For a more parable-ic version of this, see here.

Suppose I make a precommitment P to take action X unless you take action Y. Action X is not in my interest: I wouldn't do it if I knew you'd never take action Y. You would want me to not precommit to P.

Is this blackmail? Suppose we've been having a steamy affair together, and I have the letters to prove it. It would be bad for both of these if they were published. Then X={Publish the letters} and Y={You pay me money} is textbook blackmail.

But suppose I own a MacGuffin that you want (I value it at £9). If X={Reject any offer} and Y={You offer more than £10}, is this still blackmail? Formally, it looks the same.

What about if I bought the MacGuffin for £500 and you value it at £1000? This makes no difference to the formal structure of the scenario. Then my behaviour feels utterly reasonable, rather than vicious and blackmail-ly.

What is the meaningful difference between the two scenarios? I can't really formalise it.

New to LessWrong?

New Comment
46 comments, sorted by Click to highlight new comments since: Today at 4:02 PM

I think the problem is just equivalent to dividing the gains from trade. We'll look at the second example first, since it is a canonical example of a trade.

I have the MacGuffin which I value at £9. You value it at £1000. Suppose that I give you the MacGuffin, you give me £9 and you put £991 on the table between us. Then we both have the utility we had when we started and we have £991 (the gains from the trade) to divide between us. Yay!

Now look at the first example. This is a canonical blackmail, which I'm going to re-write to make it look as much like the above example as possible.

I have some letters which incriminate us both. I value the letters not getting out at £M, and you value the letters not getting out at £N. Suppose I destroy the letters (rather than releasing them) and put £M on the table between us, while you put £N on the table between us. Then we both have the utility we would have had if the letters had been released but along with this we have £M+N (the "gains from the trade") to divide between us. Yay?

I think the reason why this example seems more blackmailesque is because there is a natural Schelling point for the division of the gains, namely (£M, £N), which corresponds to me destroying the letters without asking for anything. So asking you for money is rude, because I'm greedily going past the Schelling point.

The whole point of the exercise was to create a decision theory that doesn't comply with blackmail. Well, in a division of gains problem a sensible heuristic is to always demand a fair share (where "fair share" vaugely refers to some sort of Schelling point or something, depending on the problem). In the blackmail example the Schelling point is (£M, £N) so the heuristic tells us to demand at least £M, which is exactly refusing the blackmail!

Very well said! I would only add that your point generalizes: the differences between the two cases is the extent to which it has implications for future interaction ("moving the Schelling point"): blackmail-like situations are those where we intuit an unfavorable movement of the point (per the blackmailed) while we generally don't have such intuitions in he case of trade.

My conclusion: Calling something blackmail or extortion is what the decision to defect in an ultimatum game feels like from the inside.

The above definition is approximately 2/3rds serious (in the same way that Hansonian cynicism like "X is not about Y" is about 2/3rd serious). That is, it represents what seems to be a salient point about how humans think about the subject and also about the arbitrariness that is inherit to the problem. While there are technically exceptions to the definition (some people will comply with extortion even when labelling it thus) to me at least it seems to capture a significant aspect of the meaning.

I do note that while I am calling the distinction 'arbitrariy', this isn't a criticism. Arbitrariness is inevitable in cases that inherently require some kind of arbitration and things like obvious shelling points and established norms become invaluable. In some cases the decision to defect in an ultimatum game is obvious while in others it is not and part of the way we go about making the distinction is by perceiving the situations to be inherently different despite formally equivalent payoff structures.

This seems like a satisfactory reduction. By focusing on the analogy with Ultimatum Game, we can capture both the idea of cooperating/defecting, and the idea of evaluating fairness/Shelling point, in a way that follows from the structure of the game, not via references to the informal concepts.

(In my previous attempt to unpack "blackmail", the suggestion was that blackmail is the feeling of the more vague "shouldn't cooperate", as an attempt to get away from the annoying passing-the-buck intuition that it has something to do with the relation of the decision to the status quo or Schelling points (which seem about as confusing as "blackmail" itself). The ultimatum game solution seems to address both concerns, making "defection" less vague and retaining motivation for the "status quo" part of the puzzle.)

[-][anonymous]11y70

In legal jargon, blackmail is threatening to reveal information, usually information that would damage the reputation of the person being threatened, unless that person yields to the threatening party's demands. Extortion and coercion refer to the more general act of levying threats and ultimatums to change someone's behavior. The only common exception I know of to this terminology is nuclear blackmail.

