# Market Design: Engineering Efficient Allocation Systems

**Podcast:** web3 with a16z crypto
**Published:** 2026-07-14

## Transcript

One of the things about market design is you think in advance about what you're going to do.
A lot of economics asks, how does the world work?
And market design gives the opportunity to ask, how should the world work?
Many systems, from kidney exchange to school choice, would fall apart without good market design.
There's 130,000 people who experience, for the first time, kidney failure each year, but we only do about under 30,000 transplants a year.
Most people who need a transplant die without getting one.
The challenge is to build systems that produce good outcomes even when participants have competing incentives.
A lot of kidney exchanges now are done non-simultaneously in non-directed donor chains.
People who wanted to give a kidney to someone and didn't have anyone particular in mind.
And so they allow us to organize a chain that doesn't have to loop back to the beginning.
And in that chain, every patient donor pair can get a kidney before they give one.
Hi, everyone, and welcome to First Principles, a series by the team at A16Z Crypto.
I'm Tim Roughgarden.
I'm the head of research at A16Z Crypto.
And today we're talking about market design, the careful engineering of efficient marketplaces.
Many systems from kidney exchange to school choice would fall apart without good market design.
The challenge is to build systems that produce good outcomes even when participants have competing incentives.
a problem that's central to markets, including those familiar to us in crypto.
Our guest is Dr.
Alvin Roth, the Nobel Laureate in Economics and Stanford professor who is one of the pioneers of market design.
And I'm also joined by Scott Commoners, a professor at Harvard Business School and a research partner here at A16Z Crypto, and as it happens, a former student of ALS.
Together, we explore how economic theory becomes practical market design.
from New York City high school admissions to kidney exchange, and how technology changes marketplaces, as well as the idea of repugnant transactions, which is a theme in Al's soon-to-be-published book, Moral Economics.
Here's our conversation.
So, Scott, we get to interview Al Roth, which is very exciting.
I imagine this is sort of particularly meaningful for you, given that, you know, you were a PhD advisor of his back in the Harvard days.
Oh, yeah.
And in fact, I mean, I first met him as an undergraduate the summer before I took his market design class.
You know, I was reviewing papers for this Zentralblot Math.
It's one of these, you know, sort of like large-scale math reviewing programs where they're trying to summarize every single math paper ever written.
It's an incredible Herculean project.
And I was at the time one of the people who was, you know, writing summaries for this, one of the like...
thousands and thousands.
I was reviewing number theory papers, and some of those were like, you know, sort of very run-of-the-mill.
I was getting like the tier one game theory papers sent, and I got this article.
I think it was Deferred Acceptance Algorithms, History, Theory, Practice, and Open Questions.
It was a survey article that had come out earlier that year and was about the use of this thing called the Deferred Acceptance Algorithm, which is an algorithm for finding stable matchings, matchings where nobody has any incentive to try and recontract after the fact.
And the article took you through this beautiful mathematical theory.
and then into practice, how this was used in the assignment of doctors to their medical residencies, how it was starting to be used in school choice, and tangentially sort of like how it also related to organ allocation and saving lives through kidney exchange, you know, all this incredible stuff.
And I was like, oh my God, this is the coolest thing ever.
I have to learn more about it.
And then I realized that the author...
taught at Harvard.
It was Al Roth.
He's in Stanford now, but at the time he was at Harvard.
And I had never heard of him.
I didn't know anything about this.
But I checked the courses of instruction and lo and behold, he was teaching a graduate market design seminar that fall.
So I walked in and I tell you the first like 45 minutes of that class, I knew this was what I was going to do.
And Al crystallized this set of ideas with like one in retrospect was like the simplest, but to me has always been the most profound thing I think I've ever learned about economics, which is that economics is a way of thinking about the world.
Economics is a descriptive science and a toolkit for reasoning about incentives and markets and behavior.
And market design is a way of thinking about economics.
A lot of economics asks, how does the world work?
And market design gives the opportunity to ask, how should the world work?
What is the way in which we can use economic theory and analysis to build practical solutions to real-world problems?
That idea was just so...
Like, I was primed to hear it, and it hit so hard.
I was like, wow, this is what economics can do.
And then I was a market designer.
