Whether or not you think you hold them, stereotypes shape the lives of everyone on Earth. As human beings, we lack the ability to judge each situation as unique and different…and how we group novel experiences by our past conditioning, as helpful as it often is, creates extraordinary complications in society. As modern life exposes us to an increasing number of encounters with the other in which we do not have time to form accurate models of someone or some place’s true identity, we find ourselves in a downward spiral of self-reinforcing biases — transforming how we practice law enforcement, justice, and life online. Our polarized, irrational world calls for an intense look at what it will take to humanize each other — at traffic stops, in court, on social media, and anywhere our doubt about an unfamiliar face can lead to tragic consequences.
This week’s guest is Rajiv Sethi, Professor of Economics at Columbia University and External Professor at the Santa Fe Institute. In this episode, we discuss how biases in our attention and cognition lead to unfair outcomes on the streets and on the Web, and where we can look for hope in countervailing strategies.
Visit our website for more information or to support our science and communication efforts.
Join our Facebook discussion group to meet like minds and talk about each episode.
Shadows of Doubt: Stereotypes, Crime, and the Pursuit of Justice by Brendan O’Flaherty & Rajiv Sethi (Harvard University Press).
Albert Kao & Iain Couzin on collective intelligence and modular societies.
Aumann’s agreement theorem.
“We can’t disagree forever” (Geanakopolos & Polemarchakis).
Raissa D’Souza on the Collapse of Networks.
Geoffrey West on scaling laws and cities.
Music by Mitch Mignano.
Follow us on social media:
Twitter • YouTube • Facebook • Instagram • LinkedIn
Michael: Well, Rajiv Sethi, it's a pleasure to have you on Complexity.
Rajiv: Thank you for having me.
Michael: Your work is such an interesting study of something that comes up again and again on this show, which is the inadequacy of our models in describing the true full intricacy of our world. And I think it's especially important in the economic and political settings in which you're applying your stuff. Maybe the right place to start this is to tell people a little bit about how you got into your research in the first place as a young man and how you came to care about these matters of information and belief and strategy.
Rajiv: So, I think the earliest I can think of in my life where my interest in these issues was sparked was when I was 11 years old and my family moved from India to the United Kingdom. And I was suddenly in a place where I was a minority and treated as an outsider and often in ways that were quite unpleasant and unexpected, and started thinking about identity. And how that shapes interactions between people and assumptions that people make about each other. That's been a thread that's sort of run through my research ever since.
Michael: So, how did you begin to explore this as a scientist? Like, where was your entry point?
Rajiv: So I'm an economist by training and there's a significant part of economics that deals formally and systematically with information and beliefs, and the structure that is used in economics has its limitations. And so with my quote on the book that we're going to discuss in a bit, Brendan O'Flaherty, we've tried to use both insights from economics as well as from psychology and other disciplines to try to see how beliefs and information shape incentives and behavior.
And the particular application that we've been working on, which was just published in this book in April, is to do with crime and criminal justice. But I've worked on information and beliefs more generally in other settings also.
Michael: So, let's start with the more general and then get into the more specific. What do you see as the shortcoming of the Economic Game Theory that your work seeks to address?
Rajiv: So, one of the shortcomings is what economists tend to assume about beliefs. We tend to assume, at least in most of the literature that beliefs are statistically correct, even if we have very little information. So we may be aware that we are uncertain. We're not systematically biased and we process information in rational ways based on some prior belief, we process information as statisticians would. And of course psychologists have a very different tradition in thinking about beliefs going back over a hundred years now, where for psychologists, once we form a belief about something that affects our processing of evidence. So we tend to accept evidence that confirms our prior beliefs more readily than evidence that contradicts them. And so Dan and I might go off the felt in this particular project that we needed to take those insights on board.
Michael: The book in question is Shadows of Doubt. And it's about…well, essentially structural disparity in both impact and treatment in the criminal justice system. And I think in the introduction of that book, you make this as a very important distinction. And that's one I think worth unpacking here.
Rajiv: Yes. So, disparate treatment refers to the fact that two people who was in the same circumstances are treated differently, whether it's regard to let's say, police stops or the use of force or arrest, incarceration, and so on and so forth. Sentencing, jury deliberations and decisions. So, you would have disparate impact if everything else were held constant but you change, let's say, the race or the gender of a particular individual that they would end up being treated differently. Sorry, that's disparate treatment. So disparate treatment is when two individuals who are in the same circumstances but belong to different identity groups are treated differently.
But even if people are treated the same under any given set of circumstances, particular rules that are in place in society can end up with particular groups, having different impact for a given policy. So it could be subject to the same treatment, but you could end up with a particular group shouldering the bulk of the impact.
So for example, suppose that the police treated black and white civilians in a given set of circumstances in exactly the same way, but police resources were directed towards particular neighborhoods with greater frequency than other neighborhoods. Then different groups would experience that. They would experience different impact even though in any given particular encounter they may be experiencing experiencing similar treatment.