The conceptual analysis edge-case you give is called price gouging. Lawful commercial threats (withdrawing business) can come from customers as well. Threatening to deny a seller your purchase unless offered a better deal is a threat made credible by the existence of alternative sellers and substitute goods.

Just from memory, it seem to me that unlawful, morally reprehensible coercive threats, as made by people rather than software agents, are not often very costly to the person making them, and not anywhere near as costly to the threatener as to the threatenee. For that reason, I don't think "costly to the threatening party if the victim doesn't yield" quite cuts at the joints of "the bad kind of extortion", when talking about human interactions. Some of the reasons why we don't make threats that are costly to ourselves:

We have value uncertainty. If Mr. Extortion Black kidnaps and ransoms the child of Eccentric Richman, the harmful action he threatens won't be "release a plague that eats everyone's feet", because maybe Richman hates his child and his feet.. Or he doesn't hate his feet, but he doesn't disvalue losing them because he can just buy prosthetic ones for himself (or for everyone except Mr. Black).

If I threaten to prove a contradiction and destroy reality should my friend Internet Troll fail to make a sandwich for me, there's a non-negligible chance that he'll tell me to "do it, friend". In this case I'd avoid the threat both because of my uncertainty of his values and because he doubts my commitment or my ability to act.

Blackmail in particular, among kinds of extortion, has been subjected to a long history of legal conceptual analysis. Many legal theorists take the position, contrary to popular opinion across most cultures, that while blackmail is morally wrong, it should not be unlawful, given that we don't police threats or the revealing of secrets separately. "How could threatening to do something legal be a crime?"

Conceptual analysis is difficult. Sometimes it's an incorrect to approach to solving mysterious questions. The best way I know to work with it is to keep a goal in mind that will help determine boundaries of relevant categories, like remembering that you're gathering vanadium when you sort bleggs and rubes and edge cases. So what's your goal?

[This comment is no longer endorsed by its author]Reply

What is the meaningful difference between the two scenarios?

Within the boundaries of the formal structure you've given the difference between blackmail and acceptable negotiation practice is largely a matter of social norms (and the usual complicated mess of framing games, preferences and status). We have no rule that people must sell or buy at the worst possible price that is still positive to them (in my culture at least, there are exceptions). We do have a (justifiable) instinct that revealing shameful secrets about people is an act of aggression and we are inclined to consider aggressive acts made conditional on something else that is not itself aggression to be 'blackmail' or 'extortion'.

That description is supported in my mind somewhat by the observation that people consider things to be blackmail/extortion/threats even when they are mere descriptions of a Best Alternative to Negotiated Agreement. To my best approximation, for most people the evaluation of whether X is bad dominates the formal structure, rather than merely resolving between formally equivalent cases.

What is the meaningful difference between the two scenarios? I can't really formalise it.

Why do you care? In what way is this not merely a decision about how to use a word?

Because we want to design decision theories that are resistant to being blackmailed, but that can get involved in negotiations. If there's no meaningful difference between the two...

A difference between the two scenarios you present is that in the first, the threat makes the blackmailed party worse off than if the threat had never been made, whether they yield to the threat or refuse. In the second, this is not true: rejecting the offer to sell the McGuffin leaves them in exactly the same position as if the offer had never been made.

I believe that is the difference you are looking for: making someone an offer they cannot refuse vs. making them an offer they can refuse.

Precommitting to rejecting an offer makes them worse off than if the precommittment had never been made. The difference seems more a question of degree than of kind.

And we can bring in some extra connotations to make the first scenario better and the second worse. For instance, the blackmail could be "spend an evening talking over this with me, then I'll give you back the letters", while the MacGuffin could be a cure for a universal plague or something:

"I own this bauble that can save your civilization, which will otherwise die. I precommit to rejecting any offer you make for this bauble that is less than 99.9% of the value of your civilization (so about 0.1% of your people will survive). You are, of course, at liberty to refuse."

Precommitment should not be a feature of rational agents; I think that if we can define blackmail in a land of no precommitments, we have a pretty good definition of blackmail.

For instance, the blackmail could be "spend an evening talking over this with me, then I'll give you back the letters"

Still formal blackmail. If the blackmailer would incur a cost from publishing the letters, then the blackmailer would not bother in the world where the blackmailee simply ignores such threats.

"I own this bauble that can save your civilization, which will otherwise die. I precommit to rejecting any offer you make for this bauble that is less than 99.9% of the value of your civilization (so about 0.1% of your people will survive). You are, of course, at liberty to refuse.