I mean, even that phrase, right?
Market design.
I mean, another variant of that is, Alroth has a famous paper called The Economist as Engineer, right?
And so, and really, If you look at all of the collaboration between computer scientists and economists now, market design, I think, has sort of in many ways led the way.
It's really combining sort of the practical, you know, building and problem-solving skills that engineers often pride themselves in with the sort of deep insights of incentives that economics gives you.
And even that, so that phrase market, I mean, is even that phrase sort of attributable to Al Roth?
Yeah, I mean, I think Al has really been heavily involved in essentially like the branding and the codifying of like what it is to be a market designer.
And in the paper you mentioned, The Economist is Engineer, this was really like a seminal work.
And it took as case studies different instances in which people had used economics in combination with other types of what we'd call like engineering interventions, whether, you know, whether it's specific design features that make...
an abstract auction actually work in a specific context or The theory doesn't actually quite tell you, you know, what's going to happen.
You know, stable matching theory started as a, you know, Gail and Shapley with this paper on stable marriages and stable college admissions assignments.
So when you talked about the deferred acceptance algorithm earlier, is that the same as the Gail-Shapley algorithm or?
Yeah, Gail and Shapley introduced this deferred acceptance algorithm, which is a way of finding a stable assignment of partners.
And they first talk about, these are both abstractions.
They talk about a marriage market and then a college admissions market.
But what you should really be thinking about is, you know, the matching of doctors to their medical residencies.
or perhaps students to schools, stable matching theory doesn't give a way in general to ensure that you can do stable matching when some people enter the market as couples trying to match to the same location.
It's an example of what's called matching with complementarities, and it's not guaranteed to work.
As Al showed in work with Elliot Perenson when they redesigned the National Resident Matching Program along with the American Medical Associations, in practice, for the context of the medical residency match, it turns out that matching with couples was always possible.
And they had to, like, you know, design an algorithm.
They had to sort of, like, adapt deferred acceptance, which assumes that there are no couples, or at least in sort of the native form assumes that there are no couples, to incorporate couples.
Like, how do you...
change the way the system works in a way that respects the theory and respects the objective of stable matching, right?
You still want to have this outcome where no one has any incentive to recontract after the fact.
How do you do that in a way that also incorporates this other very important real-world component?
And that way of thinking about the role of economic theory, you know, sort of Al was really a pioneer in translating and sort of building a toolkit for how to do that then other people built around.
Amazing.
Well, even just talking about this from afar is incredibly inspiring.
I can't wait to actually talk to the man himself.
And also, I mean, he's just, it'll be fun.
Yeah.
Yeah.
All right.
Let's do it.
Great.
So, Al, welcome.
I thought maybe we could chat about the role of...
sort of translating theory into practice, right?
Because, you know, over your career, at least my understanding is, you know, early on, it was really more focused on sort of the foundations on getting the mathematics right.
But obviously, you know, at this stage, you're rightfully celebrated for your practical market design work.
And so I'd just love if you could take us through a little bit about your evolution from more of the primary focus being working on papers to the focus really being on helping real people.
Well, so the joke about theory and practice is correct.
Its punchline is right on target.
You're in theory.
Theory and practice are the same, but in practice, they're not.
And it turns out there are complexities in the world that we leave out of our simple models, but that are important in the world.
So I spent a lot of time studying the labor market for doctors.
You know, it was an interesting...
problem for me to study.
And I studied it.
It had some mathematical background, important mathematical background, and it had some complexities.
And I wrote a paper about it in which sort of the final line was something like, they have a hard problem here.
And when you're writing a theory paper, that's a fine conclusion.
You know, I showed that couples were difficult to handle in a labor market.
You know, two career households are tough.
And then years later, they...
called me up.
My phone rang at the University of Pittsburgh and they said, would you lead a redesign of the market?
I still remember that phone call with a visceral feeling of, I was sort of sorry I'd picked up the phone.
But I knew why they called me, which is because I'd written a book on matching.
But I also knew that the only parts of the book that directly applied to the problem that I was about to...
take were the counter examples.
Because the book is full of simple theorems, and then you say, but of course if they were complementarities, these theorems, you know, you couldn't rely on them.
And indeed you couldn't.