Michael: Yeah, that's a really interesting additional distinction that you make in this book about stereotypes that we carry of other people and the stereotypes that we carry of places. And the complex interrelationship between those. I'd love to hear you go into that a little bit more, especially concerning the way that we think of safe and unsafe spaces, and how that landscape can change.
Rajiv: Yeah. Most of the book is about stereotypes about people, but people have stereotypes also about places. So one of the examples that we discuss in the book is that in Minneapolis for example, most calls for service from the police come from a very small set of locations. And some of these are well known to have, let's say, bar fights and brawls and so on and so forth. And people who go there know that this is likely to happen. And so that affects those who choose to avoid that place. It affects people who choose to seek out that place and so on.
And so our behavior really changes depending on where we are at any given point in time. And that's also true of police behavior. So for example, there are particular areas in which one set of circumstances may lead to the use of force even though at a different location, the very same set of circumstances would not.
Michael: To back this up a little bit, I guess it probably makes sense for us to unpack the empirical reality of the stereotyping human being in this conversation. And there's something in that distinction between disparate treatment and disparate impact that seems to get at the insufficiency of our models at both an individual and an aggregate level and how that leads to situations in which people end up falling through the cracks or places or people are not seen in their fullness, in their multidimensionality. But you also make the case in this book that doing so would be impossible, that we just can't process that much information.
Rajiv: Right. So the way we define and think about stereotypes are as generalizations. We assign people to groups, correctly or incorrectly, we might sign them incorrectly. But having assigned them to groups, then we attribute to them characteristics that we believe are typical in the groups to which we have assigned them. So that's really roughly what we think about stereotypes. That's how we define stereotypes.
Those beliefs about the group may be right or wrong. They may be more or less accurate, but we treat individuals not as in terms of their specific individuality, but as part of a collective. Now, the psychologists have recognized that to some extent that's inescapable, that we need to have these generalizations because we live in a complex and uncertain world and it would be too hard for us to treat every situation as unique.
We both walked in here and sat down on these chairs, for example, without testing them to see if they take our weight because we assumed that they would do that, even though I certainly have not sat on this particular chair before. So, this is an example of what the psychologist Jerome Bruner called Equivalence Grouping. We take a bunch of objects and treat them as essentially the same even though they each have the unique individuality. And we can't navigate a complex world unless we do that. So in that sense, stereotyping is inevitable. It's a sort of cognitive necessity.
However, our beliefs may be incorrect about a particular set of individuals just as they may be about a particular set of objects. And in that case psychologists have recognized that we treat information in somewhat biased ways. So, if we receive evidence that confirms what we believe about a set of people, we process it and accept it and update our beliefs more readily than if we receive information that is a stereotype disconfirming, that seems to fly in the face of what we already believe. And that's where the problems start to arise.
So we can't escape the fact that we stereotype. I don't think anybody can, but we can become more aware of the process and we can be more willing to question the stereotypes that we hold.
Michael: Yeah. I think ultimately that's the lesson, the takeaway from this book. And I want to circle back around into that because it's very practical and I think universal in a way that balances out a lot of the sort of more airy abstract theoretical stuff that goes on here very well. But I want to dig a little bit more into the actual dynamics of this stereotyping and how you and O'Flaherty have exposited this in the book.
One of the things I found most disturbing about this was that you mentioned that the simple awareness of a cultural stereotype linking for example, as you do, African-Americans to violence, is a predictor of the extent of bias, whether or not the subject in question personally endorsed that stereotype. You reminds me of the recent research on how even if someone is being told they're given a placebo, the placebo effect is still in place. And so there's a sort of larger game theoretical issue around matters of openness and public information sharing and transparency around exposure to what you might consider problematic or toxic ideas. That seems to be aggravating this.
Rajiv: Yeah, that's right. And I'll give you a couple of examples just to make it really concrete. So, for example, when you think about what we call crimes of appropriation — and this is not how the FBI classifies them, but if you think about crimes where people take other people's property — so burglary, larceny, motor vehicle theft and robbery, these are the four major crimes of appropriation. Only robbery involves face to face interactions. It's in this circumstance that stereotypes come into play. Robbery involves a sequence of quick decisions. You're interacting with strangers on both sides of the interaction. People can't draw up on letters of recommendation from people that others have interacted with before. So if I confront you and ask you for your wallet, you have no information on how I reacted when somebody refused to give me their wallet in the past.
Similarly, when I'm choosing whether to confront you or not, I have no idea how you're going to react to me and I can't go and ask for a reference from the last person who confronted you. So we are operating under a great deal of uncertainty.
In this kind of situation, I'm going to target people that I stereotype as being more compliant. And furthermore, you are going to be more compliant when you're interacting with people whom you stereotype as being more likely to force compliance if you were to refuse. And this ends up affecting who engages in robbery, how robberies play out, and so on and so forth as compared to let's say burglaries or motor vehicle thefts or shoplifting, larceny and so on.
