The Devil You Know, Part 1

will be worse than the known risks. And we'll get to some concrete examples and foundational research on this in a bit, but to an approximation, a good way to think about this is that ambiguity aversion is the sentiment expressed in the saying, better the devil you know than the devil you don't. This is an aphorism I feel like people often quote when they're about to justify making a really bad or indefensible decision. And the whole point of this saying is that the devil you know is

already bad. That's usually implied, and there's no indication that the devil you don't know will be worse. It could well be a kinder, gentler devil with a less pointy pitchfork. But still we often find ourselves thinking, I'd just rather stick with the one where I know how pointy it is upfront.

your stay in hell? But also I think it does have a kind of wisdom if you don't think of it as advice, but rather as an ironic observation about human nature, in which case I think it's sort of dead on this This is often how we think, and that is the core observation also of the ambiguity aversion literature.

you know, fall for tyranny. This is because they have these moments of chaos and uncertainty and they're like, well, I would just rather have this sort of known top down oppression in place rather than some sort of ambiguous chaos bubbling up around me.

and thus unpredictable to you? So it seems like, I don't know, it just seems kind of fraught, like a hassle, And instead of doing that, choosing to have an experience that you pretty much know that you will probably not enjoy.

aligned with those kinds of quests. But on the other hand, I have to recognize that like most of the really fulfilling things in my life have been because I said yes to adventure and because I took a chance on something and I went out of my comfort zone. But then again, we call it our comfort zone because it is comfortable. It is a nice place, it is a comforting place. So it's you know, it's given take.

Elsberg and this paper's impact. It was also reading a paper called Ambiguity and Ambiguity Aversion in the Handbook of the Economics of Risk and Uncertainty. This is by Mark J. Maschina and Marciano Cinneskalchi, published by north Holland in twenty fourteen. This is a sort of general overview of the concept of ambiguity aversion and a survey of a bunch of different models of it. Now, before we get into the idea that Elsburg explored in this very important paper, robbed it.

You had a bit of bio in Elsburg, didn't you.

known as the Pentagon Papers. This is a United States Department of Defense history of the US's political and military involvement in Vietnam from nineteen forty five through nineteen sixty eight.

the publication regarding the Elsberg paradox. And yeah, they detailed secret developments about the scope of the US military efforts in Vietnam, the real sticking points being the deception involving the Gulf of Tonkin incident, unreported expansion of the war, the true objective of the war post nineteen sixty five, which basically was more about containing China and preventing a humiliating US defeat rather than any specific objectives to help the people of Vietnam, and the lack of a clear

plan for victory as well. So I mentioned a lot of you have at least heard of the Pentagon papers, even if you're not familiar with the full story, because this incident this led to conspiracy and espionage charges against Ellsberg that were then later dismissed, greatly fueled the anti war movement as well as distrust in the government at the time, and led to the Nixon Whitehouse's use of

the infamous White House Plumbers to discredit Elsberg. And of course, all of this ends up feeding into the Watergate scandal, which is another one of those major historical touchstones from the time period that again, if you're not super familiar with Watergate, you've at least heard of Watergate, and you've heard about the other different gates out there that allude back to what a scandal it was.

support of various other whistle blowers. He was a vocal opponent of the use and the threat of nuclear weapons and the rhetoric around their use, and he is sometimes described as a nuclear war planner turned nuclear war activists, and his work at the Rand Corporation did involve such contemplations. Yeah.

There were simply too many variables human, technological, situational that could lead to not only catastrophic but existential unintended consequences.

Imagine that there's an urn like a jar that is filled with ninety little ping pong balls that are a mix of different colors. You cannot see inside the urn, it's opaque, But you're going to play a game where you reach into the urn and you pull out one ball at a time, and you can win prizes based on the color of the ball that you pull out. You know that thirty of the balls inside the urn are red, So since thirty out of ninety are red, there's a one in three chance that a random ball

is going to be red. The odds there are pretty clear. The other sixty balls, however, are a mix of yellow and black. And here's the really important part. You don't know the composition of that group. So those sixty balls could be thirty yellow and thirty black. It could be sixty yellow and zero black. It could be sixty black and zero yellow, or any mix in between. Since you don't know the mix, you don't know the probability of drawing a yellow ball or a black ball ahead of time.