But the theorems proved to be a surprisingly good guide for what was out there.
So that was really the beginning of my career as a practical market designer, when I agreed that their hard problems would become my problem.
You say the theorems were a guide.
These negative results, these counterexamples, and for people in the audience who don't know, there's this concept of substitutability versus complementarity, and couples are the simplest case of complementarity.
It turns out that under complementarity, stable matchings often don't exist.
If you were trying to do a centralized matching mechanism with couples, hard.
How do theorems guide us even when they tell us that what you're trying to do can't be done in general?
Well, so the phrase in general is key there, because what you said was with complementarities, stable matchings often don't exist.
And we actually didn't know that.
And it's actually not true in the application.
It's that they can not exist, but that happens only seldom.
So in my book, all the theorems had the grammatical form.
The following thing always happens.
Stable matchings always exist when you have a simple matching market.
Another grammatical form of theorem was some things are impossible.
You can never get a stable matching mechanism that always makes it a good idea to state your true preferences.
But we didn't have any theorems that said how often those problems are likely to crop up.
And it turns out that those in a market like the market for new American doctors, those problems crop up very seldom.
And therefore, we could...
start by treating the problem as if couples didn't matter too much, and then we could fix the couples, because of course the couples matter a lot to the couples.
But that's sort of what we could do.
We could say, let's ignore them until we see the problem appear, and then we'll fix the problem if we can.
And so far, we always can, because it's rare that stable matchings don't exist in a market with...
45,000 applicants and 35,000 jobs and 2,000 couples.
You know, the labor market for doctors is an interesting one in that, you know, the use of stable matching there goes way, way, way back, right?
Sort of farther than the start of your career, Al.
I mean, even sort of, I say this fully knowing that I'm on a call with two people that know literally 100 times more about matching theory than I do.
But my understanding is that even the algorithm predates the Gail Shapley paper as far as the implementation.
So since that work, we're seeing these matching techniques used in the field more and more and more and more applications.
So I'm curious about that, how that story went.
Was it that once people saw the practical matching work you were doing in the doctor labor market, then other opportunities arose?
And then you saw more places where these ideas could be used?
Or how did that happen?
Well, a couple of things happened.
And one of them was that those ideas were already in use without knowing about Galen Chapley.
in a bunch of medical labor markets.
And so fixing it in the big one just diffused it to the others.
And there's a...
A point that computer scientists will appreciate it, which is that the deferred acceptance algorithm hadn't exactly been rediscovered many times, but mechanisms equivalent to it had.
So things that if you put in the same inputs would give you the same outputs as the deferred acceptance algorithm had been discovered.
And that was the case with the medical match from the 1950s, when in the 1950s they had a simple problem.
There weren't any two career households because all American medical graduates in 1950, almost all, were men.
So they weren't married to other doctors.
But by 1970, there were already, you know, 10% women.
And today it's a little over 50%.
So there are lots of married couples and they have to look for two jobs, not just one.
And that's the complementarity that Scott is talking about.
When you're a couple, you care not just about what job you got, but what job your spouse got because of the iron law of marriage, which is that you can't be happier than your spouse.
Good practical advice for the audience.
I remember this iron law.
This is perhaps one of my most quotable Al Roth facts, the iron law of marriage matching.
I want to go back in history one more step, because we've talked about the medical match and it's fixed.
But actually, the story of how it came to be is also instructive.
Can you tell us just quickly, talk us through what problem it was solving and how it evolved to the form that you were working on?
So the medical match was unraveled.
Before they had the match, they'd have a long period where doctors were getting hired for their first job after graduating from medical school, after four years of medical school.
They were getting hired in the summer after their second year.
And that caused lots of problems.
That's too early in some sense.
There was lots of information missing on both sides.
And then later they had some other problems as they tried to fix it, some problems of exploding offers and congestion.
But those problems turn out to be quite common.
today even in many markets, like markets for lawyers, like markets in private equity.
You guys might know about some of those markets.
I don't know how.
Certainly on campus, the private equity interview dates keep moving earlier and earlier.
Well, and private equity interviews people who have just taken jobs with investment banks for two years later, which also...
you know, has conflict of interest problems.
So having markets unravel is a tough problem, and not all markets can solve it with a clearinghouse.