And the thing is that people who are stereotyped, people who are negatively stereotype, so this is…they face a lot of dangers in our society. So people who are stereotyped as being violent will be more likely to face, let's say, lethal force by police. They are more likely to be killed in escalating disputes because if people fear them, that creates incentives for them to be killed preemptively and so on. We can get into this later if you like.
So there's a lot of disadvantage in our society to being negatively stereotyped and being feared. But one environment in which being feared actually turns out to be lucrative and advantageous, is precisely robbery. If you're feared, even if there's no reason for people to fear you, it's a complete misconception. It makes robbery a lot easier than burglary and motor vehicle theft. You ask people for money and they will comply with you because they're afraid of you and so on.
Now at the same time, robbery ends up leading to arrest and prison admission with much greater frequency than burglary and motor vehicle theft. This is part of the discussion we had about disparate impact industry treatment is relevant here. Why does robbery lead to arrest and prison admission with higher likelihood than burglary or larceny? Because it can be reported more quickly. You can give a description to the police, and so on and so forth. So even if there were no bias in the operation of the criminal justice system, people who commit robbery are more likely to be caught and punished for it than people who commit burglary and motor vehicle theft.
Now, if people who are negatively stereotyped are more likely to commit robbery and they are the ones who are more likely also to be brought to justice, then the public imagination gets shaped by whom they see being arrested, whom they see being admitted to prison and whom they see within the incarcerated population. And this just reinforces the same negative stereotypes.
So you end up with a system whereby the offending population is different from the population of arrestees and the population of those who are incarcerated in ways that reinforce the same stereotypes and keep these patterns in place. That's the way in which things are going on under the surface that may not be quite so visible unless you scratch the surface a bit and look underneath the hood.
Michael: You mentioned elsewhere in the book, and you just touched on this, that similarly, feeling as though you can be the victim of violence with impunity, that you live in a neighborhood where the police are unlikely to actually show up and protect you in a case of necessity, means that you're more likely to actually strike first preemptively.
Rajiv: This is, I think one of the really important lessons actually that we are trying to convey in this book. And it's that murder is the only serious crime that you may commit simply because you're afraid somebody was committed against you. It has a strong preemptive motive. And this has been recognized at least since the work of Thomas Schelling in his 1960 book The Strategy of Conflict. Basically, homicide has a preemptive motive. Even if you gain nothing from killing somebody, you'd rather kill them than be killed by them.
And what can make you fearful of being killed? Well, one thing that can make you fearful is if you think you're going to be killed with impunity. If the person who kills you is not going to be brought to justice. And in the United States, historically for centuries, for a variety of reasons, when the victim of homicide is African-American, the likelihood that the offender, whether the offender is black or white, the likelihood that the offender will be brought to justice has been significantly lower than when the victim of homicide is white.
And what that means is that, that you can be killed with impunity more often if you're black, than you can be killed with impunity if you're white. And this changes the level of fear in interactions. Now, why is it, why is it that in this day and age today, the homicide clearance rate — that refers to the proportion of homicides which ended up resulting in arrest of an individual suspect — why is it that the clearance rate for homicides with black victims is much lower than the clearance rate for homicides with white victims? Well, there are two stories that are told about this. You can think about one as being the community view and one is being the law enforcement view.
So the community view is that people don't care about black lives. So if a black life is taken, the amount of investment of resources, law enforcement resources in trying to solve that murder and try to bring a perpetrator justice is going to be lower because people don't care as much. And so you're going to get more killing with impunity.
The law enforcement view is that, no there are strong incentives for homicide detectives to solve every single homicide, but they don't get witness cooperation. So it's much harder for them to solve homicides if witnesses are unwilling to come forward because without witness cooperation, it just makes the task of identifying and bringing a perpetrator to justice that much harder.
Now, I think there's truth to both these perspectives. Certainly, historically it has been the case that, there's plenty of evidence in the book that people just haven't cared so much when the victim of homicide is black. But it's also true that witness cooperation is less forthcoming on average when the victim of homicide is black.
Now let's think about why that might be. So, if you think about New York City's stop-and-frisk policy, which had at its peak about a decade ago, about 600,000 police stops per year were conducted. And that's down to about 10,000 now as a result of some court cases and the change in administration.
And in New York City, when you stop somebody, you're not allowed to do it simply based on race or gender or ethnicity. There has to be reasonable articulable suspicion. It's a lower standard than probable cause. But it's still a standard. You'd have to articulate why are you stopping this person? There's a form that has to be filled out.