This is a problem of dividing the gains from trade. In general we still haven't solved what is a good Schelling point for a fair division. Suppose a fair division would be to pay 50% of the value of your civilization, since the cost to the McGuffin-seller is negligible. Then you tell the McGuffin-seller that you've precommitted not to pay more than 10% of the value of your civilization, exhibiting changed source or something to make the precommitment credible. If the seller is a CDT maximizer, they say "Oh well" and sell at 10% which is the maximizing action from their perspective, since as a CDT agent they are ideologically committed not to take into account that they have caused themselves to be the target of this precommitment by being the sort of agent who would say "Oh well" and sell. It seems quite likely that if there is a 50%-of-value Schelling-point 'fair division' here then the rational action is not to accept any trade over, respectively under, 50% plus epsilon, respectively 50% minus epsilon. This may or may not end up being the same problem as formal blackmail in a completed theory, but it shares some of the same structure where you can exploit the living daylights out of CDT maximizers by moving logically first.

Precommitment should not be a feature of rational agents; I think that if we can define blackmail in a land of no precommitments, we have a pretty good definition of blackmail.

In the absence of precommitments, you can have options available that would play the same role.

This is a problem of dividing the gains from trade.

I haven't seen any theory that clearly divides what is division of gain from trade from what isn't. When two agents sit down with each other, and they both have the possibility of having built, or not having built, various objects that would have various positive or negative values for the other... Where does the blackmail or aggressive negotiations end, and the dividing begin?

Precommitment should not be a feature of rational agents; I think that if we can define blackmail in a land of no precommitments, we have a pretty good definition of blackmail.

Can you elaborate on this (especially the first part)?

[This comment is no longer endorsed by its author]Reply

Precommitting to rejecting an offer makes them worse off than if the precommittment had never been made.

It does not make them worse off than if the negotiation had never been opened.

I don't see where the difficulty is in distinguishing "I have $OBJECT for sale at $MONEY. Interested?" and "Nice operation you've got here, you wouldn't want anything to happen to it, would you? Just see that you do some business with me from time to time, and I can guarantee there won't be any trouble."

I dare say there are some radicals who see all free exchange of value as an act of violence, each party viciously withholding the good they could do for the other in order to extort a price from them, and in a virtuous society all would freely, selflessly work for the benefit of everyone but themselves without even the very idea of a return for their labour. But the world contains all manner of madness.

I don't see where the difficulty is in distinguishing "I have $OBJECT for sale at $MONEY. Interested?" and "Nice operation you've got here, you wouldn't want anything to happen to it, would you? Just see that you do some business with me from time to time, and I can guarantee there won't be any trouble."

The difference there seems to be the default (or disagreement) point: the assumed zero of the transaction. Then any deviation from the default is seen by whether it's positive or negative. It's the difference between "pay your taxes, and the police will protect you from criminals" and "pay your taxes, and the police won't smash up your shop" (or "I can sell you this for money", vs "I can sell you back what we stole from you, for money"). In all these cases, your paying/not paying imply the same futures, but it's different because of the disagreement point.

But establishing disagreement points is a tricky task, and a contentious one. I was hoping there was a difference between threats and offers that didn't involve the disagreement point.

In all these cases, your paying/not paying imply the same futures, but it's different because of the disagreement point.

I’m not sure about the others, but in the taxes/police example, the implied futures in the pay/not pay are not the same:

"pay your taxes, and the police will protect you from criminals" means if you don’t pay, P(shop smashed) = X, if you pay P(shop smashed) << X.

"pay your taxes, and the police won't smash up your shop" means if you don’t pay, P(shop smashed) = X, if you pay P(shop smashed) >> X.

(Note that X is the same for both scenarios. That is, P(shop smashed|taxes not payed) does not depend on which scenario the police chooses.)

The difference there seems to be the default (or disagreement) point: the assumed zero of the transaction.

The zero is not assumed, but objective: the state of things before the negotiation. The blackmailer specifically intends to remove the status quo as an option; the shopkeeper merely adds an option to it. Both parties know exactly what the status quo was. It is not a default, or an assumption, or anything but an objective fact that everyone involved is agreed about. An agreement point, not a disagreement point.

It is not a default, or an assumption, or anything but an objective fact that everyone involved is agreed about.

People are rarely in agreement about what disagreement point is. Especially if various entities have had a long-standing relationship with some changes.