There are markets like the law school markets and the markets for appellate court clerks that wrestle with those problems and just suffer from their consequences.
Actually, I'd love to ask a follow-up on that.
I think this may be one of the themes of the conversation, you know, when our market failure is, you know, amenable to being fixed and not.
Would you say there's something fundamental about, say, the market for lawyers such that, you know, it's just not practical to actually have a clearinghouse?
Or is it just a coordination problem of having all the relevant parties agree to sort of switch over to a clearinghouse?
That's a good question.
Market failure means lots of different things to economists.
So I've, and I use those terms in my early papers, but now I like to talk about marketplace failures.
You know, I think of marketplaces as small parts of big economic environments, small parts of markets.
So these are marketplace failures.
The marketplace is failing to serve the market in an efficient way, things like that.
But one of the problems for lawyers is that they are able to promiscuously make binding verbal contracts.
So just by talking to you, I might be entrapped into working for you under some circumstances.
For instance, if you're a federal appellate judge and I'm a law student and you say, you know.
You'd be a good clerk.
I'm going to hire you.
And the only grammatically correct answer in that circumstance is yes, sir.
Thank you, sir.
And then we're done.
You know, the job market is over for me.
So that makes it hard to have a centralized clearinghouse because you could say that to me before the centralized clearinghouse, too.
I think that's the main problem for lawyers.
And that's a very interesting one because often economists talk about the inefficiencies that result from the inability to make contracts.
Right.
If we can't make contracts, then I can't send you.
you know, 5,000 bushels of wheat with 60 days payable, you know, on delivery.
But if we can promiscuously make contracts, then you can hire me before I have a chance to interview at the other places that are interested in hiring me.
This happens in crypto markets, incidentally, right?
You can write smart contracts, which are these, you know, software programs that have immutable function.
You can basically write a smart contract to commit to an outcome up front before you participate in a marketplace.
And sometimes, you know, the incentives are such that people would...
in equilibrium, choose to do so, even if it means they're giving up rights to things they might later want from the marketplace protocol.
So, Al, so moving on to other application domains, I wonder if we could talk about school choice a little bit and how you got involved there and the work taken to, you know, convince the relevant decision makers that your ideas were worth putting into practice.
Okay.
Let me ask, do you have by any chance kids in the New York City school system?
I do.
My son is in kindergarten, yeah, at the local neighborhood public school.
Well, in that case, maybe let me emphasize that, you know, Parag and Attila and I didn't design the kindergarten entry process.
We just designed the high school entry process.
I've made a mental note to be in touch in about eight years.
It sounds great.
Eight years.
That's precocious, but okay.
Oh, that's right.
You get at eighth grade, you go into ninth grade and go through the process, which has been modified.
It was modified under the...
predecessor to Adams.
To Blasio?
Yes.
So he sort of vandalized the algorithm a little bit.
But what happened there is Mayor Bloomberg, in his first term, decided to claw back centralized authority for the school system, which had been widely diffused into what was called community control.
So there were lots of little school boards that were doing everything for the schools.
And Bloomberg decided that New York City public schools should do that, you know, the Department of Education.
So that was the second time where my phone rang and there was a guy on the other end who said, we have this problem.
30,000 of our students have to be administratively assigned.
We've tried to go through a choice process.
They have to be administratively assigned to a school that they didn't list any preferences for because we just run out of time to match them.
But he knew about...
the resident match.
And he said, this seems to me, you know, he was doing some of the market design.
And he said, this seems to me like, like a problem you've already solved.
Can you help us?
And at that time, Parag Patak was a graduate student looking for a problem.
And he came into my office one day and he said, you know, I went to Toulouse and I, I, you know, studied a lot of theory and I sort of thought maybe I'd like to get into something more applied.
And I, I said to him, how about New York City Public Schools?
And, you know, he's.
run with that.
So much of what I tell you about what's now known about what he and Attila and I did is known because of his subsequent research.
So their problem was they had a system that worked through the mail, right?
And it had some other problems too, but basically the Department of Education sent out a letter to all 90,000 applicants and said, tell us where you'd like to go.
And people applied and they listed five schools in order.
And their application was Xeroxed and sent to the schools they'd listed.
