WEBVTT - The Devil You Know, Part 2 0:00:03.040 --> 0:00:06.760 Welcome to Stuff to Blow Your Mind, the production of iHeartRadio. 0:00:12.960 --> 0:00:14.680 Hey you welcome to Stuff to Blow your Mind. My 0:00:14.760 --> 0:00:15.480 name is Robert. 0:00:15.360 --> 0:00:18.319 Lamb and I am Joe McCormick, and we're back with 0:00:18.440 --> 0:00:21.599 part two in our series called The Devil You Know, 0:00:22.040 --> 0:00:26.600 which is about ambiguity aversion. That's the observation that we 0:00:26.800 --> 0:00:32.920 prefer risks with known probabilities over risks with unknown probabilities, 0:00:33.000 --> 0:00:35.559 even if you have no reason for thinking that the 0:00:35.680 --> 0:00:39.879 unknown risks will be worse. This preference is captured by 0:00:39.880 --> 0:00:42.400 the folk saying, better the devil you know than the 0:00:42.400 --> 0:00:46.040 devil you don't. For the most part, people don't like 0:00:46.280 --> 0:00:48.879 taking bets when they don't know what the odds are, 0:00:49.200 --> 0:00:51.640 and will pay up just to avoid having to deal 0:00:51.800 --> 0:00:55.680 with ambiguity. And this is funny in a way because 0:00:55.800 --> 0:00:58.680 almost all of the decisions that we make in our 0:00:58.720 --> 0:01:03.280 actual lives are made with incomplete knowledge and some significant 0:01:03.360 --> 0:01:06.720 amount of uncertainty about our likelihood of success. Very few 0:01:06.720 --> 0:01:10.240 things in reality have clear objective odds, like a fair 0:01:10.319 --> 0:01:14.240 coin flip or a dice roll. However, it's pretty well 0:01:14.360 --> 0:01:17.360 established that most people, most of the time, do not 0:01:17.640 --> 0:01:20.760 like making decisions under ambiguous conditions, and we will go 0:01:20.800 --> 0:01:25.240 to great lengths to avoid taking risks with unknown probabilities. 0:01:25.400 --> 0:01:29.560 So in the last episode, after we introduced the concept, 0:01:29.800 --> 0:01:32.319 we talked a good bit about the original piece of 0:01:32.360 --> 0:01:35.959 writing that made ambiguity aversion famous. This was a nineteen 0:01:36.040 --> 0:01:39.920 sixty one paper by the American economist, anti war activist, 0:01:40.000 --> 0:01:44.800 and whistleblower Daniel Ellsberg. The paper was called Risk Ambiguity 0:01:44.840 --> 0:01:49.000 and the Savage Axioms, published in the Quarterly Journal of Economics, 0:01:49.600 --> 0:01:53.280 and to briefly recap, this paper proposed the concept of 0:01:53.320 --> 0:01:57.360 ambiguity aversion using a number of thought experiments where you 0:01:57.400 --> 0:02:01.200 could take different bets based on like which color ball 0:02:01.360 --> 0:02:04.640 you would draw by chance out of an urn. Elsberg's 0:02:04.640 --> 0:02:07.520 point was that people would tend to prefer bets with 0:02:07.600 --> 0:02:10.520 clear odds of winning, say a one in three chance, 0:02:11.080 --> 0:02:14.840 over bets with unknown odds, where you could have anywhere 0:02:14.880 --> 0:02:17.480 between a zero percent chance or a two out of 0:02:17.520 --> 0:02:22.840 three chance. Elsberg's intuition here has been broadly supported by 0:02:22.880 --> 0:02:26.880 real world experiments. There are a few exceptions, but most 0:02:26.880 --> 0:02:29.600 of the time most people would rather take bets with 0:02:29.639 --> 0:02:33.480 clear odds, even in ways that end up implying self 0:02:33.520 --> 0:02:37.919 contradictory assumptions about the unknown odds. This was an important 0:02:37.919 --> 0:02:41.480 discovery in the economic field known as decision theory because 0:02:41.520 --> 0:02:45.880 it violated a framework known as Savage's axioms, which were 0:02:45.880 --> 0:02:49.800 widely used to model how people make decisions in situations 0:02:49.840 --> 0:02:54.720 of uncertainty or ambiguity, and this self contradictory behavior is 0:02:54.760 --> 0:02:58.680 now known as the Ellsberg paradox. And then, finally, in 0:02:58.680 --> 0:03:01.880 the last episode, after talking about the theoretical origins of 0:03:01.960 --> 0:03:06.720 ambiguity aversion, we also talked about this phenomenon in practice 0:03:06.720 --> 0:03:09.480 