I can't fit this to any of the examples you gave. The Baron comes to the Countess with a threat to publish their correspondence: it is clear to both that he has unilaterally introduced a change to the status quo, to the Countess' detriment. A McGuffin owner comes to a McGuffin collector offering a McGuffin at a fixed price: it is clear to both sides that this has introduced a new option, taking no existing options away. Everyone is in agreement. What situations are you thinking of that make "rare" the clarity of this distinction between threatening to injure and not threatening to injure?

And what about the variant when the winged sandal was going to be given to charity, but the Baron rushed in to prevent that, arriving just in time?

Here it's clear that the Baron still has legal ownership (just!), but that it's the Baron who's changing the status quo.

You could argue that a lot of law is about specifying what the disagreement point is (generally through ownership rules and contract law), but that doesn't mean that our legal system's choice of disagreement point comes from any intrinsic definition that makes sense (see the difficulty with intellectual property).

I rather lost interest in the winged sandal story, but for all the attempted complications, it remains quite clear. The Countess never owned it, and the Baron wants to secure his ownership first before offering it for sale. Whatever this is, it is not blackmail. Engrossing, forestalling, regrating, badgering, or cornering), perhaps, which aren't even illegal any more in English law.

A lot of law is about specifying exact rules. The difficulty of doing so, precisely enough to decide cases, does not imply that there is anything philosophically problematic.

Ok, try my example here:

http://lesswrong.com/r/discussion/lw/i07/semiopen_thread_blackmail/9dt9

What is the status quo there? The black car, or the green car, or just a car (colour unspecified)?

The agreement they had: an explicit stipulation of a car for £100, and a reasonable presumption on both sides that the car would be black. Agent A is breaking the contract by demanding more. This is not a difficult example.

The zero is not assumed, but objective: the state of things before the negotiation.

If before the negotiation, a landslide was already closing on your (uninsured) country house, then after the negotiation the "state of things" is going towards the negative, for reasons unrelated to the negotiation. The question here is about the supposed distinction between that landslide and your opponent's decision algorithm.

If before the negotiation, a landslide was already closing on your (uninsured) country house

I can't find the example this is from.

Can we just use that as the definition?

Define a pre-commitment by you to be blackmail if it makes me wish that I'd pre-pre-committed (and, of course, let you know that I'd pre-pre-committed) to not do the thing that you want in the event that you made that pre-commitment.

How does that do?

EDIT: Thinking about it more, this problem is just division of gains from trade. I'll explain that more in a top-level comment.

That's the problem - every example I've come up with is covered by that definition.

Perhaps I am missing something, this seems to be an accurate model of opportunism of which blackmail itself is a subset. Namely, both examples given are a type of opportunism which the agent exploits the target's desire for a particular value (either reputation or a MacGuffin) for profit, despite the risks of lost value balanced against the target's anticipated value. Blackmail, colloquially, is typically used to denote the above by use of incriminating information (such as steamy letters), while purchasing the MacGuffin may be called any number of things depending on the character of the act.

How about something like this as a definition: There is a "status quo", an outcome for each agent if neither takes any special action, that serves as reference point. Agent B is blackmailing A if s/he announces to A that the status quo is no longer a possible outcome, and the possible outcomes (relative to the status quo) have utilities either "-x for A, +y for B" or "-z for A, anything for B", with z > x. (ETA: typo that said A instead of B corrected in the last line.)

I don't think it is possible to define blackmail without a concept similar to the status quo, as this example I posted in the other thread suggests.

So is robbery blackmail? "The status quo is not to harm or steal. Give me your money, or I will hurt you and take your money." Wouldn't most crimes fall under the same definition?

In fact, isn't the social threat of negative publicity just blackmail against would-be public adulterers? "-affair for A, +prudish satisfaction for Bs" or "-status for A, whatever Bs get out of shaming A". And justice is probably just blackmailing people into being nice against the status quo of natural selection to avoid punishment...

The robbery case is certainly blackmail according to my definition. I agree it sounds strange to call it that way, but I am quite happy to call it "extortion". And I think the distinction between blackmail and extortion in ordinary language (is it that in blackmail the threat must be to reveal information? I'm not sure, not being a native speaker) is slippery and of little theoretical significance.

The other two are not blackmail because the status quo I am talking about is an actual state, vague to define but roughly composed of the reasonable expectations of A before B makes a concrete threat and changes them. It is not a counterfactual "state of nature" in which the public is not scandalized by adulterers and justice does not punish criminals.

Yes, I'm having great trouble defining blackmail without a status quo...

