule set, if the gatekeeper leaves before 2 hours, it counts as an AI win. So it’s a viable strategy. However ---
I am sure that it would work against some opponents, but my feeling is it would not work against people on Less Wrong. It was a good try though.
I was not aware of Tuxedage’s ruleset. However any ruleset that allows for the AI to win without being explicitly released by the gatekeeper is problematic.
If asd had won due to the gatekeeper leaving it would only have demonstrated that being unpleasant can cause people to disengage from conversation, which is different from demonstrating that it is possible to convince a person to release a potentially dangerous AI.
I kind of agree upon reflection. Tuxedage’s ruleset seems tailored for games where there is money on the line, and in that case it feels very unfair to say GK can leave right away. GK would be heavily incentivized to leave immediately, since that would get GK’s charity a guaranteed donation.
The more natural option seems to be to treat that as a draw. The AI’s not getting out if you leave the conversation, but there’s not much point in going to the trouble of building an AI if you’re not going to talk to it.
I’ve always thought the gatekeeper should have a ‘shutdown’ option that results in both the gatekeeper and the AI losing money (but less loss for the gatekeeper than releasing). That should make verbal abuse strategies a good deal harder.