in the real world, and our main example here was 0:03:09.560 --> 0:03:14.079 the negotiation strategy famously associated with the Nixon White House 0:03:14.560 --> 0:03:19.120 known as madman theory, a strategy where you intentionally cultivate 0:03:19.520 --> 0:03:24.880 a reputation for volatility and unpredictability. In essence, you make 0:03:24.919 --> 0:03:28.160 your peers and your counterparties worry that you are a 0:03:28.200 --> 0:03:32.160 madman who is capable of anything, and the idea here 0:03:32.600 --> 0:03:34.560 we ended up talking about a lot of reasons for 0:03:34.639 --> 0:03:37.040 thinking that this is actually not a good strategy, but 0:03:37.120 --> 0:03:40.160 the idea here is that you will exploit their natural 0:03:40.200 --> 0:03:43.440 ambiguity aversion to get them to make concessions that they 0:03:43.480 --> 0:03:46.240 would not make otherwise, And then we did get into 0:03:46.320 --> 0:03:51.839 some international relations research on Madman theory, like reasons why 0:03:51.880 --> 0:03:54.440 it might not actually be a good strategy, especially for 0:03:54.480 --> 0:03:56.960 achieving long term goals. In fact, I think in the 0:03:57.000 --> 0:03:59.920 long term most experts seem to agree that it is 0:04:00.280 --> 0:04:04.360 harmful to everyone involved, including the practitioner. But again, the 0:04:04.440 --> 0:04:07.640 reason it may sometimes work for achieving short term wins 0:04:08.000 --> 0:04:11.480 is by taking advantage of this fear people have of ambiguity, 0:04:11.520 --> 0:04:13.680 the fact that a lot of people would rather just 0:04:13.760 --> 0:04:16.960 take a bad deal than keep negotiating with somebody who 0:04:17.040 --> 0:04:18.240 is highly unpredictable. 0:04:19.040 --> 0:04:21.360 You know, I was thinking after our recording that another 0:04:23.600 --> 0:04:29.000 often cited bit of supposed wisdom along these lines is 0:04:29.279 --> 0:04:33.840 the idea in generally you'll see this reference out of context, 0:04:34.000 --> 0:04:37.760 so someone will not be talking about actually going to prison. 0:04:37.839 --> 0:04:40.000 They'll be talking about your first dai of new school 0:04:40.120 --> 0:04:42.080 or something, and they'll jokingly say, well, you know what 0:04:42.120 --> 0:04:43.520 you're supposed to do. You're supposed to go up to 0:04:43.560 --> 0:04:46.239 the biggest person around and hit them with a steel chair, 0:04:46.720 --> 0:04:49.160 And then everyone will be like, WHOA, that guy's crazy. 0:04:49.160 --> 0:04:53.039 We better not mess with him. It seems to follow 0:04:53.080 --> 0:04:57.679 similar logic, and also seem equally, if not more flawed 0:04:57.800 --> 0:05:00.560 in its approach. It does not sound like a tactic 0:05:00.640 --> 0:05:04.400 that would actually generate like, you know, long term benefits. 0:05:04.839 --> 0:05:07.560 It's one of those pieces of advice that is probably 0:05:07.960 --> 0:05:11.400 more often repeated because it's like memorable and funny and 0:05:11.520 --> 0:05:14.240 makes for a good story because that many people would 0:05:14.279 --> 0:05:17.640 think it's actually good advice, especially if you have relevant experience. 0:05:17.839 --> 0:05:18.200 I don't know. 0:05:18.200 --> 0:05:20.400 I mean, I don't know a lot about like prison strategy, 0:05:20.440 --> 0:05:23.719 but that does seem like you could have some real downsides. 0:05:24.880 --> 0:05:26.760 You know, it makes me think of that the you know, 0:05:26.800 --> 0:05:29.800 the the this sense another bit of wisdom, another nugget 0:05:29.800 --> 0:05:33.960 that's Offen dished out, and that is getting into the 0:05:33.960 --> 0:05:36.920 philosophy of Nietzsche. Uh, you know that which doesn't kill 0:05:37.000 --> 0:05:39.680 us makes us stronger. But I forget who it was 0:05:39.720 --> 0:05:41.320 who did a spin on this and said, well that 0:05:41.320 --> 0:05:45.520 that which doesn't kill us nearly kills us, and that 0:05:45.560 --> 0:05:47.680 should be a reason to give us pause. 