For red, it's one in three. For either yellow or black, it's anywhere between zero and two out of three. Now, imagine it is your turn to draw a ball out of the urn, and you can take one of two bets. Bet a you get one hundred dollars if your ball is red, So if you pick this option, your chance of winning one hundred dollars is one in three. In bet B you get one hundred dollars if your ball

is black. Since you don't know the mix of yellow and black, you don't know your odds of winning, but they are somewhere between zero and two out of three, with the midpoint being one out of three, which is the same as the odds on red. Which one do you pick?

that's just what we think we should do. That in itself is kind of weird and interesting, like why do we prefer to avoid ambiguity even when there is no clear advantage in doing so. Then there's a really interesting wrinkle. Imagine you're playing the same game. You are drawing a ball out of an urn, and this time you're given assurance that the mix of the balls is the same

as it was the first time. So whatever was the case with the mix of yellow and black in there, it's going to be the same as it was with your last draw. And you get a chance to make a second bet. You can take either bet C, where you get one hundred dollars if the ball drawn out is either yellow or red, and this one has an unknown chance of winning, because of course you don't know

how many yellow or black balls there are. You know that your chance of winning is somewhere between one and three if there are you know, the thirty red balls and zero yellow balls, or one hundred percent if there are sixty yellow balls, Or you can take bet D where you get one hundred dollars. If the ball is either yellow or black, this one has a guaranteed two thirds chance to win, because we know that together yellow and black make sixty of the ninety balls that we

don't know what the mix is. In this second trial, people will tend to pick bet D once again, because the probabilities are known. With D, it's a guaranteed two thirds chance to win. With bets C, it's somewhere between one one in three and one hundred percent. Picking bet D allows them to avoid placing bets C, which, again, which has ambiguous chances. Here's the really interesting thing that Elsberg pointed out. There's actually a contradiction in people's behavior

between the first and second round of this game. Assuming you are actually doing your best to win the money. In round one, if you bet on red instead of black, that implies you believe that less than thirty of the unknown sixty balls are black. Right, That makes you it implies that red has a better chance of winning, so less than half of them have to be black. However, in the second round, if you take the second option picking yellow and black, it implies that you think that

more than thirty of the unknown balls are black. Otherwise the red and yellow combination option would have the better chance of winning. Right, These two bets in combination make no sense because they imply self contradictory beliefs. It says that I believe that more balls are black and less balls are black at the same time. Obviously that can't

really be the case. But there could be another issue driving this behavior, which is simply that people don't like ambiguity, and they will pay up for known odds, even if doing so implies mutually exclusive assumptions. I was trying to think, you know, as I said a minute ago, I think I would probably pick red if given the option to

play this game. Obviously, I sort of already knew the deal here, so I can't know what I would do going into it blind like most people, but I think I would, for no rational reason, prefer the uncertainty reducing options.

I think the most likely reason for that is even if I were given assurances that this game was on the up and up, I would suspect some kind of trick, Like if I'm being asked to bet on uncertainty, There's some part of me that just kind of rises up and says, uh, there's a scam here that you are not seeing.

other people they put his name on it. Anyway, Elsberg's paper about ambiguity aversion was not just about observing a weird little quirk of human behavior. It was especially of interest to the fields of economics and decision theory modeling how people make decisions, because it sought to undermine some of the major underpinnings of the fields, specifically providing evidence against some of the tenets of a framework called subjective expected utility theory or savage as axioms, named after the

American mathematician Leonard Savage who formulated them. Subjective expected utility theory or SEU theory is a way of modeling how people make decisions when we don't have perfect or complete information, which, of course in life we rarely do. Right, So, you know this is going to be relevant in understanding how people behave in most real world situations. Most situations in life are not a casino table game where you have no odds of winning.

Maybe you've played through the game before. You have something at least approaching perfect knowledge of the simplified world, and our real world is not so simplified and there's so many variables to it.

likelihood of different outcomes are. Subjective expected utility theory has several rules, but a simplified version is that it says, and then when people don't know the objective likelihood of an outcome, they form beliefs about how likely that outcome is, and then they act consistently with those beliefs in order to maximize their personal benefit. I came across a passage by the American economist Frank Knight which I think summarizes

this idea well. Night writes, quote, we must observe at the outset that when an individual instance i e. A one time event only is at issue, there is no difference for conduct between a measurable risk and an unmeasurable uncertainty. The individual, as already observed, throws his estimate of the value of an opinion into the probability form of a successes in B trials a slash B being a proper fraction, and feels toward it as toward any other probability situation.

