On the subject of blackmail and precommitments, I find the third season of 24 interesting (spoilers below):
The president’s ex-wife is at the house of an influential power broker who is standing in the way of the president. While there, the power broker has a heart attack. The ex-wife convinces the power broker’s wife to not give him his pills, and the power broker dies. (The two women act like they have committed a crime, but I’m not sure that they have, but that’s a bit of a separate issue). The ex-wife then goes to the president and gets his help in covering the incident up. Now, the ex-wife wants to blackmail the president, but cannot make any credible precommitment to tell what she knows; while the president would be in trouble if this gets out, she would be in far worse trouble. So instead she sells a crucial piece of evidence to the candidate running against the president. She gets a payoff, and the candidate can make a credible threat to release the evidence if the president doesn’t withdraw from the race; it’s not even exactly a “threat”, since if the president doesn’t comply, the candidate’s interests will be served by releasing the evidence. So the president will almost certainly comply, and if the president complies, then the candidate no longer has a motivation to release it, so there isn’t much risk on the part of the ex-wife.
This is an excellent example of blackmail that avoids the blackmail equation payoff-matrix. In this case both the interactions are mutually beneficial exchanges. In the case where each of the three is a CDT agent this strategy this allows the ex-wife to overcome her inability to credibly threaten.
By my analysis if the president is a TDT agent and the usual mutual knowledge of decision strategies is present then he will refuse to pay the blackmailer. The rival would of course still want to purchase the evidence because it remains valuable to him and he will still release it because it is beneficial to him to do so. However the ex will not be willing to sell it because she knows that doing so will inevitably result in her incarceration. So this example helps narrow down just what “that thing related to doing stuff others don’t like which sane decision theories are not vulnerable to” is. Even things that rely only on mutually beneficial exchanges (when everyone is CDT) can still qualify.
So the president will almost certainly comply, and if the president complies, then the candidate no longer has a motivation to release it, so there isn’t much risk on the part of the ex-wife.
There is no risk of the information being released (given certain assumptions). However there is a significant risk of going missing without a trace some time in the next year or two, never to be seen again. This kind of combination of entrapment, deceit and betrayal is the kind of thing which provokes vengeance. Vengefulness is a rather common attribute among humans and in this case I would even argue that it is a rational disposition to have. Being the kind of person who kills people who pull that sort of stunt against them has clear benefits. Especially when the killing could likely be accomplished with one phone call to an extraordinarily resourceful friend.
I’m reminded of another character who relies on “It is not in your immediate self interest to harm me therefore I am safe” one too many times. Even Jack, who is one of the most ruthlessly practical characters in all of fiction, just shoots her dead. And Jack is someone who would definitely kill babies if it was the right thing to do (all the babies, for epsilon rightness). This is the kind of response that should be expected, and not just because “humans are irrational”. It should be expected in some cases from even idealised rational agents.
On the subject of blackmail and precommitments, I find the third season of 24 interesting (spoilers below):
The president’s ex-wife is at the house of an influential power broker who is standing in the way of the president. While there, the power broker has a heart attack. The ex-wife convinces the power broker’s wife to not give him his pills, and the power broker dies. (The two women act like they have committed a crime, but I’m not sure that they have, but that’s a bit of a separate issue). The ex-wife then goes to the president and gets his help in covering the incident up. Now, the ex-wife wants to blackmail the president, but cannot make any credible precommitment to tell what she knows; while the president would be in trouble if this gets out, she would be in far worse trouble. So instead she sells a crucial piece of evidence to the candidate running against the president. She gets a payoff, and the candidate can make a credible threat to release the evidence if the president doesn’t withdraw from the race; it’s not even exactly a “threat”, since if the president doesn’t comply, the candidate’s interests will be served by releasing the evidence. So the president will almost certainly comply, and if the president complies, then the candidate no longer has a motivation to release it, so there isn’t much risk on the part of the ex-wife.
This is an excellent example of blackmail that avoids the blackmail equation payoff-matrix. In this case both the interactions are mutually beneficial exchanges. In the case where each of the three is a CDT agent this strategy this allows the ex-wife to overcome her inability to credibly threaten.
By my analysis if the president is a TDT agent and the usual mutual knowledge of decision strategies is present then he will refuse to pay the blackmailer. The rival would of course still want to purchase the evidence because it remains valuable to him and he will still release it because it is beneficial to him to do so. However the ex will not be willing to sell it because she knows that doing so will inevitably result in her incarceration. So this example helps narrow down just what “that thing related to doing stuff others don’t like which sane decision theories are not vulnerable to” is. Even things that rely only on mutually beneficial exchanges (when everyone is CDT) can still qualify.
There is no risk of the information being released (given certain assumptions). However there is a significant risk of going missing without a trace some time in the next year or two, never to be seen again. This kind of combination of entrapment, deceit and betrayal is the kind of thing which provokes vengeance. Vengefulness is a rather common attribute among humans and in this case I would even argue that it is a rational disposition to have. Being the kind of person who kills people who pull that sort of stunt against them has clear benefits. Especially when the killing could likely be accomplished with one phone call to an extraordinarily resourceful friend.
I’m reminded of another character who relies on “It is not in your immediate self interest to harm me therefore I am safe” one too many times. Even Jack, who is one of the most ruthlessly practical characters in all of fiction, just shoots her dead. And Jack is someone who would definitely kill babies if it was the right thing to do (all the babies, for epsilon rightness). This is the kind of response that should be expected, and not just because “humans are irrational”. It should be expected in some cases from even idealised rational agents.