0:05:48.040 --> 0:05:50.400 That's funny because that Nietzsche saying is also when I 0:05:50.800 --> 0:05:52.279 know this has come up on the show before, I 0:05:52.320 --> 0:05:55.440 know that like in huge number of cases that's not true. Yeah, 0:05:55.560 --> 0:05:59.679 like sometimes it makes us stronger, like we do sometimes learn, 0:06:00.160 --> 0:06:02.960 learn and grow and gain strength from adversity, but that's 0:06:03.000 --> 0:06:06.000 like some subset of adversities. A lot of adversities leave 0:06:06.040 --> 0:06:08.760 us much weaker and deeply shaken. 0:06:08.800 --> 0:06:11.719 Yeah, shaken, or less trusting or just yeah, or just 0:06:11.800 --> 0:06:14.280 completely weakened. And I know this may weaken us. We 0:06:14.320 --> 0:06:16.400 survive it, but it's weakened our system and then we're 0:06:16.400 --> 0:06:19.200 more susceptible to the next thing and so forth. So yeah, 0:06:19.320 --> 0:06:21.359 I mean, it's great to take solace in some of 0:06:21.400 --> 0:06:24.560 these sayings, but you can only take them so far, really, So. 0:06:24.480 --> 0:06:27.280 Anyway, we're back today to talk about some more angles 0:06:27.279 --> 0:06:30.280 on ambiguity aversion, and one of the first things I 0:06:30.279 --> 0:06:33.680 want to get into is that in the last episode 0:06:34.720 --> 0:06:37.200 we did talk a bit about the distinction that Daniel 0:06:37.240 --> 0:06:41.400 Elsberg tried to make between ambiguity and risk. But I 0:06:41.400 --> 0:06:43.640 think one thing that's really important to point out that 0:06:43.720 --> 0:06:47.039 we didn't get into last time is the distinction between 0:06:47.240 --> 0:06:52.080 ambiguity aversion and risk aversion. I think risk aversion is 0:06:52.080 --> 0:06:54.680 a concept more people are probably familiar with, you know, 0:06:54.720 --> 0:06:57.800 that comes up in conversation. Sometimes these are not the 0:06:57.880 --> 0:07:03.240 same thing. One who is risk averse doesn't like to 0:07:03.320 --> 0:07:07.480 make bets. They would prefer a sure thing, even if 0:07:07.520 --> 0:07:12.440 the sure thing is statistically less valuable than the risk. So, 0:07:12.720 --> 0:07:16.400 for example, imagine a fair coin flip game. We stipulate 0:07:16.440 --> 0:07:19.320 that this is a fair coin. There's no cheating involved, 0:07:19.480 --> 0:07:21.800 and on each round when I flip the coin, I 0:07:21.840 --> 0:07:24.840 can either just give you forty dollars no matter how 0:07:24.880 --> 0:07:28.120 the coin lands, or I could give you one hundred 0:07:28.160 --> 0:07:31.280 dollars every time it lands heads and nothing every time 0:07:31.320 --> 0:07:32.280 it lands tails. 0:07:33.120 --> 0:07:33.600 So with the. 0:07:33.520 --> 0:07:37.720 Gambling condition here, you will on average win fifty dollars 0:07:37.800 --> 0:07:40.120 per round. It's a fifty to fifty chance of winning, 0:07:40.200 --> 0:07:43.080 and the payout is one hundred dollars. So the odds 0:07:43.120 --> 0:07:46.000 here are clear. If you play this game a thousand 0:07:46.000 --> 0:07:48.560 times in a row, you will make more money by 0:07:48.600 --> 0:07:51.800 gambling than by taking the guaranteed forty dollars every time. 0:07:52.200 --> 0:07:56.400 It's statistically all but a sure thing. But still some 0:07:56.440 --> 0:08:00.200 people would rather just take the actual sure thing. They 0:08:00.200 --> 0:08:03.480 would rather just take the forty dollars, and that is 0:08:03.600 --> 0:08:08.160 risk aversion. It is a psychological tendency to dislike and 0:08:08.360 --> 0:08:11.760 avoid taking a stake in uncertain outcomes, even if you 0:08:11.840 --> 0:08:14.600 know the exact odds of those outcomes and potentially even 0:08:14.640 --> 0:08:17.920 if those odds will statistically bring you more benefit than 0:08:18.000 --> 0:08:20.640 the sure thing. And another way of thinking about risk 0:08:20.680 --> 0:08:25.000 aversion is that sometimes to some people, the feeling of 0:08:25.160 --> 0:08:28.680 security that you get from taking the sure thing is 0:08:28.720 --> 0:08:32.560 actually more valuable than the money. It's more valuable than 0:08:32.600 --> 0:08:35.240 the extra money that you would take home on average 0:08:35.520 --> 0:08:39.120 by placing a bet with a higher expected value. And 0:08:39.160 --> 0:08:43.080 this emphasizes something that is true in decision theory and economics. 