So we essentially we form a belief for a feeling about how likely something is, and then we act as if that was like you know that we know the odds of a coin flip coming up heads are one in two, whether or not we're actually right about those feelings or beliefs. Elsberg used the principle of ambiguity aversion to say, actually, no, people do not always behave as if they have consistent beliefs about what is more or

less likely to happen. And you can prove this because they make bets that would imply self contradictory beliefs in the same situation, meaning they either don't actually have or act on consistent beliefs about a situation, or they don't always act to maximize their own monetary benefit. Elsberg's point here was that we can have obscure previously unacknowledged motivations like the desire to avoid dealing with ambiguity, and that particular desire is so strong in some cases that it overrides.

It's our consistent thinking and a simple dollar value understanding of rational self interest. And by the way, the technical name of the savage axiom that Elsberg was attacking here was it's called the sure thing principle. Another thing to understand about Elsberg's paper was that he was arguing that we need to make a stronger distinction between two different concepts. Those are risk and ambiguity. Elsberg gives the definition that risk is when we have a stake in an outcome

and we know the theoretical likelihood of that outcome. So this is like the casino games or a coin toss. You know, a fair coin toss is fifty to fifty. We know what the outcome likelihood is, and we can take a risk. A fair die roll is one in six to get a particular number. Ambiguity is when we don't have an objective way of knowing what the probability

of our desired outcome is. To Elsberg, ambiguity was quote the the nature of one's information concerning the relative likelihood of events, a quality depending on the amount, type, reliability, and unanimity of information, giving rise to one's degree of confidence in an estimation of relative likelihoods. And so it's worth noting that you can have confidence or even what feels like certainty, or act as if you have confidence or certainty without actually being correct. You can go through

life having high confidence in objectively low probability things. One of the actually really useful things about SEU theories, the subjective expected utility theory, was that it made things simple and easier to understand because it treated risk and ambiguity the same. So a person making bets on a die roll will know that the probability of rolling a specific number is one in six, and they will behave accordingly.

And a person who has a stake in an outcome with an unknown probability will, according to Savage's axioms, mentally even subconsciously form an internal belief about the probability of that outcome. So example is you imagine there is a one in four chance it's going to reign today. That might not have anything to do with reality, but you just take that belief on and you behave as if that is that you make all of your decisions based

on that belief. They will act according to that belief the same way they would act to knowing that a coin flip is fifty to fifty, and Elsberg argued, no, that is not how we make decisions. SEU theory with savagees axioms failed to account for other complexities in human behavior, for example, our deep desire to get away from ambiguity, Like we don't like ambiguity so strongly we will make

decisions that imply self contradictory beliefs just to avoid it. Now, there have been a lot of arguments over the years about how to interpret ambiguity aversion. Is it actually rational in ways that were not previously recognized, or is it more like some kind of systematic error or cognitive bias. We might come back and revisit that debate in a subsequent part of this series, but we've got a couple

other things we wanted to talk about today. First, Rob, I know you have a really interesting place you want to take this, but first I wanted to mention the idea just of experimental confirmation, because one important thing to understand is that Ellsberg's original paper did not have an experimental component. It was a thought experiment, and he realized the phenomenon needed to be tested with real human subjects.

In the decades since his paper, it has been tested many, many times, many different ways, and generally it has proved robust. There are some nuances and exceptions, but generally people behave the way Elsberg predicted. They would a majority of people structure their choices to void ambiguity, even when there's no clear objective benefit to doing so, and they even make

bets that would imply against self contradictory beliefs. According to SEU theory, if those bets can get them out of dealing with ambiguous probabilities, so I don't have to mess with that. Regarding those experiments, my main source here is again that survey article by Machina and Siniskalchi. Again that's ambiguity and ambiguity of version from twenty fourteen. The authors here list a whole bunch of experiments where people carried out versions of like the three color urn or this

other experiment that Delsberg described, the two urn experiment. They list Felner in nineteen sixty one, which found that people would rather take fifty to fifty odds than unknown odds, which could be better or worse. They also cite Becker and Brownson sixty four, m. Kremen sixty eight, Slovak and Taverski from nineteen seventy four, Curly and Yate from eighty nine, among others, all generally finding confirmation of Elsberg's predictions about

ambiguity aversion quote. Although most of these experiments used students as subjects. Researchers such as McCrimmon sixty five, Hogarth and Kunruther eighty nine, Einhorn and Hogarth eighty six, Viscusi and Chessen ninety nine, hoe Keller and Celtica two thousand and two, and Mafioletti and Sentori two thousand and five have examined the ambiguity preferences of business owners, trade union leaders, actuaries, managers, and executives with the same overall findings, so it appears

