RICHARD THALER: Here's a list of potential topics. We're going to cross this one off, because that's what Danny's going to talk about now, and so there are two things that I could talk about. One is a paper I've done about the National Football League draft that is showing that team decision-making doesn't seem to be too good, and they over value picking early. That's the one-line summary of that.

MYHRVOLD: Cross that one off.

THALER: Are you going to say that for all of these? The second one is I could do something about the fear of the behavior of finance, and then Sendhil has a talk he could give about mental models. Say a sentence or two about what that's about.

MULLAINATHAN: It's about how there's a large set of programs where we're trying to encourage behavioral change in some form, and one of the underlying themes that emerges that unifies them is can we effectively elicit the mental model people operate with and then use that to then create behavioral change. It's a structured way to think about the behavioral change literature.

THALER: I'm not sure how we go about aggregating preferences. We could pick for you.

MULLAINATHAN: We should draw lots and have a dictator.

THALER: Tell you what. You can think about this and submit an anonymous vote of preference if you have one that we'll aggregate.

MYHRVOLD: But then later you're going to show us we made a dumb choice.

THALER: But those are your choices, or you can sleep in. If you have a preference, why don't you write it down on a piece of paper and give it to us, and then we'll, at five, we'll figure out which one we're going to do. Okay?

KAHNEMAN: Why can't we have a vote?

THALER: That's what we're doing.

KAHNEMAN: In a complicated way.

THALER: In an anonymous way.

LECLERC: In an anonymous, complicated way.

THALER: Yes, okay.

MULLAINATHAN: Vis-a-vis, the sleep in. You can also sleep during the talk.

THALER: Are there unfinished aspects of ...

MULLAINATHAN: I should clarify one thing. Given how I started, I didn't make one point clear. I'm not trying to argue that strange things can happen in markets, our conversation clarified this. I'm not trying to argue that strange things can happen in markets. I'm trying to ask the question that you put very well. Can we put structure onto this problem so that we can start to define when "strange things" will happen and what those strange things will be. That's the idea that this can happen, not surprising to some, perhaps surprising to some, but it‚s not that useful so much as to be able to articulate when and what, and these experiments are helpful by being able to tell us that. That's the hope, and to be very clear upfront, we're there yet. I'm going to try and end with conjectures that Dick and Emir, who we're working with, and I have. We can throw those conjectures around a bit, but that's what puts structure on this problem.

HILLIS: Last time we did this event I thought that we learned that the value that people assigned is not. It's interesting that they depend on the way that the choice is presented, but it even depends on information that's presented to them after, of how the question is put to them after they've made the choice, way far later or something like that.

MULLAINATHAN: One of the things that's very, very difficult about bridging psychology with economics is that economics is too trivial, and psychology is too rich, and so the challenge is to figure out ... since we're trying to build some structure ... is to figure out, to explain a class of things what part of the psychology we use. If we open the bag or the box of valuations, there's a lot of things that go into that component, into that box, and what I've tried to do is I've tried to pick the ones that are most important for this set of examples that I'm talking about. That's a choice that I may have made incorrectly or correctly, but that's the reason why I'm presenting only a subset of these things.

HILLIS: Yes, that answered my question.

ROMER: He's also saying, to some extent, that there's a lot of problems that classic economics solves really well, perhaps too well, and that behavioral economics, in order to say anything, needs to be able to define a set of problems that it can solve better.

MULLAINATHAN: That's exactly right.

THALER: And complicating the model by making the Econs more like Humans makes the economics a lot harder. There's a reason why economists invented Econs. It is that optimization problems are easy to solve, and so, you have a hammer ...

KAHNEMAN: Especially if you make it uni-dimensional.

THALER: Well, yes. There's a story I like telling. Twenty-five years ago, giving a talk about this stuff at a National Bureau of Economic Research conference, and one of the economists there, after hearing me talk, says, "If I understand what you're saying, what I know how to do is solve optimization problems, and what am I supposed to do?" And I said, "That's your problem. Not my problem." Right, it's harder, but your comment was very well taken.

DYSON: I got a quote here that goes back to the beginning of this. I used to work on looking at how John van Neumann started "Theory of Games and Economic Behavior", and at the beginnings of it, if you disregard his early paper on the Minimax Theorem, is a letter, unknown, unpublished, dated December 9, 1939 to Stan Ulam where he says ... these things always start with a question. "I refuse to accept, however, the stupidity of the stock exchange boys as an explanation of the trend in stocks. Those boys are stupid all right, but there must be an explanation of what happens which makes no use of this fact."

MULLAINATHAN: That's fabulous.

DYSON: That's how his 525-page book starts.

MULLAINATHAN: I never heard that, but that's a great quote.

DYSON: It's unknown.

MULLAINATHAN: Should I continue ...

THALER:... is that in the book?

DYSON: No, no. That's a letter.

MULLAINATHAN: Let me continue here. I will try and be quick. What I wanted to do is give you a sense of a few more experiments. For no reason, we built this case that choice sets have an effect, but I want to have these experiments so you have these in the back of your mind. They're interesting in and of themselves as useful experiments. And so far, what we have is it's hard to translate some characteristics into value and the choice set can affect these translations. Okay, that's where we are.

Here's another interesting example where the choice set affects translation. This is abstractly put. I always remember this example. If you're looking for apartments and there are two apartments, and one is close to campus, more expensive, and the other further from campus, less expensive. Which one do you want? You pick one.

I add a third apartment that is dominated by this one. You can imagine when I add this, a lot of people switch to that, and that's another situation. It's not an exchange rate issue. This is an issue where the presence of little p is adding a feature. I use this example. At least one way to understand this example is, so far we've been talking about how do we translate characteristics back to values, but they're often characteristics that you construct from within the choice.

Here's a way that people may go about it. Gosh, do I want big Q or big P? Well, big Q is closer to campus. Big P is cheaper, but big P has another good feature it's better than little P. I'm creating a feature in the choice set, and so it's not just on the translation of the effect, I'm now creating a feature, and that's a pretty common thing. I'll go a little faster over this, but if you introspect, you'll see there are many examples where you create.

MYHRVOLD: How different, though, is that from saying that you're learning something about the world? In fact, what you're doing is you're saying that my prior distribution over what's possible might have been wrong, because it turns out there's this thing that's off the line.

THALER: Our co-author thinks that's important.

MULLAINATHAN: There's definitely an element to this, which is learning, but we have to be precise in what you're learning about.

HILLIS: But what are you learning about the world that's relevant as a choice? You're learning something, but it's an irrelevant something.

MULLAINATHAN: Exactly.

MYHRVOLD: Well, you're learning a relative ranking of those things. I mean, the likelihood that big P is the worst thing out there is now gone.

KAHNEMAN: Well, but how much value does this have if you know that somebody put that thing in there in a strategic context. It's worth nothing. You've learned nothing, because it's strategic. You should be able to discount it.

MYHRVOLD: Well, there are two questions. One is, is there an Econ on the other side of this game trying to fool me? Yes. If you take the paranoid perspective, you're right; you didn't learn anything about the world that bastard didn't want me to know. Therefore, I shouldn't trust him. I'll know to act that way around you from now on.

THALER: No, just always assume, always assume there's an Econ trying to screw you. That's an axiom.