Why is this a problem? I see an obvious status quo in every one of your examples, and if you have to go to such lengths of complication to arrive (or not) at anything doubtful, what does that prove? Look at any border closely enough and it will look blurry. That need not stop us from drawing them.

It's not necessarily a problem, it's an interesting result!

[-][anonymous]11y00

Does a definition with "status quo" look something like this?

"The blackmailer sets up a situation in which the blackmailed has two options {A, B}. Both blackmailed and blackmailer prefer option B to option A. The status quo is a situation C, which the blackmailed prefers to B, whereas the blackmailer prefers B to C."

Buying up a MacGuffin that someone else wants, and offering to sell it for an exorbitant price (B) doesn't really fit here. Presumably the status quo (C) is where the MacGuffin lover doesn't have the object of desire in the first place, so they still prefer B to C, and it's not blackmail. However, stealing the MacGuffin and then offering to sell it back (or destroy it otherwise) does fit the description. This means that "ransom" is a subset of "blackmail", which I think makes sense.

[This comment is no longer endorsed by its author]Reply

What about if I bought the MacGuffin for £500 and you value it at £1000? This makes no difference to the formal structure of the scenario (note that me valuing the MacGuffin at £11 would change the formal structure: then I might do X even if I knew Y was impossible: I would still reject an offer of £10). Then my behaviour feels utterly reasonable, rather than vicious and blackmail-ly.

I suggest moving the parenthetical to the end of the paragraph. I had to re-parse that in programming mode before I interpreted the ambiguity in the last sentence correctly and I wouldn't have noticed my confusion at all were I not familiar with the distinctions you are trying to express.

Thanks for the comment - I simply removed the parenthetical, it's not adding much useful in such a short post.

But suppose I own a MacGuffin that you want (I value it at £9). If X={Reject any offer} and Y={You offer more than £10}, is this still blackmail? Formally, it looks the same.

One big difference I see is what counts as an "action"; arguably X={Publish the letters} "feels" more like an action than X={Reject any offer}. It seems that when what distinguishes a "blackmailish" offer from a "non-blackmailish" one is a discrete distinction (either you precommit to publish or you don't), making the "blackmailish" offer is "bad", but when there's a continuum between the most and least blackmailish versions (as is the case with the MacGuffin negotiation), the offer doesn't seem to be as much of a transgression.

I'm not totally happy with this tho, and to me the general problem of "negotiation in situations with no focal points" has no satisfying solution beyond racing to who precommits first.

I'm not totally happy with this tho, and to me the general problem of "negotiation in situations with no focal points" has no satisfying solution beyond racing to who precommits first.

That trick works against CDT agents, anyway. Against agents that implement a modern decision theory it leaves the negotiation problem approximately where it started. (This means that if you put ideal CDT agents and ideal TDT agents in a room with devices that offer similarly ideal pre-commitments and give them exchanges to make the CDT ends up accepting the worst possible positive trade.)

Why not taboo 'blackmail'? That word already has a bunch of different meanings in law and common usage.

I'm trying to define threat/blackmail or similar concepts in decision theory. In the two examples above, one seems a clear negative situation, the other doesn't, and I can't figure out what the difference is.

You need a different definition for 'blackmail' then. Action X might be beneficial to the blackmailer rather than negative in value and still be blackmail.

The whole point of this post is to find a formal definition of "something like blackmail" (maybe "threat" instead of "blackmail" would have been a bit better, but "threat" also means a lot of other different things).

I agree that maybe starting with an informal definition of "what is meant by 'blackmail' here" may have been better.

A problem with blackmail is that it can be repeated. After you do Y (pay me money) and I don't do X (publish letters), I can go back and once again ask you for money. With your second example, after not rejecting the offer I no longer have the MacGuffin, so I can't repeat.

EDIT: The only way to rule out repetition is to include it in the precommitment. But note that for human agents it is impossible to precommit in an enforceable way without side-effects. They can't self-modify, so they would need a trusted third party. Then this trusted party will get to know at least the fact that there are some juicy secrets out there about Y.

But if I give you the letters in exchange of the money, this doesn't make it any less blackmail...

And having extra MacGuffin's to sell shouldn't change it, either.

You could have made copies of the letters.

And after selling you one, I will still have one less MacGuffin. Presumably these can't be copied easily. And if they can, e.g. if it is a digital file, then what is it that you really want to get? Access to the information in the file or preventing access to the information for others?

This made me think of some related concepts: NDAs and exclusive rights contracts.