0:08:43.360 --> 0:08:47.480 Value isn't just about getting or keeping money. We use 0:08:47.559 --> 0:08:51.480 money because it's an easy way to quantify value and experiments. 0:08:51.520 --> 0:08:56.160 But sometimes we would pay money just to feel certainty 0:08:56.240 --> 0:08:59.160 and comfort. That feeling is of higher value to us 0:08:59.200 --> 0:09:02.679 than the money is. But then, of course again, other 0:09:02.720 --> 0:09:04.880 people at other times would not pay that. Some people 0:09:04.960 --> 0:09:07.800 love the feeling of risk. I mean people gamble for pleasure. 0:09:08.559 --> 0:09:11.439 Yeah, absolutely, but yeah, but other times it's like I 0:09:12.200 --> 0:09:13.680 just don't want to take a chance on this. I'd 0:09:13.720 --> 0:09:17.680 rather just go for what's going to be straightforward and guarantee, right. 0:09:17.800 --> 0:09:20.960 So that's risk and risk aversion. Ambiguity a version is 0:09:21.040 --> 0:09:24.040 different from risk aversion because it is not about a 0:09:24.080 --> 0:09:28.640 dislike of risk, but specifically a desire to avoid ambiguous 0:09:28.760 --> 0:09:32.320 risks with unclear odds. You might love gambling on a 0:09:32.320 --> 0:09:35.280 fifty to fifty coin toss, but you might not want 0:09:35.320 --> 0:09:37.120 to play the game where you have to guess how 0:09:37.160 --> 0:09:39.280 many black balls or yellow balls there are and there's 0:09:39.280 --> 0:09:42.600 no way to know. So a person could technically be 0:09:42.880 --> 0:09:45.840 not very risk averse at all, but highly averse to 0:09:45.880 --> 0:09:50.400 ambiguity the interaction. The more I thought about it, the 0:09:50.440 --> 0:09:53.479 more I thought that the interaction between these two tendencies 0:09:53.559 --> 0:09:58.000 in real life is interesting because I think to many 0:09:58.080 --> 0:10:01.920 of us, it's not always if a fear or a 0:10:02.000 --> 0:10:07.120 hesitancy we're experiencing about making a decision is rooted in 0:10:07.480 --> 0:10:11.680 risk aversion or ambiguity aversion. For instance, this came up 0:10:11.679 --> 0:10:14.000 because I was trying to think of examples of risk aversion, 0:10:14.840 --> 0:10:17.680 and I thought of the idea of what to do 0:10:17.960 --> 0:10:20.640 with savings. Imagine you save up a little bit of 0:10:20.720 --> 0:10:23.079 extra money at work from your paycheck, and you want 0:10:23.120 --> 0:10:26.319 to put that toward retirement, and you have multiple options 0:10:26.360 --> 0:10:28.880 of what to do with that money. Imagine a person's 0:10:28.920 --> 0:10:32.040 in that scenario and they discover that they would really 0:10:32.120 --> 0:10:35.400 feel better just keeping the money in a bank account, 0:10:35.600 --> 0:10:39.760 earning a relatively low rate of interest, rather than investing 0:10:39.800 --> 0:10:42.920 it in like a stock market index fund, where with 0:10:43.000 --> 0:10:46.120 the index fund, it is commonly assumed it is standard 0:10:46.160 --> 0:10:49.360 financial wisdom that the balance will grow at a faster 0:10:49.480 --> 0:10:51.680 rate than it will just sitting in a savings account, 0:10:51.880 --> 0:10:54.040 and you will end up with more money years down 0:10:54.080 --> 0:10:56.040 the road if you put it in the index fund. 0:10:56.760 --> 0:10:59.800 This pattern of behavior could be taken as an example 0:10:59.880 --> 0:11:03.679 of risk aversion, because it's like, I would choose to 0:11:03.840 --> 0:11:06.360 probably end up with less money just so I can 0:11:06.440 --> 0:11:11.440 avoid taking risks with what I have. But then I thought, well, no, 0:11:11.600 --> 0:11:15.880 not necessarily. What if that same impulse is actually a 0:11:15.880 --> 0:11:20.800 better example of ambiguity a version than risk aversion, Because 0:11:20.960 --> 0:11:24.040 how do we actually know that a stock market index 0:11:24.120 --> 0:11:27.280 fund will grow more over time than the balance of 0:11:27.320 --> 0:11:31.360 a savings account. The answer is historical patterns. It's like 