Semi-open thread: blackmail

I agree, it just sort of accumulated in this way. Can threads be merged?

(Please note that in the following, I'm using "blackmail" for all sorts of attempted coercion or extortion)

I'd say this formulation of yours is very useful for drawing some boundaries around the word "blackmail". Namely, the cost to the blackmailer in case the threat fails should be somewhat less, but at least comparable than the cost to the blackmailee in case the threat succeeds, or the threat will be simply viewed as stupid (although you could, technically, still call it "blackmail").

Or rather, this assumes a probability of success of 0.5. If this probability (as judged by the blackmailer, and historically checkable) is different, the threat and costs imposed on both blackmailer and blackmailee also have to change accordingly.

For example, whole governments are known to work on a no-blackmail-basis, as officially announced. For example, when hostages are taken, government officials (at least in my country) frequently announce that their country will not let themselves be blackmailed. So in order for a threat to succeed, according to my model the costs to the threatener must be very low (e.g. they have safe refugee or asylum in a third country, and don't have to fear retaliation in case the threat fails and they have to act on it) and the costs to the country or government (incremental over succumbing to the threat) very high (e.g. the leaking of diplomatically relevant documents which can seriously damage international relations).

In your example from above, I'd argue that the threat (with $150 green paint) has a chance of succeeding if and only if:

• (a) The customer has been successfully blackmailed many times before, and this is common knowledge and/or
• (b) The customer really needs a black car because of some obscure reason, and only this car, only in black, will somehow give him huge profits. This is also common knowledge.

The question is whether one can reason about why this is so, utilizing some form of TDT or -derivative.

Blackmail. A wouldn't make the threat if A believed that B ignored such threats, and A would have no motive to paint the car green if B were a rock. Pricing of green makes no difference so long as the price is not negative.

Blackmail or not?

Yes.

And would it be different if the price of green was high, but not so extreme?

No. (At least, only marginally.)

What is the reason that Agent A asks for an extra £10? Is it because he had a bucket of green paint laying around that is about to go bad, and he doesn't want to spend £9 to drive to the store to buy black paint? Or does he believe he can get away with demanding more since the contracts needs to get amended?

That's more theft than blackmail. When B refuses to do laundry, B is acting on the intrinsic value of not doing laundry. That's different from the car paint example, since threatening to paint the car green is not based on the intrinsic value of painting the car green.

I think I agree with the status quo point.

You and me live in the open desert, and our dear mother is trapped there (assume some mythology).

No mythology necessary, actually. Two daughters and an aging mother. Which of the two gets to marry and move away from home, and which is left behind? My mother, her sister, and their mother. All dead now, so I can't ask any of them if there was any conflict over it.

This is not something I would call blackmail, any more than I would call theft blackmail. Fait accompli: one of the two simply does something to the disadvantage of the other. It may of course be morally wrong, but there are many varieties of wrong, only some of which are blackmail.

But this is just a terminological question. The real question here is presumably: how can a rational agent be designed not to be exploitable by fait accompli?

What if the problem were entirely symmetric, and we both had exactly the same preferences. One of us has to stay, one has to go. Is it blackmail if I go first, and thus force you to stay?

Well, If I have a chance to react to that before your decision is final, I could say that if you do that then I'll go anyway (Thus leaving mother to burn) Then it seems to resolve to Chicken. (We both want the other person to back down first, so that we can go all the way to Olympos, but if neither backs down, mother burns in desert heat.)

And if I don't have a chance to react to that before your decision is final, then it's sort of like "What do you do if you are playing Chicken against someone who has already smashed his steering column with a rock?"

And I think that makes sense, since in both cases (blackmail, chicken) the goal is make the person back down because of some sort of threat. Does that help?

As I said elsewhere, I think identifying blackmail depends on identifying an agreed status quo state. This is fairly easy in ordinary life (if we had an affair, the ordinary human expectation is that you would keep the letters secret for free, etc.), but more difficult in outré cases like this one and the other one you suggest with the letters in the cave. It is not surprising therefore that in cases like this we do not have clear intuitions.

That said, following the definition I proposed, I'd say that in your first example if it is a plausible status quo that your brother would go to Olympos (maybe he got a job offer letter form there?) then it is blackmail for you to go instead. If he just happened to come up with the idea of leaving, then my intuition is that this is not a plausible status quo; there is no stability, no ordinary human expectation that you would abide by his decision of leaving you stranded. A more plausible status quo (which applies also to the perfectly symmetric case when both come up with the idea together) would be flipping a coin to decide who leaves. Relative to this, either of you leaving is blackmailing the other.

This wasn't a comment thread - it was supposed to be for people to submit examples of blackmail.

I'm very aware this wasn't the best organisation! I wrote the long post, then realised a summary could work, then thought maybe people would like a separate example thread... :-(

I apologise.

1. What does "an action that harms another agent" mean? For instance, if I threaten to not give you a chicken unless you give me $5, does "I don't give you a chicken" count as "a course of action that harms another agent"? Or does it have to be an active course, rather than act of omission?

2. Is it still blackmail if it's "justified"? For instance, if you steal me car, and I threaten to call the police if you don't give it back, is that blackmail?

Harming the blackmailer is simply a proxy for "this isn't a course of action the blackmailer would otherwise engage in; they do it only for its blackmail effect"

I own the land the cave is on, and refuse to let anyone fix the cave, unless you pay me. Blackmail?

Yes.

What about if the only reason the cave entrance was crumbling was because I diverted a river. I didn't do this with the intent of causing the entrance to crumble (I have other reasons), but I was fully aware of this side effect, and let it happen. Blackmail?

No. Assuming that protecting the cave is free (because someone else wants to come and do it) the diverting of the river is not itself blackmail. Subsequent blackmail is still blackmail. It becomes close to blackmail if the cost to repair the cave is greater than the value of the river diversion. Depending on the various value instantiations possible this example does a good job of exemplifying the 'blackmail vs gains to trade' blurry boundary.

What if I approached you before I diverted the river, and asked for money then. I still want to divert the river, but if you pay up enough, I will desist from it. Blackmail?

No, assuming the right values for various outcomes this is exactly what "split the gains from trade equally" looks like. The inconvenience for keeping the secret is divided evenly. Another excellent example because to most humans this will feel like extortion and if the numbers are tweaked a little it could be extortion by my standards too.