MULLAINATHAN: I want to table this question for a second, because your response is very important, because we have to be very precise about what's being learned, and it's more subtle than just learning, but I don't think it's important for the story that we're telling here for the reason that whether people are engaging in this behavior very sensibly or non sensibly, what we're interested in is, what is the emergent market outcome, and the very micro foundation of this stuff is a bit of a distraction. It's important, I'm not saying that, but it's a bit of a distraction.

We did this experiment with milk, of all things. I'm showing it with Powerade and Gatorade and orange juice, but you can do this with quarts and liters of something as mundane as milk. If you have a quart and a half- liter ... half-gallon ... a quart and a half-gallon of milk, by adding an eight-ounce orange juice at 140, demand moves towards a 16-ounce orange juice. That's intuitive.

That's what I just showed you. But think of the funny pricing elasticity this produces. If I'm this orange juice producer, it gives me an elasticity of demand that's very odd, and that elasticity is if I raise this price, I'm raising demand over here. Right? Does that make sense to everybody? There's this very strange substitution, and I'm raising demand by drawing demand from here to here. You'll notice that's very similar to a theme we had before where the most important feature of a choice set effect is not that I'm making my product look good. It's that I'm affecting the other guy. Right? There's a strategic interaction element now.

MUSK: You see this in popcorn at the movie theater where there's a buck difference between the jumbo and the little one.

MULLAINATHAN: You know where you see this?

THALER: The little one being this size.

MUSK: The jumbo is like the size of your body.

MULLAINATHAN: Where this drives me nuts is Krispy Kreme. I have to tell the story, because it annoys me. I like Krispy Kreme donuts, but they're horribly tempting. I go in, and I always say the same thing. I say, okay, I'm going to buy four. I pick this one, this one, this one, and this one. And then the friendly person at the other end of the counter says, "It's only 39 cents more for a dozen." I don't want a dozen. But what moron would not buy a dozen for 39 cents more? Oh, well.

BEZOS: By the way, they give you one to make sure you don't buy just one.

MULLAINATHAN: Exactly right.

BEZOS: That's an Econ.

MULLAINATHAN: Exactly.

MULLAINATHAN: I've gotten my pet peeve out of the way.

BEZOS: There's a famous strip club in Seattle. It's been there for decades. It's called ...

MUSK: You know it well?

BEZOS: It's called "The Pink Lady", and their slogan is "99 Beautiful Girls and 3 Ugly Ones".

MULLAINATHAN: The contrast effect rule.

BEZOS: It's been there for decades.

MULLAINATHAN: Let me show you one more study. I'll tell you where all this is going, and then I'll move to another topic. The one more study is that a very, very important feature of choice sets is conflict. People hate to make decisions. That's very intuitive. But one consequence of people hating to make decisions is that you can move them from one part of the choice set to another if that allows them to reduce the amount of decisions they have to make. Okay?

Here's a good example. Patients are scheduled for a ... and I won't even bother saying this ... I have no idea how to say this ... but operating room slots are taken by emergency cases, who has a higher priority? This is a question that was given to a bunch of doctors. They were given the dossiers to patients, patient MS, patient AR, long descriptions. Here I've summarized them for you. There's a 52-year-old and 72-year-old, and they were asked, if asked for your opinion on which patient would you operate on first? Most will say patient MS. Okay? That's intuitive. Why? Because patient MS is younger.

A lot of times doctors say if they have to choose, age is not a bad indicator, a utilitarian value. The utilitarian value changes quite fast when we add patient PK. Patient PK is also a young 55 year-old employed bartender who us different but also young. We add patient PK. Who do you think gets chosen? Patient AR. Why? Because by using my youth utilitarian feature, I now have to decide which of these two. That's a pain in the ass. AR stands out.

BEZOS: Am I the only one who's very depressed by these examples?

MULLAINATHAN: Depressed? Is that what you said?

BEZOS: Yes.

MULLAINATHAN: Oh, sorry.

HILLIS: We can tell how depressed you are.

MYHRVOLD: No, this is an Econ that's in our midst, folks.

THALER: You should be happy, Jeff. These are the people that are making you rich.

MULLAINATHAN: One reason this is important, and this is the theme I want to close on so you see why this is important, is think of the fact that it's in patient AR's interest to have patient PK enter the picture. Okay? If these were three different firms, when PK enters, AR benefits.

PK also benefits because he is stealing some market share from MS. If we were to put this into a competitive landscape, you'll notice that the entry of PK is both beneficial for PK and for AR at the cost of MS. If we held these dimensions constant, that's what would happen. Okay, let me close the loop and say where we are, and then I'll stop this topic.

The thing that matters is what I want to call a choice set effect, where the choice set affects valuation, and so with no choice set effect, you could be wrong. You could be making lots of mistakes. If there are no choice set effects, there's a very simple prediction, which Dick had called catering, and which is intuitive. Whatever the utility function you happen to think you have, the market will maximize that. Okay?

The market will cater to your needs. With choice set effects, it will no longer cater, and there are two scenarios. Scenario one is what I want to call stability. Scenario two is what I want to call instability. Here's a good way of thinking about stability, and somebody raised this earlier.

Suppose that when I'm looking at paint colors, it comes out that if you put a lighter paint in front of me, I move towards that, that's a choice set effect. If you put a darker paint in front of me, I move towards that. Well, even though that's a choice set effect, there is a stable solution. I get the range of paint colors, and I go to the middle. That's a situation, and I can make this precise, where there is a stable choice set where it's no longer in anyone's interest to enter and upset the preference order that comes from that choice set.

Those are situations where we don't get this funny stuff I talked about before, even though there are choice set effects. What we get is equilibrium. Granted, hedonically we may think they're mistakes or whatever, but we won't get anything other than an equilibrium with a choice set and a fixed set. But there's another case, which is very important which is the case of instability. You already see instability potentially in an example like this. Here I did it with one dimension. In one dimension, it looks like what would happen is PK enters and takes a market share. Well, now SK, an older patient should enter and start to steal market share. This looks like this will also tends towards stability.

But if you have two dimensions now, you can end with cycling, and cycling has this essential feature that every time someone comes in, it changes the preference ranking, and then the equilibrium is upset, so then some other guy has a different decision to enter, and it turns out one of the things that emerges is that you can get no equilibrium. The features that generate no equilibrium cycling ... which is the notion ... that's what I mean by people want simplicity but can never get it. That's the sense in which I mean it. That is, if there's no ex post valuability, if there are choice set effects and if there are choice set effects that are good at this particular type of instability.

And what are the markets that have that feature? Well, we've talked about mortgages. Mortgages clearly have a difficultly of ex post valuability. They clearly have choice set effects, because there are characteristics there that are very hard to evaluate. They can have instability if firms have the ability to do the following, okay, which is what I think of as the megapixel problem. The megapixel problem is can they create a characteristic that simultaneously does the two things. One, it looks like it's a characteristic you ought to have, but two, it causes you to under weight some other characteristic. Right?

I add megapixels, and you say, "Gosh, I guess we should have more megapixels," but now I guess weight doesn't matter as much to me, and now you see the germ of cycling begin to happen, and in financial contracting you can get this sort of cycles. Because I add something else, no load, and I say, well, I should get no load, but then I don't pay as much attention to this thing, and the next guy will say I've got no load and I've also got this other thing. You can start to end up with these markets that are inherently always in this soup.