0:11:31.640 --> 0:11:34.680 statistically true over the last one hundred years or so, 0:11:34.800 --> 0:11:38.040 you know, within certain economies in living memory. But it's 0:11:38.080 --> 0:11:41.760 not actually like a fixed odds casino game or a 0:11:41.880 --> 0:11:45.600 lab experiment where the odds of winning are fair axiomatically 0:11:46.160 --> 0:11:48.679 in the real world. As the fine print always says, 0:11:48.800 --> 0:11:53.439 investing involves risk, including loss of principle, and past performance 0:11:53.440 --> 0:11:58.240 does not guarantee future returns. So maybe actually preferring the 0:11:58.240 --> 0:12:01.679 savings account is a form of ambiguity aversion rather than 0:12:01.760 --> 0:12:06.280 risk aversion, responding to the idea that, well, what are 0:12:06.320 --> 0:12:10.160 the odds that suddenly, somehow investing in the stock market 0:12:10.280 --> 0:12:12.800 becomes a bad idea in a way that it has 0:12:12.920 --> 0:12:16.600 not within living memory. It's hard to understand what the 0:12:16.640 --> 0:12:20.480 odds of something like that are. Seems pretty unlikely, but 0:12:20.720 --> 0:12:24.280 it's fundamentally ambiguous. You just can't say what the objective 0:12:24.280 --> 0:12:26.760 odds are. There's really no way to calculate them. So 0:12:27.000 --> 0:12:32.920 sometimes there's actually ambiguity about whether there is ambiguity, you know, 0:12:33.000 --> 0:12:35.880 because like I think an invest an investment advisor might 0:12:35.880 --> 0:12:38.720 tell you on one hand, like, well, you know, historical 0:12:38.880 --> 0:12:41.280 performance of the stock market, you can pretty much depend 0:12:41.320 --> 0:12:43.640 on that that's a known risk when you know you 0:12:43.640 --> 0:12:47.320 have you have statistics that tell you that you'll most 0:12:47.440 --> 0:12:49.720 likely get this rate of return on average, and you 0:12:49.840 --> 0:12:53.040 just don't know how seriously to take the possibility that 0:12:53.080 --> 0:12:54.840 the future will not be like the past. 0:12:55.720 --> 0:12:58.800 Yeah. Absolutely, And another angle that I was looking at 0:12:58.800 --> 0:13:00.839 here on all this is that it's not just the 0:13:00.880 --> 0:13:04.320 ambiguity and risk, but it's perceptions of ambiguity and risk. 0:13:04.880 --> 0:13:08.440 So you know, it ultimately isn't necessarily coming down as 0:13:08.480 --> 0:13:11.959 much to like what are actually the hard probabilities here. 0:13:12.480 --> 0:13:15.520 It can skew greatly depending on what a particular individual's 0:13:15.520 --> 0:13:18.640 worldview is. So you could have somebody like you could 0:13:18.640 --> 0:13:23.880 have like a you know, a Washington Bartholomew hog Wallop figure, 0:13:23.880 --> 0:13:26.040 and he's decided not only is he not going to 0:13:26.040 --> 0:13:28.120 invest in the stock market, he's not even gonna put 0:13:28.160 --> 0:13:29.880 his money in the bank. It's better off in the 0:13:29.920 --> 0:13:31.640 tin can buried in his backyard. 0:13:31.760 --> 0:13:32.480 Yeah, there you go. 0:13:32.559 --> 0:13:36.160 Yeah, even though there may be very good statistics that 0:13:36.240 --> 0:13:39.120 show that is less safe. Yeah, and that is a 0:13:39.160 --> 0:13:43.920 greater risk. Your worldview and perception could be Oh no, absolutely, 0:13:43.960 --> 0:13:46.880 you can't trust the banks at all. I can only 0:13:46.920 --> 0:13:49.960 trust myself, and therefore it's safer in my yard with 0:13:50.000 --> 0:13:59.880 me standing over it with a shotgun. 