quite robust. The authors here also report some interesting findings of McCrimmon and Larsen from seventy nine which they looked into it, and they found that while the majority of people did try to avoid ambiguity, some minority of people did not and even just chose to embrace it. They also found, as you might expect, that our relative tolerance for ambiguity went up or down depending on how good

the known odds were in the known odds bet condition. So, for example, if you do this three color earn experiment, if you take the number of red balls that's the ones where you know how many there are, If you take that down to zero or even just down to five or ten, most people will take the gamble with the unknown odds on yellow and black because the odds

on red are clearly very bad. If you make the odds on red really good, even more people will stick with the no nods, and ambiguity of version becomes even stronger. So that shouldn't be all that surprising. But yeah, in the middle category, where the odds on red are somewhere in between good and bad, you're just kind of like, I don't know, But still people tend to go for

the known odds rather than the unknown odds. However, the authors also found that even when the idea of subjective expected utility and choice consistent see was explained to the people doing this experiment explained to the subjects, or even in cases where the known odds were bad, some subjects just stubbornly stuck to red and avoided betting on any ambiguous conditions. So most of us seem to dislike ambiguity in the majority of cases, and some of us just

don't like it at all and will not tolerate it. Now, as is often the case with stuff and decision theory and economics and stuff, you start off playing these little like gambling games with like balls in a jar or something like that, and so it seems like, well, how consequential could this really be? But actually, I think you can take the idea of ambiguity aversion to number one.

It has been applied to very consequential situations in real life, and thus understanding it can have major consequences not only for people's individual lives, but for world events.

on all of this. But you know, in that book the subject of ambiguous data comes up numerous times how will a nuclear power react in response to ambiguous data about a potential incoming attack? And additionally, it talks about the purported advantage of ambiguity in intense slash response, namely in the case of mad Man theory. This is very closely associated with Richard Nixon, but also tied to various other figures domestic and international, including the current US president.

A kind of supposed perceived madness or volatility that would lead adversaries to second guests any plans to move against them.

other side of a bargaining table. This goes back a long ways. Machiavelli wrote about it in fifteen seventeen, and it's probably as old as the first prehistoric human to like go around hitting trees randomly with a club, you know, and say, who knows what tharg is going to do? Why would you mess with throg on this? We?

a winner in mutually sured destruction. So if the White House was trying to get the Kremlin to do something like I'll nuke if you don't do it, that was not a credible threat because bo both parties knew that the other party knows that war would be the destruction of them both. So no rational actor would ever strike first. Between rational actors, I'll knew Q if you don't do

XYZ is an empty threat. The goal of madman theory is to gain leverage in negotiations by introducing some amount of fear that your counterparty may not be a rational actor. He might be crazy enough to do it, we don't know. Therefore, the madman strategy makes direct use of the opponent's ambiguity

version as a tactic to get leverage over them. The madman player in this game is betting that the fear and uncertainty about how to handle their unpredictability will cause their counterparty to make concessions they might not make otherwise, Like you end up thinking, well, I'll just take a bad deal rather than wander into territory with unclear levels of.

if it were only as easy as we're laying it out here, it would almost be a different situation totally.

risk into playing games like this. There are a lot of reasonable criticisms of mad Man theory, but I think a major one, at least from my point of view, is that at best it is a strategy for short term gain at the expense of long term stability and negotiating power, because one way or another it undercuts your credibility and your ability to be seen as an honest and reliable broker in future negotiations, either because you come to be seen as actually irrational, in which case people

don't want to deal with you, and they you know, they will find what you know. They may be they may be motivated to do something bad to strike first to you because you're too dangerous to be let loose, or they may be, you know, they may try to find ways to cut you out of negotiations, so that could be one consequence, or just because you are revealed to have been bluffing, in which case you lose your credibility.

more successful international coercion? By Roseanne W McManus. This came out in twenty nineteen from Cambridge University Press, and in this the author points out that a reputation for madness would seem to be more often harmful than helpful in international coercion, So it undercuts the leader's ability to make believable peace commitments, treaties and so forth, which and it really almost feels absurd that we need to stress this. These are all vital for peaceful relations between nations. No

one knows what I'll do easily slides into. No one knows what I'll honor, No one knows what treaties I would well I actually stand by, and so forth. You know, to bring this back to Dungeons and Dragons, I think any dn D player worth your salt knows that while you might well enter into a pact with a lawful evil devil, you never enter into a pack with a chaotic evil demon because the lords of the nine hell are going to stick to their letter of the contract,

if not the spirit of the contract. But demon lords such as the mcgorgan will honor nothing, and there is no coherence even within themselves. They're just pure ambiguity, and you can't strike any sort of deal because they won't stand by it no matter what. Whereas the devil's being lawful evil, they may look for that wiggle room, they may find ways to, you know, to avoid the spirit of the deal, but they're still bound to the letter.