Now, if you look at those forms and you look at the cases where the suspicion was that the individual had a weapon, so you're stopping them under suspicion of having a weapon. About 97% of those stops did not result in weapon recovery. So you were mistaken about the fact that the individual had a weapon, which means you're stopping a whole bunch of people who are innocent and are going to be aggravated by that. And to the extent that these stops disproportionately affected African-American and Latino civilians, which there is evidence that they did, you’re going to have a whole bunch of innocent people that distrust and even despise the police.
Those people are not going to be likely to come forward as witnesses, even for very serious crimes like murder. And so you get this situation where generalized aggression in policing can give rise to a greater ability possibly to solve minor crimes, while making it harder to solve the really big ones like murder.
Michael: This is a real Gordian knot, it seems, a real rat king of problems. A perfect example of complexity in that there doesn't appear, at least to me, to be an ultimate cause that we can point to or an ultimate single silver bullet solution to this. But…
Rajiv: Well, there's been work actually by some computer scientists at Stanford and others trying to see, well, what would be the consequences of having more carefully targeted police stops? Reducing the number of police stops, applying a probable cause standard rather than the reasonable suspicion standard? And their claim, they tried to predict the likelihood of contraband recovery or weapon recovery in any particular stop. They have models that do this, and they feel based on their data analysis that you could recover most weapons with substantially fewer stops.
So, there are policies that are suggested by this kind of reasoning that one could scale back, apply a strict standard, so that fewer innocence have their lives disrupted. Because ultimately better police-community relations are critical if you're going to have witness cooperation.
Now, there are many other reasons you may not have witness cooperation. People may have a historical memory that prevents these policies from leading to substantial improvement in police-community relations. But there are policies that are suggested by the data that aren't all that difficult to implement, or all that complex.
Michael: I'm thinking, I guess, of…we had one of our postdocs here, Albert Kao, who published recently on how the longer the memory of the individual nodes in a network, there's a threshold at which the network actually starts getting dumber in the aggregate. So I wonder, that seems to extend to issues of what you might call like a kind of overfitting of people that continue to select into the reinforcement of stereotypes. And almost a point in favor of the half-life of cultural memory and allowing us to be flexible enough as a society to update our beliefs about each other.
Rajiv: That's a really interesting point, and I should look at that work actually, because I do think that the longer the cultural and historical memory, the more difficult it becomes to adjust to changes in policy. You react as if the policy had not changed essentially. And this goes back to what we discussed earlier about confirmatory bias, this phenomenon that psychologists have identified, where evidence that confirms stereotypes is more likely to be accepted and evidence that disconfirms.
And so they may have been a time where a certain stereotype, whether it's stereotype of bumps, civilians by police or whether it's stereotypes about police by civilians, there may have been a time in which the stereotype was reasonably accurate. And if that memory isn't allowed to fade in the face of changes in behavior and changes in policy, then it's going to cast a shadow on current interactions in ways that make policies actually less effective. You're absolutely right. That's a relevant reference to bring up.
Michael: There's another thing that I'm really curious about here because so much of the horrific exchanges that you talked about in this book as occurring between people who meet each other face-to-face seem to be mirrored in exchanges that people have with one another online. And when people…just, as someone who has to spend as much time as I do on Twitter watching the carnival funhouse mirror, nightmare of inflammatory identity-political conflict, online. I wonder how you see that setting being similar to, and being different from, in-person settings.
Rajiv: Well, one of the things we point out in the book is that, the Internet has caused certain things about individual histories to become permanent in ways that they weren't in the past. If you did something bad or something good and your community or your neighborhood knew about it, they'd have a more balanced picture of what you've done. And what happens with the way in which information is stored and shared now is that certain acts get permanently assigned to certain identities and you don't get a complete picture of individuals and what the consequences of this are really have yet to be fully thought out and understood whether or not Twitter leads to greater polarization or not, there are studies ongoing about this. We're not really quite sure yet. Certainly you see that the impression one gets is that it's extraordinarily polarized.
But I'll give you an example of something that's on Twitter just yesterday and it was by a journalist. I forget the name, but describing a whole bunch of direct messages that they got from Latinos and Latinas in response to some of the rhetoric about immigration that's been very salient over the last couple of years, especially in the light of the El Paso massacre. And one of the things that was described was about a couple of individuals who had a Latino taxi driver and gave him a fake hundred-dollar bill as a tip and it was folded so that it wouldn't be realized soon enough and they were laughing as they got out of the cabin. So on and so forth. And that made me think a little bit about what we assume about online interactions as opposed to the older, more offline interactions.
If you think about, let's say just taxi cabs, there was a lot of discussion at one point and perfectly legitimate about taxi cabs profiling passengers for a variety of reasons. What they expected about the size of the tip, the likely destination, the likelihood of being held up and so on and so forth was causing individuals to stereotype past some folks in favor of others and so on. And that was a terrible thing in New York City has had countless complaints about this historically and as attempted at various points in time to try to address this problem. And there was some folks who suggested that, well, once you have Uber and Lyft and in your online cabs, perhaps the anonymity of that process may cause that discrimination to diminish.