THALER: One footnote on this ... notice if we continue on mortgages as an example, 30 years ago there was essentially one kind of mortgage; 30-year fixed rate. The government passed the Truth in Lending Act that said you have to report interest rates in a uniform way which was APR, and that basically solved all the problems. Choosing the best mortgage, they gave you a sufficient statistic. You look at that number, and you're done. We have zillions of kinds of mortgages. APR is no longer a sufficiency statistic, and now all these things that Sendhil has been talking about are possible in a way that weren't when we had one single yardstick.

MULLAINATHAN: And just to make clear, the reason APR is no longer possible is exactly because there's a strategy which allows you to say sure I have higher APR, but I've got this thing. Don't I look more attractive as a result? You need to have a strategy that breaks the sufficient statistic condition, and that's what starts to generate,

THALER: Or I have the low APR for one year.

MULLAINATHAN: Exactly. Right. You need something that starts to break that which is then a substitute generator. In other words, the stability allows markets to settle on the sufficient statistic, and the instability prevents markets from generating in a single sufficient statistic. And the issues we talked about, the other issue that plays a role here is, in markets where there's homogeneous preferences like roughly, by and large, there's a few characteristics that you evaluate that tells something, then it's very easy for information providers to create those sufficient statistics, for cleanliness of rooms, etc. We've got that, and you had raised this earlier.

When there are markets where there's a strong interaction between my characteristics and the products that are produced, it's not possible for third-party providers to generate that sufficient statistic. Does that make sense to everybody? Those are the four elements.

BEZOS: I'm wondering also if an additional characteristic might be that for some reason in that market or in that industry a competitor cannot emerge who is dominant in all of the attributes.

MULLAINATHAN: That's exactly the instability component, and it's because ... you see, you were taking the valuation of attributes as somewhat given, and ... you're right. If there's somebody who's better on everything ...

THALER: Better on everything. And in many businesses, it's possible for somebody to be better on every single dimension in supporting customers. But there are some, perhaps like mortgages, where you can't do that, because there's a finite amount of money that you have to collect to get a certain return, and you can move it around and shape it differently, but you can't be the lowest APR over 30 years and the lowest APR over one year.

MULLAINATHAN: Right. You know what's a good example of this is that I've often puzzled why are MP3 players so different from digital cameras? You can make an argument about MP3 players. I don't know how to value this. I don't know how to value this. And it's true that the market for MP3 players look like digital cameras 10 years ago, but if you have a company that's able to come in and produce something that uniformly dominates the entire sector, then they have the ability to simplify.

BEZOS: Every variable that customers care about, a product is genuinely better.

PARKER: That's not the case with the iPod. This is an interesting case study because what happened with the iPod is that Apple came along in a market hack, and the hack was that industrial design trumps everything else. If you look at the iPods that came before\, look at the MP3 players that came before the iPod, they, for their time, offered more storage potentially, because you had more flexibility and control over the device, which was thought to be an extremely important consumer dimension, and the ability to put multiple SD cards in, the ability to swap components in and out. These were thought to be important, and Apple came along with a device that's in many ways more limited, but the key recognition was that it's a lifestyle product, and it's something that you're going to be seen with, and so you want to be seen with a product that looks cool.

BEZOS: But you just failed your first test which is you can't hedonically judge the value after the purchase.

MULLAINATHAN: The ex post evaluability is pretty good.

BEZOS: It doesn't even get past gate one.

THALER: Well, also there's this issue of what features do people value. This is the mistake Microsoft continuously makes which is adding features that three guys in the world want and annoy everyone else.

MYHRVOLD: That's why they're so unsuccessful. For years, when people would say things like that to me, I'd always say, "That's so great that you're so concerned about Bill. I'll pass this along.

MULLAINATHAN: Any more issues on this? We can have a brainstorming session later about how to help Bill. I'll pass around the hat.

MYHRVOLD: How well do you think you characterized the set of things where you can get stability and instability? You've looked at some set of things, and the first set of points you made I would characterize as, "Hey, an existence proof!"

MULLAINATHAN: Exactly. That's exactly where we struggle.

MYHRVOLD: But there's got to be some ... if you're looking at this in mathematics, you'd say, "Hey, there's a set of things where the problem is ill-posed." Well, then you got to go back and say, "Well, how many things are like that? Are there new different ways?" How far along are you in the catalog of things?

MULLAINATHAN: One place where we are not so far along is being able to say what are the varieties of instability, which is important. The place where I feel we're further along in the thinking is the empirical end of this. What's nice about this space is that you can offer hypothetical product trade-offs like this. You can start to understand where are the inconsistencies.

The statements that we're making around the table ... that mortgages look like this or mutual funds look like this ... they need not be intuitive statements that we hope are true. They're completely empirically validatable statements where we look at 100 consumer and say, "Please make trade-off between mortgage A and mortgage B," and we can see if, as we hypothesize, are we seeing cycling in the data for mortgages, but not seeing cycling in the data for MP3 players. The empirical validity of this is further along than being able to say, "Here are the varieties that we see."

MYHRVOLD: But to take the MP3 example, one could imagine a parallel future or parallel universe where, in fact, Apple didn't do the product that it had. In fact, there is something almost exactly like digital cameras that happens to be going along. And so it isn't something that's intrinsic, and you could ...

MULLAINATHAN: No, no. I agree with you it's not intrinsic in the following sense, is that there is an important innovation element here which is not about innovation of product features that matter. Your Apple case study is perfect. It's about innovation of product features that also fundamentally reshape the valuation of the rest of the space, and that's ultimately a creative activity. There's no way we can put a model on it. We might say digital cameras look like this. There might be a brilliant insight that someone then has out there that's a hat, market hat.

MYHRVOLD: The Steve Jobs of digital cameras may not have shown up yet.

MULLAINATHAN: Agreed. It may not be about this as a lifestyle device. Maybe there's something else about digital cameras that once you have that feature, people feel like who cares about megapixels. Who cares about all this? This is what I want a digital camera for. That would be very powerful, and in turn what is an unstable market right now is a stable market.

That's a Meta lesson, which says, "Where would I put my innovation money in some of these sectors that look like they're unstable?" I would put my innovation money into finding those dimensions that don't make me attractive relative to the choice sets that are out there right now.

BEZOS: Because that's going to be changed.

MULLAINATHAN: Yes, exactly.

PARKER: Remind me what the problem in an unstable market is.

MULLAINATHAN: Within an unstable ...

PARKER: What is the actual problem in unstable market? Why is it bad?

MULLAINATHAN: Yes, so wait a minute. Bad in a sense of...

BEZOS: Bad for society or bad for ...

PARKER: Yes, well, that's my question.

MULLAINATHAN: The sense in which we think it's bad ...

PARKER: Well, maybe I'm picking up on the implication that there's a value judgment there, and there's not.