0:14:01.600 --> 0:14:05.120 But anyway, so this all brings us back to this 0:14:05.240 --> 0:14:09.880 distinction between risk aversion and ambiguity aversion, and how difficult 0:14:09.880 --> 0:14:11.960 it can be to tell the difference between the two, 0:14:12.120 --> 0:14:15.199 Like when we feel aversion to making a bet of 0:14:15.240 --> 0:14:17.760 some kind, which by way of analogy. That doesn't have 0:14:17.800 --> 0:14:19.760 to mean literally a bet with money, but just any 0:14:19.800 --> 0:14:24.800 important decision in our lives when we feel that, like fear, hesitancy, 0:14:25.000 --> 0:14:28.880 or resistance. We often think of this as risk aversion, 0:14:29.440 --> 0:14:33.320 but could it actually be more often ambiguity aversion That 0:14:33.360 --> 0:14:37.200 we're responding subconsciously to the fact that we don't know 0:14:37.240 --> 0:14:40.160 how to calculate the odds for the decision we're about 0:14:40.160 --> 0:14:43.520 to make, and that makes us uncomfortable. Now, one of 0:14:43.600 --> 0:14:46.360 the big things we left open after we introduce the 0:14:46.400 --> 0:14:49.400 concept of ambiguity a version in the last episode is 0:14:49.960 --> 0:14:53.640 how is decision theory supposed to make sense of this 0:14:53.840 --> 0:14:57.280 finding of like the Elsberg paradox? And there has been 0:14:57.360 --> 0:15:00.000 a great amount of inks billed in answering this question 0:15:00.080 --> 0:15:03.280 in the fifty something years since Elsberg. I don't think 0:15:03.320 --> 0:15:05.520 I can even attempt to cover the whole landscape of 0:15:05.560 --> 0:15:07.680 responses here, so I want to be clear this will 0:15:07.680 --> 0:15:11.440 not be a comprehensive technical survey of the ambiguity of 0:15:11.520 --> 0:15:13.720 version literature. I'm just going to mention a couple of 0:15:13.720 --> 0:15:17.360 broad trends and then a few ideas within each. So 0:15:17.920 --> 0:15:21.960 within the responses there are two main categories of thought. 0:15:22.360 --> 0:15:27.600 One assumes that ambiguity aversion is actually a component of 0:15:27.720 --> 0:15:31.680 rational decision making, that it is, in a subjective sense 0:15:31.960 --> 0:15:35.760 wise to avoid ambiguity even when you can't really prove 0:15:35.880 --> 0:15:40.040 that the ambiguous bets will generally turn out bad. The 0:15:40.160 --> 0:15:43.560 other main way of thinking is that ambiguity a version 0:15:43.720 --> 0:15:48.120 is some type of error or cognitive bias. And here 0:15:48.160 --> 0:15:50.000 I'm going to mention a couple of sources. The first 0:15:50.000 --> 0:15:53.080 one I cited in our last episode. That is a 0:15:53.520 --> 0:15:58.440 book chapter by Mark J. Maschina and Marciano Siniscalci called 0:15:58.480 --> 0:16:02.240 Ambiguity and Ambiguity of Version in the Handbook of Economics 0:16:02.280 --> 0:16:06.440 of Risk and Uncertainty published in twenty fourteen. And then 0:16:06.440 --> 0:16:08.040 the second source I want to mention here is by 0:16:08.080 --> 0:16:12.800 Nabil I al Najar and Jonathan Weinstein published in the 0:16:13.320 --> 0:16:16.920 Journal of Economics and Philosophy called the Ambiguity of Version 0:16:16.960 --> 0:16:20.880 Literature a Critical Assessment. This one is more in that 0:16:21.000 --> 0:16:24.880 second category I mentioned. It's an approach arguing that ambiguity 0:16:24.920 --> 0:16:27.360 of version is actually just a type of error basically. 0:16:28.440 --> 0:16:30.080 So first I want to look at what are some 0:16:30.160 --> 0:16:33.160 of the ways that ambiguity aversion and like the Elsberg 0:16:33.240 --> 0:16:36.880 paradox might actually be a form of rational, self consistent 0:16:36.960 --> 0:16:42.040 decision making. One theory here can be called the multiple 0:16:42.160 --> 0:16:46.680 priors model, also known as max men expected utility max 0:16:46.720 --> 0:16:51.840 men is max n max men expected utility or MEU. 0:16:52.680 --> 0:16:55.360 Let's think about it like this. Imagine you are asked 0:16:55.400 --> 0:16:58.640 to make a bet involving ambiguous odds, like we'll go 0:16:58.720 --> 0:17:00.760 back to the balls in the urn example from the 0:17:00.760 --> 0:17:03.840 first episode. So am I more likely to pull a 0:17:03.920 --> 0:17:06.400 red ball or a black ball out of this urn? 0:17:07.560 --> 0:17:10.680 And so to remember the details there? The three color 0:17:10.680 --> 0:17:13.880 eurn experiment was that inside this urn you can't see inside, 0:17:13.960 --> 0:17:16.480 you don't know what color you're reaching into grab, but 0:17:16.560 --> 0:17:20.359 there are exactly thirty red balls and then sixty balls 0:17:20.359 --> 0:17:23.359 that could be any mix of yellow and black. So 0:17:23.720 --> 0:17:27.200 they could be zero yellow and sixty black, sixty yellow 0:17:27.200 --> 0:17:30.120 and zero black, thirty and thirty, or anything in between. 