But once you think about how the reward system works in these online networks, it's possible that somebody who gives you a perfectly good service but belongs to an ethnic group to which you have any version might end up receiving a low rating from you for no reason other than the fact that you have some animosity towards them. Similar to the idea of playing a joke on somebody by giving them counterfeit currency as a tip. And that can give rise to very distorted statistics such as ratings of cab drivers or delivery persons on online.
So this is all very new and we're still trying to understand these effects. And I don't think we do much in the book with this, but we do mention it.
Michael: Well, one way that I see a really strong correlation between the online interaction and the in-person interaction on the street is that, as you make the case, so many of these...you give an example in the conclusion of a young man who was in a failed carjacking and accidentally shot himself and then in the ambulance confesses to two crimes earlier that week and is sentenced, at the age of 15, to 25 years in prison.
And there's something about the lifelong consequences of choices that are made under conditions of duress or coercion, or at least decisions that have to be made very fast about each other. And that there seems to be an issue about the time horizon and like you said at the beginning that, so much of this is about us not having the time to proof and check everything and investigate every experience as a unique experience and investigate every person we meet as a unique person. And that there's something about the way that we...whether it's in person or it's on the Internet, about the way that we signal and that we present to one another. That it seems almost as though the book is a map of the tragedy of the limited time that we have to evaluate each other and the decisions that we make.
Rajiv: Yeah. Actually in a review of the book The Economist, Diane Coyle called it sobering. And so in a sense it's not most uplifting book to read. We just laid out the situation as we saw it.
The incident that you described, actually it's the person there who as a juvenile was tried as an adult because in the ambulance he confessed it to, two robberies, involving very small amounts of money. And then was prosecuted under the Three Strikes law in California. That person actually is the cousin of the great political philosopher, Danielle Allen, who is a University professor at Harvard. And the incident is described in her recent book, Cuz. I won't go into too much detail about the story of her cousin Michael, but he died at a very young age after release from prison. He was tried as an adult and really it was a botched robbery, not a carjacking. But it was, yeah, somebody who was attending to their car. Right. So there was a car in the picture. And that's a book I highly recommend.
But what Danielle does in that book, first of all, she humanizes her cousin. She knows all the different aspects of his character. What's in the public record is just a very limited story. He was a very courageous firefighter, a lot of incarcerated population in California is called upon, to come out and fight fires in an emergency. And he did that to great effect. He was also a poet. He wrote some remarkable things, but had Danielle not humanized him in that way, the rest of the world would just know him as a juvenile offender.
And actually there's another very popular and very important book by Bryan Stevenson, which many of your listeners would know, Just Mercy, which does this over and over again with many, many people. And this process of humanization is absolutely critical. I think if we're going to have substantial movement towards the criminal justice system that fits with international norms. We’ve departed so far from it. About two or three generations ago, the US incarcerated population was not that great relative to the broader population when you made international comparisons. Now it's such an extreme outlier. There's nobody even close, I think, Russia second, but it's still far behind.
And this process of humanization, partly because of all the way in which the racial and ethnic divisions in our society are mapped out in the way in which they reflected in the way in which people have attitudes towards those who are incarcerated, the humanization processes is really important.
And one other point I'll make is, this is a reference to a different book by John Pfaff, who's at Fordham Law School as a very good recent book called Locked In, where he points out, although a lot of the discussion nowadays is about decarceration of nonviolent offenders. But you're not going to have, John Pfaff points out, and correctly points out in my opinion, that are not going to have significant decarceration unless you deal with the length of sentences and the severity of punishment for people who have committed crimes that are classified as violent. They may have been present when murder was committed or an aggravated assault or a robbery which is classified as a violent crime.
So the incident that you mentioned is part of this story of humanization that I think is really important if decarceration efforts are going to really take hold going forward in time.
Michael: To pseudo-mathematize this, it seems as though everywhere we look in this conversation we have runaway positive feedback loops.
Michael: And that the solutions, plural, that have to be brought to bear on this situation are about interventions as early as possible in the runaway positive feedback.
Rajiv: Yes. In fact that's what makes the situation sobering, actually because positive feedback loops lock us into equilibrium. If you want to describe it that way in mathematical terms that are hard to shake, it's hard to escape. We are stuck in some ways. Marcellus Andrews, another economist who's a friend of mine, describes the world we live in as a stereotype trap. The stereotypes have incentive effects that cause people to behave in particular ways and give rise to certain patterns in the data that, because of what is hidden versus what is revealed and because of our psychological machinery end up reinforcing those very stereotypes. And I think the idea of a stereotype trap, it's a very good way to describe the condition that we're in, makes it very hard to escape.
I mentioned to you earlier some policies that could be implemented fairly easily, at least with regard to generalized aggression in policing, greater clearance rates for homicides, better community-police relations in order to elicit greater witness cooperation.