destroy themselves. So the idea here is that no one would take this hypothetical nation or the leader of this hypothetical nation seriously unless they presented an air of madness, and then perhaps, hey, they might do it anyway if provoked. They they you know, they're they're unhinged, and thus it can be a form of asymmetric leverage. So we're talking

low probability but high consequence. And MacManus also points out that there would also seem to be more of an advantage if you had a mild reputation for madness rather than an extreme one. You don't want to come off like, you know, a complete da da you know, eye rolling maniac.

In one of these cases, the idea would be like, well, you know, we can, we can still get this individual to the negotiation table, but we just have to be hyper sensitive and I guess again, kind of avoiding that complete fall off into chaotic evil demonhood and instead dealing with devils that can still be bound by some sort of law. All right, I have a hypothetical example here. So we have demigodic hero Hercules and we have his mortal cousin Eurystheus, and they're grabbing lunch at the local

euro place. Great, So Eurystheus warns his cousin that if he tries to steal any of his fries, he's going to flip the table and all their food's going to wind up on the floor, and he acts just really

sensitive and unhinged about the whole thing. Yes, Okay, Hercules obviously knows that Eurystheus can't take him in a fight and might not even be able to flip that table over before Hercules stops him, and rationally, Eurystheus should realize this and just let big Hurk have a few extra fries if he wants, like that's what does it matter. But from Hercules's standpoint, does he really want to risk his lunch winding up on the floor. Maybe he should

just let Eurystheus keep all of his fries. Okay. The downside of course here is that maybe Hercules just won't invite you risk the us out for lunch next time, and he's certainly not going to pick up the check. And on top of all that, how are they going to work together to defeat the hydra? Because, as you remember from past episodes and from Greek mythology in general, Hercules can't do that on his own. He has to have his cousins help to burn the stomps after each head of the hydras cut off.

and you don't care about long term goals. You are only interested in this situation right now, in extracting a short term advantage.

that situation. There's still the domestic response, the domestic relationship. And so one of the papers that I was looking at this one is titled mad Man or mad Genius. The International Benefits and Domestic Costs of mad Man's Strategy by Joshua A. Schwartz, published twenty twenty three in Security Studies, and in this the author found that the Madman approach can work in negotiations with foreign adversaries, but quote entails

significant domestic costs that potentially erode its efficacy. He also points out that mad Man theory may simply not work against major powers, because while it might make a threat more credible, that doesn't necessarily make the threat more effective and make the adversary willing to cave to those demands. So he points out that this is perhaps why Nixon's use of mad Man theory didn't work against the Soviets. Like, Okay, it made Nixon's threat more credible, but did it actually

make it more effective? Did he actually get what he wanted to out of these bluffs these threats, and then you have the domestic side of things, with the hypothetical leader's own citizens not loving the heightened stakes because you know, obviously I mean, I say obviously, but this is one of those things that we often have to be reminded of in a case of mutually assured destruction, or even a case of asymmetric exchange involving nuclear weapons or some

other kind of just you know, horrible weapon of mass destruction, a leader is always bargaining with the lives of their own citizens. You are the chips on the board in

a no one knows what I'll do wager. Yeah, And there also notes that, okay, you know, this is going to also differ depending on how much of a voice the people have in a given nation versus how much power the ruler has, and there's going to be less backlash in places where the people have less of a voice in the ruler has more absolute power.

the long term. But you know, unfortunately this ends up rolling out into so many things about human experience and human perception. Is that we are so focused on the short term and we don't think about the long term.

ambiguity aversion offensively and negotiations. But I think one way that it can definitely be useful is to have knowledge of ambiguity aversion to use defensively to think about in analyzing your own behavior and being aware of your bias to avoid ambiguity and sort of check yourself and think, like, wait a minute, am I actually making the right decision here?

Is this actually what's rational? Or am I just having an irrational bias against situations with unknown probabilities that are having an outsized scariness in my mind because of the ambiguity involved. Like the example we talked about earlier with people being afraid to try new experiences because there's some amount of ambiguity.

be better understood? So yeah, well we'll talk about that kind of stuff next time.