MULLAINATHAN: There's less of a value judgment in the way I'm laying it out right now than you think there is, because it's hard For me to go further and make a value judgment, I have to make one more assumption, and here's the assumption that sometimes we are going to make and sometimes we might not. The assumption is that people, that instability leads to bad hedonic consequences, and that requires two things. First, that the outcome we get is not an outcome that's hedonically pleasurable; that is, they may not choose the product that's right for them. Or two that the actual act of choosing in an unstable situation is painful. Those are the two ways we could get it.

BEZOS: But if our first gate is that you can't judge the hedonic value you have post purchase, how does anybody ever know that they've made the wrong decision.

MULLAINATHAN: That's why I'm saying we're not necessarily ... that's exactly ... that's why I'm saying you're presuming more than I'm saying right now, because one of the challenges is what do we do there. Dick's suggestion at the beginning we can view as saying, are there a class of examples where we feel comfortable making hedonic judgments, whereas the consumer is confused and doesn't. Digital cameras? Could be very difficult.

BEZOS: That's the absolute truth.

MULLAINATHAN: Right. Digital cameras could be very difficult, because ... and , extremely difficult. I don't know how you should trade off weight versus quality. What the hell? How would I know that? Mortgages, I might say the consumer is going to the mortgage market with a very trivial proposition. I want to get the best mortgage I can to allow me to buy the biggest house so that I can live a comfortable life. I don't know how to translate that easy, simple preference to the characteristics I'm being given.

We may disagree that that's true. I'm putting it out that there might be examples where we, as outsiders, find it easier to do the translation. Those examples where we find it easier to do translation are the examples where we might be more willing to do interventions of the type that Dick proposed. Does that answer your question? And I'm not saying that those are everywhere. There are many markets where we'll see instability, and we'll say, "C'est la vie". But it's a useful thing to know, because it's a descriptive. Many times we're as interested in description as we are in intervention. In fact, description is perhaps even more interesting.

MYHRVOLD: Isn't fashion an example of something that's intrinsically unstable? People revise it forever.

MULLAINATHAN: Hemlines, for example, are unstable, right? Because there‚ it's a choice set effect.

MYHRVOLD: Sure, that's one example of that but more broadly. The whole point about the fashions two years from now is they ought to be different from today, because in fact, there's a value in that market for being different, so it's got to cycle.

THALER: Instability can be just another name for innovation, and we tend to think innovation is good. It needn't be pejorative. It's just descriptive. Here is one other footnote on the Apple example. Notice that that's a case where, because somebody has some monopoly power, they've been able to rationalize the market in a way that at least has some good qualities that Sendhil talked about, though we're probably missing out on innovation that might exist if there was another viable player, and the same with Microsoft. I'm not going to go back to Seattle.

HILLIS: I had one more thing to say about instability which is that it's not just maybe a characteristic, but it may be an inherently desirable characteristic, because if you're trying to solve an optimization problem in a complicated, multidimensional ...

HILLIS: But in fact, what you generally do, most optimization algorithms tend to find multiple minimums or maximum, depending on which way you do it. What you do is you deliberately introduce instability. You do things like annealing, which on purpose prevent the system from settling down.

MULLAINATHAN: I see where you're going with this. I do. Another way of saying this...

HILLIS: If you think of the societal problem as being we want to find the best solution for society, then you do want...

MYHRVOLD: In particular, evolution is an unstable.

HILLIS: Predator-prey systems and parasite systems are a classical example, and you can in fact demonstrate that fitness finds better solutions if you introduce parasites than it does if you don't.

MULLAINATHAN: Here I make one caveat which is let me give you two descriptions of instability. This is a great point. I want to give two descriptions of instability, one which matches your description and one which doesn't, which is because valuations change so much with the choice set, new entrants always coming in, reshaping the landscape and there's a lot of churning of things. An alternative version of instability is there are four cell phone carriers. There are always four cell phone carriers. It's that their pricing structures keep changing in the circle. That second case doesn't have nearly the competitive, anti-incumbent advantage that the first case does.

HILLIS: It does only happen if you keep the number of dimensions there, so if you're innovating by adding dimensions and things like that, you don't have that kind of simple situation.

MYHRVOLD: Right. In particular, the three broadcast networks had that in broadcast TV, but cable broke that, and then all of a sudden you've got Fox, and you've got this whole explosion of things, because a technology broke the bottleneck.

MULLAINATHAN: And the only thing I'm saying is that there's a two by two instability changing participant mix, and that the cell ...

HILLIS: That's a different instability than the one that you were talking about where somebody comes along and they're just some new dimensions that hasn't been a characteristic. There are two different kinds of instability.

MULLAINATHAN: That's a good point.

THALER: One footnote to this whole discussion, this project that we've been talking about during this session, that there's a hidden third author.

MULLAINATHAN: He's under the table.

THALER: Yes. In some ways, one way this project, to describe the way this project started, was asking the question, when do more choices make consumers better off? And the naive economic answer to that is, always. At least assuming away search costs. And if you take all the stuff that Sendhil showed you, and you say, all right, now we have this now very complex set of choices, and we realize that sometimes people find the task hard and don't pick the thing that's hedonically best for them, then when does this complexity lead to good outcomes. And, we don't have a one-line answer to that question, but in a sense, I'm only adding this to give you a context of how we came to this project and what question we're trying to get our hands on an answer to.

MUSK: Does everybody know the classic joke ... there's three guys on an island, a chemist, an engineer and an economist, and they come across some canned food. The engineer says, "Well, I can open that can for you by banging it on a rock," and the chemist says, "I can find a way to corrode that can open." The economist says, "Assume a can opener."

MYHRVOLD: Okay.

THALER: Assume poverty.

MULLAINATHAN: Can we switch now?

MYHRVOLD: Yes.

MULLAINATHAN: This will be a slightly different kind of exercise. What I'd like to do is, in this one, I would like you guys to co-create in a sense that this is very early stage, even earlier stage than the thing we just presented. What I'm going to do is I'm trying to make a soup, and I'm going to tell you all the ingredients we have, and you can tell me one of the ingredients doesn't fit or there should be other ingredients, but I have some strong views so I'll fight you down. That's a tension that I'm hoping you get.

Let me start by saying here's an interesting observation about how we tend to understand poverty. Our current theories and approaches to poverty are not about poverty at all. They're about the things that happen to co-vary with income.

What do I mean by that? It so happens the poor live in environments where education is particularly bad. It so happens that the poor live in environments where there are bad health systems. It so happens that the poor live in environments with crime, etc. It can go on. I'm not saying that's a bad way of studying poverty, especially from a policy point of view.

For example, if you're interested in dealing with HIV, one useful thing you may do is to deal with tuberculosis, because many HIV patients die of TB. But conceptually, you're not really studying HIV. You're studying one of the things that happen to co-vary with the existence of HIV. One of the starting points of this project is to ask the question, is there something intrinsic to poverty in and of itself? Okay?

ROMER: Can you give us a context? Are you thinking inner city U.S. or Haiti?

MULLAINATHAN: Everywhere.

ROMER: And you think it's the same? Inner city U.S. is the same problem as Haiti?

MULLAINATHAN: That's a question I want to pose, and I want to argue that we have an answer that is "yes".

MYHRVOLD: Wow.