There's a bunch of stuff we discussed in the book about gun laws as well. This is topical now with what's happened recently in El Paso, in Dayton. And so there are things that can be done. Things aren't as sobering, I guess, as the first appear. But you're right, there's a sense in which the positive feedbacks lock us into a situation that then becomes hard to escape. But our approach is that we need to first understand the processes if we're going to do something about them. And most of the book is just trying to lay out things that are below the surface and that aren't easily exposed.
Michael: Earlier you mentioned the street level interaction of a robbery and how both parties in that interaction are sizing each other up. Listening to you talk about that, I had this sort of absurd thought that, like, what if we had a Yelp for criminals, you could be like, “In spite of the fact that I was coerced into giving that person my wallet, it was relatively pleasant exchange.” But then of course you get into this whole thing about how that is just another venue for another potential runaway chain reaction where the ratings end up reinforcing a public image for someone else. This gets to some of the other work that you've done more broadly on public disagreement.
Michael: And on what it means for there to be public information and private information. How we rate each other in terms of being well understood about our transparent biases, or well-informed. I'd love to hear you open up into that.
Michael: And talk a little bit about that and how you see those two issues relate.
Rajiv: Sure. I'd be happy to do that. And this is joint work with Muhamet Yildiz who is an economist at MIT. And really it's also connected to what we've been talking about because it involves identity, in some respects. So again, to go back to what economists typically assume about the statistical accuracy of people's beliefs, there's an assumption — I won’t go into it in too much detail. That's referred to as “the common prior assumption” in Economics, which essentially says that when two people disagree about something, it's because they have different information. So it's not because they have different biases or because they have different prior beliefs in a Bayesian or statistical sense. This is called the common prior assumption. We both reason from the same prior belief, we receive different information. We'll update and have disagreement.
But one of the consequences of that way of thinking, which was highlighted very beautifully in the 1976 paper by Robert Aumann, was that although we can disagree, we can't agree to disagree. In other words, the fact of our disagreement ought to be informative. The fact that we disagree or to tell us that we have different information and we ought to then update our beliefs and come closer to each other actually. And if we still disagree, that means we haven't fully absorbed the information that the other party has received and so on and so forth.
There's a very nice paper called, “We Can't Disagree Forever” by John Geanakoplos and Heracles M. Polemarchakis. And John Geanakoplos as a longstanding association with the Santa Fe Institute, as you know. And this paper really takes Aumann's idea and says, look, suppose that two people with a common prior have different information. And they just announced their beliefs repeatedly, sequentially, what's going to happen in the long run? And they show that, in a well-defined sense, they will converge to the same belief in a finite number of steps, under a certain set of circumstances.
Now that's a very nice mathematical result, but that's not what we see in the world. We have public disagreement all the time about various issues and in fact. And recognize this and the purpose of his paper was really to lay out the assumptions under which you would not get that. So that when we see the public disagreement we can start to think well which of Aumann's assumptions as being violated.
And my work with Muhamet sort of departs from the standard setup by allowing people to have heterogeneous prior beliefs, so that the starting point on the basis of which we then process information may differ across individuals. And in that case we could end up with different beliefs even if we have the same information, and we could end up with public disagreement. And a lot of what we do is to think about, well how do we then update? When I observe your opinion about something, it's conflated because your opinion is the result of two different things. It's your prior, which from my point of view is just a bias, right? If it differs from my prior belief.
But you also have information and I understand that you have information. So for me to make an inference from your opinion, I need to try to untangle what is a bias from my opinion, from my point of view, and what is genuine information that is valuable to me. And so what Muhamet and I have done is try to look at this process.
And one of the questions that we look at is, well what does that imply for, where are we going to seek out information? So this goes back to your point about Twitter and polarization and so on. And really this is one of the areas of my work that I still find very absorbing and interesting. So I'm glad you brought it up.
This work illustrates a certain tradeoff that hasn't been explored much in economics and we think not much in social science more generally. And it's a tradeoff between people who are well-understood and people who are well-informed. So this is a tradeoff that you mentioned.
What do we mean by well-informed? Well, when informed just means that you have good information about the world. So, for example, an economist may be well-informed, let's say about the unemployment rate or the process by which the unemployment rate is computed and so on. A climate scientist may be well-informed about the degree to which a business-as-usual scenario is going to give rise to an increase in global temperature and which areas will be most effected. So this is what we mean by well-informed, which is the common understanding of the term.
But it's also important to think about people who are well-understood. So somebody is well understood in our setup if their biases are accurately understood by an observer. So it's a relational property. So, I may know that you're very well-informed because you're an expert in climate science, but you'll still be very poorly understood by me if I don't know your biases. If I'm very uncertain about whether you have a predisposition to believe one thing versus another about climate change. And if you are poorly understood by me, despite being well-informed, I may decide that you're not a good source of information for me. I don't know where you're coming from. I don't know where your biases are and I might actually go and seek information from somebody who is less well-informed than you, because they are better understood by me.