And that was their application.
So that has a lot of problems right there.
As a principal, you could see that you're my third choice.
And you might decide that- And this is in the 2000s, right?
Just to be clear, just to the timing.
This is in the 2000s.
And they were still using a process like this.
Yes.
And that would show you that- You were my third choice.
And you might decide, you know, I have a great high school.
I'm only going to consider people who list me as their first choice.
So right away, there were reasons to have to strategize and think about what my first choice should be.
But also, then the high schools made their decisions.
They forwarded them to the New York City Department of Education, which sent letters to all the lucky people who had been admitted to at least one high school.
And some of those letters said, you've been admitted to three high schools.
choose the one you want.
And they waited for the decisions to come back.
And then they sent them to the schools and the schools made new admissions offers.
They only had time to do that three times.
It was August by that time and 30,000 students were still unmatched.
And they just, you know, if you're 14 years old, you have to go to.
school in New York.
So they assigned them to the nearest school that had an empty seat.
And people were very unhappy with that system for good reasons.
It's like chess by mail, except it's deferred acceptance by mail with like, you know, truncation.
It's deferred acceptance by mail with a very short time clock.
With a very short time list, right?
Short list, short time frame.
So the games all end before checkmate.
Right.
Exactly.
Three move, chess by mail.
Yes.
So we looked into it.
We wanted to make sure that we weren't going to be the economists who drove the last middle-class students out of the New York City public school system, things like that.
We did a lot of talking to educators.
But eventually we built for them a deferred acceptance algorithm that met their needs.
And they've been using it since then.
The part that got vandalized by Mayor de Blasio was even when you have a deferred acceptance algorithm.
Instead of having 30,000 students who you can't match, you have about 3,000.
And they're often students who didn't have good communication with the Department of Education.
They may not have submitted preference lists.
To get down to 3,000, we had maybe 10,000 students who didn't match to one of the places they'd listed.
And we gave them another opportunity to submit lists.
And that part has been truncated and replaced with waiting lists, which are not well defined that I believe have...
Their procedures have changed each year.
That's part of the story about not being well-defined, how the waiting lists are run.
But the main match, there were three parts of the match.
One was to the exam schools, because there are regulations that make it sensible to do them first.
Then there was the main match that most students get matched in.
And then there was the supplementary match in which people who didn't get any of their listed applicants got another chance at schools that were still...
had empty spots.
And I believe now only the first two parts are operating.
And then there's something happening in that third part to get those students into schools.
I'm curious, what was the alleged reason for doing away with a supplementary match that you mentioned?
There wasn't a good reason.
They never said what they were going to do.
I think there was generally some anxiety about going into the third match.
I mean, for sure, going into the third match was bad news.
It meant you hadn't matched any place you'd listed.
And they decided that since it was bad news, they were going to do away with it.
And I believe they hadn't thought about what they were going to do because they've changed each year since.
One of the things about market design is you think in advance about what you're going to do.
I have a quick footnote because you mentioned it a couple of times.
The blog, you started that when I was taking the market design class in 2009, or fall 2008, rather.
And...
It has continued essentially daily ever since then.
First of all, we should plug it because it's awesome.
Like, you know, if you're not following marketdesigner.blogspot.com, Al won't advertise it, but I'll tell you, you should go look at it.
But also, can you talk about how it's related, like how it's...
linked in with your work?
So I started it for my class in market design.
I wanted people to know that market design problems don't just occur to you when you're reading papers in Econometrica.
They can also occur to you when you're reading the newspaper about markets that are having problems.
Also, market design has lots of different applications and markets can have lots of different problems.
There are lots of themes.
It became a habit.
And so now I just do it.
But it's a source of ideas and it's a source of memory.
It used to be that if I read an interesting New York Times story, I would forward it to you if I thought you would be interested.
And I'd copy it to myself.
And I would hope that when I wanted to remember it, I would find it in my emails.
And that didn't work.
But now I can find it in the blog.
So I've just written a book.
I have a book coming out called Moral Economics.
It's about repugnant transactions and, you know, so controversial markets.
And a lot of it is I've been blogging about.
controversial markets for years.
And when I think of something I want to write about, I look to see whether I've blogged about it.