MULLAINATHAN: That seems like an odd statement to make when you start with the previous approach, because inner city U.S. has crime. India doesn't. The health systems are okay in inner city U.S., they are bad in India. You would think from that approach that this is a stupid way to approach it. But what I want to argue is, in fact, there is something intrinsic to the state of poverty, and it's even broader than that, and to just preview it, I want to say that there's something intrinsic to the state of scarcity, whether it be scarcity of income or scarcity of other things. Okay? That's where I'm coming from.

Before getting there, let me show around a few interesting puzzles. I like these puzzles, because they help me think about the problems. So push back. These are things that we're doing a lot of work on, so if you have more questions, I can talk about them. The first is a project we have on fruit vendors and vegetable vendors and flower vendors. In development, you have to show photos of the people you work with. You also have to have a map of the place where you work, so I haven't worked out the second technology, but at least I have worked this out. The way fruit vendors work ...

MYHRVOLD: Good megapixels.

MULLAINATHAN: Yes. It's good, no? I don't know how many of you have ever seen a fruit vendor in a developing country, but the typical vendor has the following business model. They get up at three in the morning. They go to the local big aggregator market. The big aggregator market can be as far as an hour or 45 minutes away if it's a big city. They go there at three in the morning. Why?

Because all the fruits and vegetables and flowers are being shipped from the farms all around the urban center into that big market, and you've got to get it early while it's fresh. You buy that early in the morning, then you take it back, and you start selling from around five in the morning as customers come before they go to work to buy the fruits and vegetables, and your day is usually over by two. What's interesting is, think of the business models these guys have. This is a high working capital business.

Let's say a typical fruit vendor might use 1,000 rupees of capital, so they buy 1,000 rupees of fruit. They might sell that during the day, and then they accumulate, they convert their fruit back to financial capital, and the next day the whole thing starts over again. It's a continuous debt or savings demand business. It's classic economics. Why am I emphasizing this? This is from a survey of about 525 vendors that we have. We can see, as I was saying earlier, on a typical day they use about 1,000 rupees of capital. This used to be $20, now it's $25, and that leads to a profit of about ...

THALER: You mean we are even losing to the rupee?

MULLAINATHAN: Dick, we're losing to everything. Everybody. It's harder for me to do research in India now. I must be the only idiot who, I got a grant from this big Indian foundation. We agreed on it. We agreed on it in dollar denomination. I didn't even think.

THALER: All right, sorry.

MULLAINATHAN: Thanks for bringing up hard memories. The profit per day is about 10 percent. 100 rupees per day are the gross profits. Okay? I'm trying to get you into the business model.

Okay, let's now look at how they finance this working capital, and this is interesting. About 70 percent of them use what's called a "daily loan". Daily loan means when they show up at the market at three in the morning, there's a moneylender there, and he says, "Here are your 1,000 rupees." You take the 1,000 rupees, you buy the things, and the average amount of time they've been doing this is about nine and a half years. This is a pretty steady state financing model. And the average interest rate on this is about 4.9 percent or 5 percent.

BEZOS: Daily.

MULLAINATHAN: Daily.

BEZOS: Go compound that.

MULLAINATHAN: Exactly. That's the thing I want to think about. I want to think about, what an odd state of affairs. You're borrowing in steady state at 5 percent a day. Okay? Let me put this in perspective so you get a sense. Five percent a day means that on 1,000 rupees, they're paying 50 rupees of interest. Remember, the gross income is 110, so half that income is going to servicing debt.

Normally if we could find a development program that would double incomes, we would be thrilled. Well, here's a fun one that seems to double income potentially. How could they double their income? Five percent a day is such a powerful compounder that these women often buy a cup of tea during the day for three or four rupees. Okay and they might buy two cups of tea or one. If they simply had one less cup of tea a day for a month, they can go back to drinking to tea. Took the five rupees they used for that cup of tea and simply plowed it into their business' loan capital and did that every day, in 30 days, they'll be debt-free, and they'll double their income. How can it be? What are we to think of this population?

BEZOS: It's a very good question.

MULLAINATHAN: Right? And this is not a unique population. This is the most common thing that you'll see. I want to call this the debt trap—consistent borrowing at high rates.

HILLIS: Are there any other benefits from the moneylender?

MULLAINATHAN: Let's break it apart. One possibility is the money lender could be providing insurance, could be providing other features that they might value. In fact, we followed that through, and they're not getting anything at all. The moneylender will provide the following form of co-insurance. If you're not able to sell all your fruits, I'll roll over your debt for a day at five percent a day.

BEZOS: What is the return? What returns do the moneylenders get?

MULLAINATHAN: I'm so pleased that this question didn't come earlier. In an economics audience, when people see 5 percent a day, the first question they ask is, "Gosh, why no competition?" Which is a useful question, but it's a bit beside the point because the question is, why are people even borrowing? Forget competition. But in fact, the returns the moneylender's get are not that good.

BEZOS: Because they have a lot of defaults.

MULLAINATHAN: No, defaults are low, very low. The reason it's not that good is this is a big illusion in space. Five percent seems big, but when you have all this infrastructure where you set up and you've got to find these women and you've got handle it, in the end the transaction costs are huge for a loan that is $25, and you're making maybe a dollar a day on this loan in interest, and you've got to sit there.

You've got to have a boy. You've got to manage the cash. You've got to get the cash yourself. The most telling thing of this is, in doing this project, I took one of these incredibly aggressive micro finance institutions with me, the CEO. And I went with him, and we interviewed a bunch of these women, and it was fun and we had a good time and then I saw him a few months later, and he said, "Oh the interviews we did.

We've set up a branch in that market, and we're lending." I said, "Fabulous. You're competing with these guys." He said, "No, no. I would never lend to the women. That's too high transaction costs. I'm lending to the money lenders." That's great, great, and a smart idea. Because their cost of capital goes down, and then they'll filter down, but still it shows that there isn't that much money to be made, and it's because of the transaction costs.

This is the first puzzle I want to talk about which is steady state high interest rate borrowing. It's a stunning puzzle, and we ruled out this possibility that maybe there's something else that goes with this loan, and we see it in lots and lots of places. We see it in crop finance. Every year you borrow to finance your crop, but yet you borrow at high rates. Why don't you save your way out of borrowing?

That's one of the puzzles. We see it in the U.S. You see it with payday borrowers in the U.S. The typical payday borrower might pay 18 percent interest on a loan that lasts two weeks, and the average person is getting eight of these loans per year. If you go to the top quartile, they're rolling these things over again and again and again. How could it be? Why not save your way out?

MUSK: I feel like there's a delayed gratification problem.

MULLAINATHAN: Good. One hypothesis we can have, and we'll put this up in a second, is that you can imagine some sort of impulsivity problem. I want to put that on the table, because that's the prominent hypothesis that we often have, and the reason I went through the rich description of these vegetable vendors is that there's phenomena that are inconsistent with a simple delay of gratification problem.

BEZOS: For one thing, they're working their butts off.

MULLAINATHAN: Three in the morning they're getting up! If you told me I had to get up at three in the morning, I'd be dead now. Right?

PARKER: Is it a banking system issue that they literally can't cash the checks? That's certainly a problem in the U.S. where they go to payday loans because they're not citizens, and they can't open up a bank account.