So this tradeoff gives rise to patterns of observation both online and offline, whose blogs I read, which newspapers I subscribed to, which TV shows I watch, whom I follow on Twitter and so on, which might reflect the fact that I follow people who are more poorly informed than others, simply because I have a better idea of what their biases are, where they're coming from. I can better interpret their opinions and back out the information that those opinions contain.
And this is something that arises even if my only goal is information. It's not just reinforcement of my own views, which may also be that a phenomenon that is at play here. But even if my only goal is to seek information, I may still seek it from people who are less well-informed because their biases are more transparent to me. This is just sort of an overview of the kind of themes that we're exploring.
Michael: It's interesting that there's an economy in that decision.
Rajiv: There is.
Michael: Which is to seek out sources of information that require the least effort to understand.
Rajiv: That's right. And actually...so here's something that we find in our research, which is that, when I observe your opinion and I try to make an inference from it, I learn about three things actually. I learn about the world, which is what I care about. I learn about you. I actually learn something about your bias too. And I learn also about what we call your culture. I learned something about the biases of people who have similar biases to you. And what that does is it makes me more inclined to seek information from you in the future. By observing you, you become better understood by me and not only you, but people who are similar to you also become better understood by me.
And so there's this, again, we talked about positive feedbacks. There's a tendency for people to seek information from sources that they have consulted in the past with greater frequency as they move forward into the future. So there's another path dependence, another positive feedback effect, in that respect. And there's an irony there because you may be actually well informed today about an issue that I care about. And so I seek information from you that allows me to better learn your biases and tomorrow I may seek information from you even though it's about an issue about which you're quite poorly informed, simply because your biases are more transparent to me. So I hope that was reasonably clear.
Michael: Yeah. There's another angle to this that I've observed in interpersonal relations, which is: a lot of people, whether it's in this very real and grave criminal justice context or whether it's just in intimate relationship, most people want to feel understood. However, there's labor involved in becoming understood.
When we talk about the economics of this, like part of it is the effort expended in educating a stranger or even in an intimate relationship. And that seems to be another obstacle toward the kind of process of restorative justice more broadly. And maybe more personally just people coming together in synthesis and good faith dialogue.
Rajiv: Yeah. So let me give you another example from the book that's also very well known actually. This goes back to a memoir by Brent Staples who's the writer for The New York Times. He's also a psychologist by training as a PhD in Psychology from the University of Chicago. And there's a very famous incident that described in a memoir that he published in the 1990s where he would take walks along the lake while he was a graduate student and would find people visibly afraid of him. They would be clutching their bags or they would be trying to move away across the street, or just seemed to be fearful of him. Although he knew in his own mind that there was nothing to be afraid of. And he's a graduate student. He happened to be African-American, tall, and so on.
And he found himself inspiring fear in passers-by as he walked down past the lake, and realized that he's being misunderstood, and it bothered him, it upset him. And there are similar incidents that are described over and over again. You can find many instances of this there's a book by Paul Butler who's at Georgetown Law School who describes very similar kind of feeling of being feared over and over again. There was a speech by the South Carolina Senator, Tim Scott, who is very conservative on most issues with a very emotional speech on the Florida Senate where he himself was treated with suspicion even by staff who were screening people entering the Senate building. So this is a very common occurrence.
So what's interesting is the investment that Staple's made in trying to shift this stereotype, and you talked about investment in trying to become better understood. So what Brent Staples did, and he describes this in his memoir is that he started to whistle — initially out of nervousness. And he realized that he was in tune and very good at whistling and he started to whistle Vivaldi, The Four Seasons by Vivaldi, and tunes from the Beatles. And this whistling in tune, which people recognized completely shifted the stereotypes. So people would smile at him, people would nod at him, instead of being afraid. So he was able to shift the stereotype to an investment, if you like, into something that caused him in his own mind to be better understood, which was one of the goals that he was trying to achieve.
But what's less well known in the very same memoir is that, Brent Staples just got a bit fed up with this — and you can imagine the frustration having to make an investment so that people are not afraid of you. And he decided for a time to act in stereotype-confirming ways, and just to act in ways that would make people more afraid. I think it was just the frustration with the fact that he was having to bear these costs in order to dispel stereotypes that he felt should not have been applied to him in the first place. Not going out and threatening assaulting or robbing people, but simply not taking the trouble to sound and look harmless and friendly.
And he saw that he was able, just by his posture and just by his demeanor, to instill fear in people. And he came to the recognition, actually, and he says this explicitly in his memoir, that if he wanted to rob somebody, all he would have to do is just stand there. Not even asked them for the money they were just handed over. They were that afraid of him.