And often I have, and that reminds me and helps me get going.
This is funny.
I will say, I also use your blog as an index of market design problems and information about them.
So I guess it's heartening to learn that you do this as well, because I just sort of assumed you had it all in RAM all the time.
All the time.
Absolutely.
Remember everything.
We'll definitely return to Repugnant Markets in the new book.
But let me just let me stay a little bit longer with the school choice application.
In particular, you know, when you speak about the economist as engineer, you speak about the importance of spending time in the trenches, really studying the specific details of the application.
And just, you know, quick personal story.
I mean, I had that very front in my mind, actually, when I started.
doing research in the sort of blockchain protocol space.
It's one of the many benefits I get from my association with A16Z is really seeing what's in the trenches and having that inform my research.
And really talks I've seen you give over the past 15 years are in my mind when I think about the time that I spend that way.
So returning to school choice, what were the...
application-specific details that were really important there?
What made the problem different, for example, than the work you did matching doctors to hospitals?
So some students were getting multiple offers, and we wanted to know whether maybe that was important to them, and that if we ended up just giving them the best offer that they could have gotten according to their preferences submitted beforehand, whether that would do them harm.
So we looked at where they went, the students who got multiple offers, and mostly they went.
to the place that they had said they liked best.
And so we decided that we wouldn't harm them by taking that away from them.
And one of the things that speeds up the process, it's not just that it works on a computer, although that is faster than the mail, it's that we ask them for their decision before they begin.
That is, we ask them to submit preference lists, and those are their decisions.
And what was taking time, I think, in the letters was not the mail, but the decisions.
So only 17,000 students got multiple offers, if I remember correctly.
But those 17,000, what a relief.
You know, you applied to high schools.
You got in not just to one, but three.
The neighbors should know.
Your friends should know.
You should talk to everyone about it.
You should think what to do, even though you'd already had a rank order list.
And I think that took takes time.
And having those decisions made first speeds it up.
You still need computers.
I mean, Congestion is a really important part of markets, and I've studied markets that tried to do deferred acceptance by telephone.
And that turns out to be very hard just because of the way rejection chains propagate.
So actually, maybe for the audience, could you elaborate a bit on when you say congestion?
So congestion has to do with not having enough time to evaluate or transact or do all the things you have to do in the market.
The nice thing about a centralized clearinghouse like the medical resident match is you make a bunch of decisions up front and then everything happens algorithmically.
But if you were trying to do that by telephone, I studied a telephone market for professional psychologists.
And they tried to implement the deferred acceptance algorithm by telephone.
So at 9 a.m.
on selection day, all the phones are ringing as employers are making offers to applicants.
But by 9.30, most of the phones are silent because most of the applicants have received an offer, but they're waiting to see if they receive a better offer.
Now, how do they receive a better offer?
Well, somewhere in the country, someone gets a better offer, and then he calls back the place that made him.
The offer he was holding and he says, I got a better offer.
You know, thank you so much.
I'm not going to come.
And they have to put down the phone and call someone else.
And in the market we studied, they were so experienced and fast at that, that that exchange took only six minutes, about a minute for me to say, no, thank you.
I'm not coming.
You know, it was really nice of you to make me an offer.
And about five minutes for you to call Scott and say, Scott, we've been thinking about you all morning.
You know, we'd like to make you an offer.
And he already knows the nature of the offer.
You know, all that stuff has been.
So you're just saying to him, now we're making an offer.
And he says to you, that's great.
You know, you're one of my top choices.
I'll get back to you later today.
You both hang up.
And now he has to call someone and do another thing that takes six minutes.
And so you can only do 10 of those an hour.
That's the rejection chain that's taking time.
Even though each transaction is very, very fast in the world of labor markets.
You can't do a whole country's labor market by phone.
This one has so many resonances in different contexts.
I mean, I at least have done market design in, you know, both in crypto land and outside, right?
Like, you know, with cross-country vaccine allocation, you face versions of this, right?
Like, you know, vaccines expire in...
Crypto, often there is literal latency in the system because of the way the computational time and power is rationed by the system.
You literally can't do more than a certain number of computations per block, which then means that you can't do more than a certain number of market operations per unit time.
I was a customer in some NFT sale that had to space out over time because there were so many individual transactions people were doing.