MULLAINATHAN: Well, keep in mind I'm not putting up check cashers here. There are also people who pay for check cashing services, but that's nowhere nearly as expensive as payday loans. Check cashing is a slightly different thing, but payday loan really is about delay. It seems like we're moving money up. How could you be moving money up at such a high rate?

MYHRVOLD: Well, the personal discount rate is higher yet.

MULLAINATHAN: That's the delayed gratification issue. One explanation that jumps out at you is these people are highly impulsive, and I think the thing I'm trying to point out is, simultaneous with this odd behavior is also behavior that looks like incredible foresight, like getting up at three in the morning. Like, for example, all these women also save in what's known as a ROSCA (Rotating Savings and Credit Association).

A ROSCA is the most beautiful financial institution in the world, and I'll tell you why. There are very few things in economics that are universal, but in every country in the world, even in some primitive tribal societies, they have created the ROSCA independently. Unbelievable. A ROSCA is, the 10 of us are going to get together. We're each going to put money in each week, and one of us will take turns taking the whole pot.

Does that make sense? Each week you put in $100. There are 10 of us. Today I get the $1,000. Tomorrow you get the $1,000. Next day you get the $1,000. What's remarkable about that institution is that most of us are savers. The first guy gets immediate gratification. Everyone else is engaging in savings. All these women are in ROSCAs. There's something oddly inconsistent in their premises.

MYHRVOLD: When it's your day, you don't have enough capital to stop going to loans?

MULLAINATHAN: That's right. What we have to ask is, when they win the ROSCA, why don't they pay off the debt? One of the experiments that we're doing now ... well, it's done now ... is we simply took 1,000 of these women or 500 of them, and we paid off their debt. I haven't seen the data. It's a real question. Six months later, how many of them are debt free? How many of them have fallen back into debt? Introspect about that question. You can place your bets now, and we'll check.

ROMER: Do you know what they do when they do get the money in the ROSCA? Separate from your experiment.

MULLAINATHAN: Yes, they don't pay off the debt.

ROMER: But what do they use it for?

MULLAINATHAN: They buy durables. A washing machine is too high for this population, so they might buy a ...

MYHRVOLD: Pot.

MULLAINATHAN: A pot. Exactly. Okay, so this is one problem I want to present. I'm going to circle around the issue and tell these are the issues I'm thinking about.

THALER: Yes, we ought to have a poll on what percentage of the ...

MULLAINATHAN: Women fall back?

THALER: Yes, so can you write, when you make your vote as to the talk you want tomorrow, you can also write down what percentage of the people will be back in debt, and we'll report next year.

KAHNEMAN: So, you will tell this group why they're exactly like those women. Right?

MULLAINATHAN: Yes, I'm getting to that. There's a second interesting problem. I call those debt traps. The second interesting problem I want to talk about is failures to plan. Here's an example. A lot of women die every year. It's one of the biggest sources of mortality amongst non-elderly women. if you're not a baby and you're not old, one of the biggest sources of death, pretty close to HIV, is maternal mortality.

Women die in giving childbirth. Why do they die? It may come as a surprise, but delivering at home in the dirty environment without access to medical care so when you hemorrhage and there's no one to deal with it, and you can get infected. Often times they lose the child, and often times you lose the mother. Why deliver at home? One possibility, and in our survey we find this, is lots of women want to deliver at a hospital, but it's expensive, and more importantly than expensive, it's expensive to finance it.

MYHRVOLD: It's cheap compared to flowers.

MULLAINATHAN: Pretty cheap compared to flowers. Very interesting. That's right. I agree, and I'm not sure why this heterogeneity is there on this side. One of the reasons there's this heterogeneity is, I can tell you an explanation.

THALER: The flower vendors should pretend to be pregnant. It's an arbitrage here.

MUSK: It's a larger amount, so the transaction cost as a percentage of the amount is smaller?

MULLAINATHAN: Totally right. There are two reasons. One is it's a larger amount, so oftentimes we look at interest rates, but you should look at service fees. Sizes go up and interest rates drop fast. The second reason why it is, is these women are women who also borrow for crops and for all of these things. In terms of for the moneylender, the exposure is he has like a big loan portfolio. For a vegetable vendor, those women are much poorer than this sample. His exposure to you is the 1,000 rupees. That's his whole transaction fee, so you can't amortize the transaction costs of the fee over a big portfolio.

Okay. Putting that aside, we think that some of those who do not deliver are put off by financing. If you ask them, they even say, "Yes, I would love to go, but when I delivered, I didn't want to borrow at that rate, etc." That's fine. Seems somewhat sensible. There's an oddity here. The oddity is, you've known about the pregnancy for six months. Why didn't you save your way up? Odd. It's related to the issue of the debt trap, but here there's a very clear date that's coming, and you're not making the moves to save up to it. Okay? Does that make sense to everybody?

MYHRVOLD: You know who are not? There's not a set of women that are avoiding this?

MULLAINATHAN: There are definitely women who save their way up, but in our sample about 80 percent of them don't do it.

MYHRVOLD: How much are all of these things selection effects? The fruit vendor women may be the remnants, and all those who got it got the hell out of fruit vending and they went into something more lucrative.

MULLAINATHAN: Definitely, though we see in our sample about 40 percent of them are internal financiers. Remember I said 1525? 31 percent of them are also steady state internal financiers.

BEZOS: Internal financier means they finance themselves?

MULLAINATHAN: They do it with their own capital.

HILLIS: Do they have a reliable mechanism available for saving?

MULLAINATHAN: That's a great question. In this population, they do have something called a self-help group, which seems to be a pretty reliable mechanism for saving. Another mechanism they have for saving, ironically, is the moneylender who will charge them about 4 percent or 5 percent to save.

MUSK: They charge to save.

MULLAINATHAN: Most poor people pay to save. The self-help group doesn't charge.

ROMER: That's per year?

MULLAINATHAN: Yes, it's per year. It's a lower rate. You're better off saving. Okay, what I want to point out here is why do some foreseeable events appear as surprises? And we see this again and again. These two things ...

MYHRVOLD: With a high enough discount rate, you answer both questions.

MULLAINATHAN: Exactly. One possibility is we call this "impulsivity of the poor".

MYHRVOLD: "Time premium" is less judgmental than impulsivity.

BEZOS: In the pregnancy example, is it practical for them to save in that six-month period, to do a hospital birth? What percentage of their income would they have to save?

MULLAINATHAN: A good way of thinking about it is the women who end up borrowing at such high rates, they often pay it off in about six months. The funny thing about debt versus savings, is savings is exactly like debt, but earlier and cheaper.

BEZOS: I got it.

HILLIS: The other thing that would make the discount rate be different without impulsivity is if you have a high accurate judgment of the uncertainty of the future. If you have a highly uncertain future, then in some sense, the option value of spend it now is rationally worth more.

MULLAINATHAN: That theory, unlike the high subjective discount rate, that theory we rule out for the following reason. Economists have postulated that theory, and they talk about this theory as spending theory preferences, and the idea is, if you don't if you're around tomorrow, your discount rate is delta times your probability of being around. But the thing is we didn't calibrate that, and while this population does have higher mortality than we might, it's nowhere near the order of magnitude.

HILLIS: It's not just being around. It's also you don't know what you're going to need tomorrow. You don't know how much they're going to ... you don't know what your capital ...