And this actually got me and Dan also thinking about the incentive effects of stereotypes, and really, for somebody like Brent Staples — and of course he was getting Doctorate in Psychology; he had no interest in robbing people — but if somebody were interested in taking appropriating the property of others, if people were that afraid of you, even though there's no reason for it, it becomes much easier for you to engage in robbery than burglary or motor vehicle theft. And if you are somebody who cannot instill fear in people, then you either have to make investments in menacing facial tattoos or something, or you're going to go and choose other crimes like burglary or motor vehicle theft or larceny.
And so you're going to end up beneath the surface with the sorting of individuals and their crimes where people who instill fear are going to choose robbery. And people who don't are going to choose burglary. And as I said to you earlier, robberies are much more likely to lead to rest and prison admission than burglaries. Burglary is a much harder to solve. You don't even realize you've been burgled sometimes until you come back from vacation. And, that reinforces these stereotypes. That's back to the positive feedbacks that you described earlier.
In the positive feedbacks have been a theme that has been explored at the Santa Fe Institute all the way back from its inception. Brian Arthur's work in the 1980s and so on. And so in a sense this sort of fits with that tradition, but it is sobering as Diane Coyle said to think about it.
Michael: There is another SFI External Professor, Raissa D'Souza. And Raissa’s work when she recently presented at the Science Board Symposium, this video’s up on YouTube about The Collapse of Networks and how as a network grows, the number of low degree edge relations grows and it becomes easier and easier for a cascading failure to propagate through the network. And not to flog this issue, but it does seem as though the scale to invoke Geoffrey West, another important SFI researcher and his work on the scaling laws of cities and how the more interactions we have per day the faster our lives move, certain properties emerge out of that. And one of these properties seems to be that, we are having more and more of these ephemeral, fleeting, seemingly inconsequential encounters with strangers. And it seems to me as though one of the things that we see in the polarization of a global conversation is people feeling as though, again, that they can speak with impunity to one another.
It's precisely the opposite of what you've observed in this conversation, that the consequences are even more grave because of the sort of permanent record that we now all possess. And so, I mean, how do you see this as linked to this particular issue of the structure of networks and does that offer…I want to find a way to segue this into what I think is the most important place where we could possibly end this. It’s where you and Dan write in the book on the issue of hope. Does seeing the problem clearly inspire hope, and does the fact that it seems as though our society is only getting faster, and that modern society perhaps gives us fewer affordances for the kind of deep and human relationships we need in order to undo this kind of a thing. Like where are we in this mess?
Rajiv: I wish I knew. I think it's the crucial question really of our age, right? It's really going to determine where we go from here. I used to be more optimistic generally about communication that the truth will win out. I’m a bit of a First Amendment purist in and disregard and have become increasingly pessimistic about the ability to instill false belief in others strategically. The ability for us to withstand that effort seems to be limited. We need mechanisms in place whereby we can clearly identify trusted sources. And I don't think we have those yet.
Going back to the issue of hope in the book, again, I think I'm predisposed to be optimistic, and so maybe I'm biased over here. But to us in the book, it's connected to humanization and we feel that there are efforts — by Danielle Allen as I mentioned to you, by Bryan Stevenson — to humanize people who may have committed even heinous crimes in the past to reveal them, to have redeeming features, to reveal them, to be capable of rehabilitation.
So hope and humanization are really linked to each other. Now we live in a time where rhetoric involves a lot of dehumanization. And recently we've heard a lot about people being characterized as vermin, about infestations, about invasions, and this is deeply dehumanizing. And the question really is whether or not it's going to stick or it's going to provoke a backlash that will cause us to find this kind of rhetoric intolerable. And I think we're going to see that play out over the next few months and years. And that's going to tell us, really, whether there's basis for hope. But we have a battle right now between dehumanization and humanization. And I am, I would say, pessimistic in the short run and optimistic in the long run. But that may reflect the stereotype on my part, and I'm open to revisiting that and changing my view on it as time unfolds and as evidence becomes available.
Michael: Yeah, I mean when you talk about there being, in the two scenarios, one in which members of a disagreement have common priors and one in which they don’t, obviously it's not a binary. And if we dig deep enough, I'm sure that we can find common priors. And that the depth that we have to dig to find them is therefore the variable that allows us to calculate the actual curvature of that long arc of justice in history. Right? How far back do we have to go to find a common ground upon which we can discuss our differences?
Rajiv: Yeah, that's an interesting perspective on this. That was Harsanyi’s original defensive of the common prior assumption: that, we should if we see disagreement, we should try to understand it and trace it to particular factors. That there's no basis for assuming heterogeneous priors to begin with. We found it useful as a modeling assumption because we want to explain disagreement, which, really can't arise under the common prior assumption, and various auxiliary assumptions that I usually made. But that's an interesting way to look at it. Whether or not there's a basis for hope is really tied to the question of whether or not we can see in each other our common humanity.
Michael: Well, Rajiv, this has been an amazing conversation. It's an honor to sit with you and talk about this. Thank you for your time.
Rajiv: Thank you for giving me the opportunity.