0:17:30.240 --> 0:17:32.200 You have no way of knowing the mix in advance. 0:17:33.000 --> 0:17:36.639 In these situations, maximin expected utility theory would say that 0:17:36.920 --> 0:17:41.560 instead of forming one consistent belief about the number of 0:17:41.600 --> 0:17:45.200 black balls inside and then placing bets according to that prediction, 0:17:46.119 --> 0:17:50.919 we allow ourselves to consider multiple different probabilities and we 0:17:51.000 --> 0:17:57.199 apply those probabilities to different bets pessimistically. So essentially, for 0:17:57.320 --> 0:17:59.920 each new bet that requires us to guess the number 0:17:59.880 --> 0:18:02.879 of black balls, we ask ourselves what would be the 0:18:02.920 --> 0:18:06.080 worst case scenario for me here, and then we assume 0:18:06.240 --> 0:18:09.880 those worst case odds when considering the bet. So if 0:18:09.880 --> 0:18:13.400 you're asked to choose between black and red, there could 0:18:13.440 --> 0:18:16.720 be anywhere between zero and sixty black balls, you just 0:18:16.760 --> 0:18:20.119 assume there are zero and you bet on red. I 0:18:20.160 --> 0:18:22.720 tried to come up with a real life decision sequence 0:18:22.760 --> 0:18:26.879 that might help the contradictory behavior under MEU makes sense. 0:18:27.320 --> 0:18:30.240 So here's what I've got to imagine this. You're making 0:18:30.280 --> 0:18:32.919 plans for the afternoon and you don't know what the 0:18:32.960 --> 0:18:35.679 weather is going to be. It might rain, it might not. 0:18:35.880 --> 0:18:38.679 Don't know how to calculate that. You were thinking about 0:18:38.680 --> 0:18:42.040 maybe going out for a walk, but then again, considering 0:18:42.080 --> 0:18:44.120 that it might rain, you don't want to be caught 0:18:44.119 --> 0:18:46.919 out in a storm, so you decide instead to stay 0:18:46.920 --> 0:18:50.520 home and do some indoor activities. At the same time, 0:18:50.680 --> 0:18:53.840 you are also deciding what to do about the flowers 0:18:53.880 --> 0:18:56.760 on your front porch. They're drooping and they need some water. 0:18:57.200 --> 0:18:59.919 Now it might rain and then they would get all 0:19:00.080 --> 0:19:02.480 the water they need from that, But then again it 0:19:02.560 --> 0:19:05.359 might not rain, so you decide to make time to 0:19:05.400 --> 0:19:08.879 fill up a watering can and water them manually. This 0:19:08.960 --> 0:19:13.520 could appear inconsistent because you've made two different decisions, both 0:19:13.560 --> 0:19:15.919 on the bet that it will rain in one case 0:19:16.000 --> 0:19:17.960 and on the bet that it won't rain in the 0:19:18.000 --> 0:19:21.840 other case. It's possible that that's just self contradictory, or 0:19:22.320 --> 0:19:26.440 it could be perfectly self consistent and logical if you 0:19:26.480 --> 0:19:30.000 are considering a range of different probabilities of rain for 0:19:30.080 --> 0:19:33.919 the afternoon and then applying the most pessimistic one to 0:19:34.040 --> 0:19:38.080 each decision rather than just assuming one probability that applies 0:19:38.119 --> 0:19:40.119 to all decisions. Does that make sense? 0:19:40.760 --> 0:19:41.800 Yeah, yeah, I think so. 0:19:42.560 --> 0:19:44.840 And to be clear, there's no guarantee here that the 0:19:44.880 --> 0:19:49.480 pessimistic assumptions will be correct, because again, the information environment 0:19:49.600 --> 0:19:53.439 is truly ambiguous. You just can't know. That's what ambiguity is. 0:19:54.000 --> 0:19:57.920 But considering multiple probabilities and acting according to the most 0:19:58.080 --> 0:20:02.320 pessimistic one for each decision is theoretically a logical and 0:20:02.520 --> 0:20:06.879 consistent strategy. So the authors that support this model argue 0:20:06.920 --> 0:20:09.920 that it is actually a rational way. Ambiguity version might 0:20:10.000 --> 0:20:14.000 produce the apparently self contradictory results we observe, like the 0:20:14.040 --> 0:20:16.919 apparent contradiction comes from the fact that we're sort of 0:20:16.920 --> 0:20:20.480 coming up with different predictions for each bet instead of 0:20:20.520 --> 0:20:23.639 one prediction for the actual full state of affairs. 