MYHRVOLD: Shit happens.

MULLAINATHAN: But that should push you the other direction. If you might have very high needs for capital tomorrow or you might not, that should push you to save, not to spend it now. Not being around is a good motive to spend it all. Okay, so those are the two questions. There are a few other examples like this. I'll show you this from the U.S. This is relative caloric intake over the month. Okay? From week one, week two, week three, week four. Week one, at the beginning of it, you receive your food stamps, and you'll notice, how hard could it be to smooth over these four days, but you significantly lower caloric consumption to about 10 percent calories less of the diet. Only about 300 calories on average in week four than in week one, and this is a consistent thing you see. I'm showing this ...

THALER: In week four, they're starving?

MULLAINATHAN: No, no, no. This is relative. In other words, this is ... I've set zero to be here. This is 10 percent higher relative to that.

THALER: Okay, got it.

BEZOS: Why is week three higher than two?

MULLAINATHAN: This is basically ... the big effect is of one to four and then there's a drop.

ROMER: How carefully can they observe intake as opposed to purchasing?

MULLAINATHAN: These are pretty good. These data are pretty good because they ask you, "What did you eat?" They go through food diaries. This comes from a data set on food consumption, not on food purchases. They go through these detailed diary questions. "What did you eat there?"

HILLIS: What month? Is it the calendar month?

MULLAINATHAN: People receive food stamps every month. I calibrated the beginning of week one to be the day they receive food stamps. It's not the calendar month.

HILLIS: It's a food stamp month.

MULLAINATHAN: It's a food stamp month, and it varies between people. Different people are on different schedules

ROMER: You're saying this is only a 10 percent difference from week one to week four.

MULLAINATHAN: That's on average.

PARKER: It doesn't seem that meaningful.

ROMER: That could easily confound with purchases. Unless your diary data is ...

MYHRVOLD: Ice cream.

PARKER: The 10 percent doesn't seem like that much more. I probably go through more than 10 percent variance from week to week and day to day.

MULLAINATHAN: Fair enough.

BEZOS: This is across a whole bunch of people.

MULLAINATHAN: This is across everybody. It's not saying 10 percent variability. It's the systematic variability. This doesn't have to be  super consequential. You could argue, so you got your food stamps, you celebrate by having ice cream. That would be the innocuous solution. When combined with the qualitative data, this is far less innocuous. You hear lots of women saying in week four ... could have graphed something else. Maybe I should graph this next time, which is have you gone hungry in the last week because you did not have enough food to eat? That also goes down. I should probably graph that.

MYHRVOLD: But it's again consistent with a high discount rate.

MULLAINATHAN: Yes, exactly. All these things start to look like high discount. That's why I'm leaning on this. Let me tell you what we can do to understand all of this, which is that scarcity evokes a unique psychology in and of itself, and that scarcity is not just about scarcity of money, but it can be scarcity in time. It can be scarcity in pleasures. I want to go back and re-conceptualize everything we've seen with those of us around the table scarce in time. Just like the poor may make bad allocation decisions with money, I want to postulate that we make some pretty bad allocations with time. Let me give you an example.

BEZOS: Why am I here?

MULLAINATHAN: Exactly. That's exactly it. Every time, I'm about to go give a talk, what do I ask myself? What the hell did I agree to come see a bunch of weird tech guys.

BEZOS: It was so far in the future, you said yes.

MULLAINATHAN: Let me give you an example. We talk about the women being in debt traps. We are in time debt traps. We feel like, gosh, if I could only clear my calendar of all of these commitments, then I would be free to start thinking much more easily. But somehow, we always stay under the gun‚ we're always fire fighting. We're always working on the thing that's delayed, etc. What I want to point out is there something, and this is the hypothesis we have going in, there's something unifying, which is under conditions of scarcity we tend to make similar mistakes.

The poor do a bad job of allocating their money under these conditions of scarcity. The poor do a bad job of allocating time under these conditions of scarcity. The poor in hedonic pleasures those who are going through a very bad time, maybe a divorce do a bad job of allocating the little hedonic pleasures that they have and seeking out things. That's the analogy I want to carry through. Does that make sense to everybody? And that's the view I want to argue is the unifying principle.

PARKER: Is it possible, though, that we do a better job of allocating our time under enormous time pressure and scarcity that makes us focus on optimization more.

MULLAINATHAN: That's a good question. What I want to argue is that there is a duality here which is ... let me go through ... if I forget, please do raise it again. I want to show you one metaphor that is helpful to me and hopefully will be helpful to you guys.

MYHRVOLD: Are you going to address the question as to whether you're doing poverty or a covariant of poverty?

MULLAINATHAN: Here I'm doing scarcity, which is an intrinsic feature of poverty.

MYHRVOLD: Sure, but there's a causal question.

MULLAINATHAN: Say more.

MYHRVOLD: If you were a medical doctor, you'd say we have a discount rate disorder here. We're not properly ... we're not making a rational choice about discount rate, because once things get too scarce, we freak out about it. And yes, that could cause those women to double their incomes, but that may not get them out of poverty.

MULLAINATHAN: Right, so what I want to argue is that any condition of poverty evokes this psychology of scarcity, and so that's the sense in which I feel I'm talking about something intrinsic to poverty.

MYHRVOLD: Fine, but that isn't necessarily causing the poverty.

MULLAINATHAN: No. Exactly. I'm not going to argue that the psychology of scarcity necessarily generated poverty. It could be that exogenous features, in fact, one of the experiments that I want to tell about, we know what's causing poverty, it has nothing with the scarcity, but we're going to see scarcity responding to that.

MUSK: What do you think causes poverty?

MULLAINATHAN: I have two answers that I'm pretty confident about right now. This is less an exercise in saying X causes poverty than it is an exercise in saying, let me tell you what the background ...

MYHRVOLD: It's a feature. It's an intrinsic feature of poverty because scarcity is part of poverty.

MULLAINATHAN: That's right.

MUSK: But you said you know what causes poverty.

MULLAINATHAN: I was joking when I said that. If I knew that, would I be talking to you guys?

MUSK: Well, that you can track lack of education as the primary cause of poverty.

MULLAINATHAN: I want you to take one ride with me, which is that there are two enterprises which are different and both important. One enterprise is like the HIV/tuberculosis thing. My only goal is raising income of the poor, education you may say education, they say clean drinking water, someone said deworming.

MUSK: Education.

MULLAINATHAN: That's your view, but I'm saying that other people would say other things.

MUSK: I'm right.

MULLAINATHAN: And there's a view, and people will throw money at it and solve the problem. There's a very different enterprise that we can be engaged in. Let's put aside poverty alleviation for a second, and let's ask, "Is there something intrinsic to poverty that has value and that is worth studying in and of itself?" One of the reasons that is the case is that, purely aside from magic bullets, we need to understand are there unifying principles under conditions of scarcity that can help us understand behavior and to craft intervention. If we feel that conditions of scarcity evoke certain psychology, then that, not to mention pure scientific interest, will affect a vast majority of interventions. It's an important and old question.