0:20:24.400 --> 0:20:26.560 Yeah. Though, of course, making all your decisions based on 0:20:26.600 --> 0:20:29.320 worst case scenario is not a great way to live your. 0:20:29.160 --> 0:20:31.199 Life, right, And I'm going to come back to that 0:20:31.240 --> 0:20:33.359 in a minute, because that is part of the critical assessment. 0:20:34.000 --> 0:20:37.640 Another theory that would make the Elsberg paradox behavior rational 0:20:38.160 --> 0:20:42.360 is what's known as show qu expected utility or CEU 0:20:42.520 --> 0:20:44.840 and the show K. There. That's a French name. It's 0:20:45.200 --> 0:20:46.560 cho q u e. 0:20:46.600 --> 0:20:48.800 T show k sounds delicious. 0:20:49.320 --> 0:20:53.280 Oh yeah, so sho K. Expected utility is the idea 0:20:53.320 --> 0:20:56.920 that when we evaluate a bet and the odds are ambiguous, 0:20:57.320 --> 0:21:00.760 we basically we do come up with a single internal 0:21:00.800 --> 0:21:04.760 belief about probability, but it takes a weird form. Instead 0:21:04.760 --> 0:21:09.400 of a probability, we use what the literature calls a capacity. 0:21:10.560 --> 0:21:12.960 So most of the time when we're trying to judge 0:21:12.960 --> 0:21:17.119 how likely something is to happen, together the likelihood of 0:21:17.160 --> 0:21:21.040 that event and the likelihood of its negation of not 0:21:21.240 --> 0:21:24.280 that event have to sum up to one hundred percent. Right, 0:21:24.359 --> 0:21:27.160 So like under normal probability, if there is a one 0:21:27.280 --> 0:21:30.080 third chance a ball will be read, there is a 0:21:30.119 --> 0:21:33.120 two thirds chance the ball will not be read. These 0:21:33.160 --> 0:21:37.720 complementary probabilities should always sum to one, or, if expressed 0:21:37.760 --> 0:21:42.000 as a percentages up to one hundred percent. CEU theory 0:21:42.320 --> 0:21:48.520 says that our internal representation of likelihoods for ambiguous events 0:21:48.680 --> 0:21:53.080 is not like this. Instead, when the odds are ambiguous, 0:21:53.200 --> 0:21:56.959 we operate on the basis of a capacity, which mathematically 0:21:57.000 --> 0:22:02.600 incorporates our lack of confidence in the situation by representing 0:22:02.640 --> 0:22:06.960 the chance of an outcome and its negation together in 0:22:07.000 --> 0:22:08.879 a way that does not have to add up to 0:22:08.920 --> 0:22:13.320 one hundred percent. And I apologize to the more informed people, 0:22:13.359 --> 0:22:15.960 I'm skipping over some mathematical complexity here, but this is 0:22:16.080 --> 0:22:19.040 the simplified version. Is if you take the example of 0:22:19.080 --> 0:22:22.000 the three color eurn, a person might begin with the 0:22:22.080 --> 0:22:26.280 objective information that you know thirty one third of the 0:22:26.320 --> 0:22:28.679 balls are red, so you know you actually have a 0:22:28.680 --> 0:22:31.199 one third chance that the ball you draws red, And 0:22:31.240 --> 0:22:33.480 then for the yellow and black balls, instead of coming 0:22:33.560 --> 0:22:36.960 up with with guesses that sum up to two thirds, 0:22:37.000 --> 0:22:40.000 which would be the actual sort of objective probability. Whatever 0:22:40.040 --> 0:22:43.600 the mix is, you pick some smaller probabilities, like you 0:22:43.680 --> 0:22:46.800 add in a maybe a one fifth chance that the 0:22:46.840 --> 0:22:49.280 ball is yellow and a one fifth chance that the 0:22:49.320 --> 0:22:52.639 ball is black. Obviously, these these cannot be the real 0:22:52.760 --> 0:22:56.000 objective odds. Again with the caveat on what are actually 0:22:56.080 --> 0:22:59.080 objective odds, but they can't be the real They can't 0:22:59.160 --> 0:23:01.639 represent reality because they don't add up to one and 0:23:01.680 --> 0:23:04.760 there are no other colors in there. But again, the 0:23:04.840 --> 0:23:07.160 values we come up with in these kinds of decision 0:23:07.160 --> 0:23:11.320