Honestly, if you're interested in anti-poverty mediation tomorrow, I don't think anything I'm going to say is going to be that helpful, in that sense. There are probably people out there that could say more that would be helpful tomorrow. They could say, we know this is the best way of providing education that raises education levels. We know this is de-worming that does this. What I'm saying is that if we're interested in the deeper question of scarcity, and poverty being a manifestation of it, we conjecture that there's some unifying principle. Does that make sense? I do want to draw the sharp line. I don't want to pretend that I can answer your question, because I can't.

MUSK: It's very obvious that education is the primary contributor of poverty.

MULLAINATHAN: Then we have it back on the table.

KAHNEMAN: You need to get into time poverty.

MULLAINATHAN: I agree.

KAHNEMAN: Because there are many things to consider when you start looking at what time poverty is doing to us. Let's talk about professors. There might be people here who are more efficient than we are.

THALER: Might?

MYHRVOLD: The solution of the time poverty issue is, for me anyway, is that if I continue ramping things up more and more, eventually I'm forced to only do high-leverage things. I continually add more and more things which forces me to do everything scaleably. Elon's answer is probably pretty similar.

MUSK: Yes, I agree.

MULLAINATHAN: One view is that you guys think that you're optimally allocating your time, but we're a whole set of highly intelligent people.

PARKER: It's never optimal.

MULLAINATHAN: Here's what I want to say is that's intrinsic to scarcity. Let's put aside money for a little bit, and we're going to talk about a suitcase. I want you to take this metaphor with me. There are two people who are on vacation. Ted is traveling with a small and very full suitcase. James is traveling with a larger suitcase that's got some space in it.

They both visit a store that is offering running shoes on sale. What does James ask, the guy with the larger suitcase? He asks, in thinking about these shoes, "Do I like these shoes?" That's an absolute judgment. What does Ted ask? Ted asks himself, "Do I like these shoes, and what of roughly equal size must I take out of the suitcase to make room for these shoes, and is that thing worth losing if I buy these shoes?" Does that make sense to everybody?

In other words, James has a very different decision problem than Ted. James' decision problem is very narrowly graphed. He can simply ask the question, "Do I like this thing?" Ted is forced to make a trade-off between this thing and other things and re-optimize his entire portfolio. Does that make sense to everybody?

PARKER: It's not quite right, because if they're saying, "Do I like these shoes," then it seems to imply they've already made the decision that they need new shoes. Maybe the question is phrased ...

MULLAINATHAN: Say more? That's still a feature of the shoes. In other words, ...

KAHNEMAN: You can make them equal, whatever you say about the shoes.

MYHRVOLD: Well, of course, it also could be, "Can I afford?" but you didn't want to use that one. You wanted to pick a nonmonetary one, so you used space.

MULLAINATHAN: Exactly. The suitcase is just a metaphor. Many things are a packing problem. The thing I'm trying to point out is that scarcity induces a very different packing problem. When I talked about James having a bigger suitcase, he had some room. What does that mean? Room is slack. When you have a bigger suitcase, you can afford to leave it somewhat unpacked.

Why can you afford to leave it unpacked? It's because of diminishing marginal utility. A guy with a big suitcase, the last 10 percent that he's packing doesn't have much value. One thing he can buy, is he can say, "I've got all my stuff. I won't tightly pack it, and if I choose to not tightly pack it, what's my advantage? My advantage is, my decision problem gets much, much easier."

MYHRVOLD: Optionality.

MULLAINATHAN: Optionality. And it becomes easier in a very specific way, which is, if I created slack for myself, I no longer need to make trade-offs. I can evaluate things in isolation, whereas if I don't have slack, I'm forced to evaluate things globally, and I'm forced to make lots and lots of decisions. Does that make sense to everybody?

Here's an interesting example of this that is compelling. This is work by Crystal Hall. Imagine that a friend goes to buy an appliance priced at $100. Some people are shown $500. Some people are shown $1,000. Although the store's prices are good, the clerk informs your friend that a store 45 minutes away offers the same item on sale for $50 less. Would you advise your friend to travel to the other store to save $50 on the $100 expense? That's the question that's asked.

You can see the way this is set up. Some people are given saving 50 on 100. Some people are given saving 50 on 1,000. As you may imagine, when we run these with undergraduates or people such as yourself, it's a funny thing that emerges. People are more likely to travel 45 minutes to save $50 on $100 than they are to travel to save $50 on 1,000. Okay?

That's because they're evaluating $50 locally, and fifty dollars, percentage of 100: 50 percent. Fifty dollars, percentage of 1,000: much smaller. This is what we see in the first row are the high-income population. When we give it to people such as you, they exhibit this preference in consistency, which is 54 percent are willing to do it for 100, but only 17 for 1,000.

But look at the low-income population given the same question. They don't care. They don't care whether it's 50 on 1,000; 50 on 100; 50 on 500. They're willing to travel in whatever case.

KAHNEMAN: They need the 50.

MULLAINATHAN: They need the 50.

BEZOS: And they're doing global optimization.

MULLAINATHAN: And they're doing global trade-off. Fifty has intrinsic, specific value to them. The intuition I have behind this is that sometimes we go to a restaurant and we say, "Oh, I don't want to order that. That looks too expensive." We also have some local budget constraints, some sense of what's a right price or whatever, but when I think about that, I'm not poor. I'm making a local judgment.

BEZOS: Well, it's an aesthetic decision mostly.

MULLAINATHAN: Exactly. And a good way to understand that is when a poor person does that, what they're thinking is, if I bought this, I would have to give up that. Does that make sense? That's the essence of slack, and that's what this data is trying to tell us.

The suitcase metaphor tells us the first ingredient I want to bring to the table, which is about lack of slack under conditions of scarcity. I want to know one consequence of that, which is greater decision complexity that comes from lack of slack means greater need for decision quality. The guy with the smaller suitcase, if he makes the same mistake as the guy with the bigger suitcase, he gets hurt more often. Does that make sense? The slack acts as a buffer for our mistakes.

HILLIS: Do you know what computer scientists mean when they talk about caching? Cache management. This exact thing is a very studied problem in computer science. It's a problem when you have a fast memory and a slow memory, and what you want to do is you want to keep things in the fast memory that you're working on.

BEZOS: Small: fast memory. Big: slow memory.

HILLIS: A small, expensive, fast memory and a slow memory.

MYHRVOLD: In that case you call your suitcase, what fits in there is the working set.

MULLAINATHAN: Exactly.

MYHRVOLD: And what you have to do to manage the working set under some circumstances in economics. It's really an analogy.

HILLIS: If you have slack, then it's not a very hard problem. The big problem is what do you empty out of your small memory in order to fill in the big memory.

MULLAINATHAN: Very nice. I like it.

MYHRVOLD: That's why your computer occasionally, when you have too many applications open, it stops working. That's exactly what's going on with it, is you get to some point where it doesn't know what set of things‚ where the problem of deciding what set of things to keep in mind becomes an overwhelming management problem. Whereas if you close a few applications, and all of a sudden you have a little bit of slack, it suddenly recovers.

MULLAINATHAN: That's good. I'll look that up. That's a very similar metaphor in that you're saying, if I get this right, we could say those who are scarce have a very small cache, and through swapping it out…

MYHRVOLD: Yes. Therefore, they're in that state that their computer is caching.

MULLAINATHAN: That's nice. Stopping? Okay. Should I take a break now? I'm happy to stop.