It was actually causing running out of block space.
Market design is a way of both undoing congestion and sort of, sorry, un-nodding, I'll say, instead of unraveling, because unraveling is bad.
Un-nodding the congestion.
De-congesting.
It's like sinus decongestion.
Good.
What's the framework for that?
How do you go about it?
And again...
on the theory to practice theme.
I know this is one of the places where the theory and the practice sort of sometimes dovetail and sometimes practice requires engineering on top of theory.
Well, there's lots of ways to decongest markets because markets that fail to do that don't get very large.
So think about Amazon.
There was a time when Amazon sold only books and it's conceivable that you could had been an author or browsed by topic.
I mean, like, you know, maybe the Dewey Decimal System.
But now Amazon sells everything.
If you had to look in alphabetical order at all the things they offered before you could buy something, you wouldn't be able to buy anything.
But there's search, right?
So that decongests Amazon.
You can quickly find what you want to buy among their vast inventory.
Think about Airbnb.
When Airbnb started, you had to try to...
reserve a room that other people might also be trying to reserve.
Supposing you were trying to reserve a room in my house, you know, our guest room, supposing we put it on Airbnb.
In the old days, I would post it in the morning and then I'd go off to work.
And then I'd come back from work and look at my computer, my home computer, and see if anyone had applied to reserve our room for tomorrow.
And if four people had applied and you weren't the first, I would write to you and say, I'm so sorry, the room is taken.
And so you would have wasted a day and now have to try to get a room again.
Well...
That was fine when they were competing with other websites like Crash Pattern in London.
But today they compete with Hilton Hotels.
And that isn't the way Hilton Hotels works.
When you call up Hilton Hotels, you say, can I have a room tomorrow night?
They don't say which room.
They can tell you pretty quickly whether they've got a room.
And if they said to you which room, you'd be astonished.
And you'd say, well, how about room 2112?
And they'd say, I'm so sorry.
You know, that room is taken.
And they'd hang up.
And then you'd call back and say, how about room 2111?
They'd say, oh, yeah, sure.
But that was how Airbnb worked.
You didn't get my room, you had to get someone else's room.
Well, Airbnb has done a lot of stuff since then.
First of all, if you have a high reputation and if I'm willing to take people with high reputation, then you could just click and not wait for me to confirm your reservation.
But even if I want to confirm reservations, once someone is online, they take my listing away from your webpage, so your app.
You won't get second in line.
You'll only see places that you could get.
So they've gotten rid of a lot of congestion, and indeed, they compete with hotels now.
But they had to do that in order to compete with hotels, because all of a sudden, they had a lot of rooms.
And congestion would take time to be rejected from one room and apply for another.
It's interesting that in both this and the school choice context, resolving sequencing issues up front, sort of like agreeing on a fixed order given what's available, actually was the core congestion mitigation, which is definitely also, of course, sort of familiar to our blockchain listeners.
We're dealing with a big congestion problem right now in allocating deceased donor organs for transplant because organs don't last that long after they've been recovered.
But it takes time for people to decide whether...
You make an offer to a transplant center for a particular patient.
So there's a big controversy over what are called out-of-sequence offers, which is when the organ procurement organization starts getting worried that maybe this organ isn't going to find a home while it's still viable because being on ice isn't good for it.
They start making offers out of sequence.
There were some initially agreed on sequence, but they sort of say, we better get this organ transplanted quick.
And they jump ahead to someplace that they think we'll better get.
And actually, I think that comment's a great excuse to zoom out a bit and talk about kidney exchange, because some of our listeners may not even be aware that a large number of kidney transplants are actually sort of the product of a market which L.U.
helped design.
So I know you've told the story many times, but it'd still be great to hear the background again.
Well, one part of the background is you can get a kidney transplant from a deceased donor or from a living donor because healthy people have two kidneys and can remain healthy with one.
Another part is you can't pay a donor for a kidney.
That's a repugnant transaction that's against the law almost everywhere in the world, with the singular exception of the Islamic Republic of Iran, where you can pay a donor for a kidney.
There are black markets, but mostly, and certainly in the United States, Kidneys are gifts.
Their price is zero and they're very scarce, therefore.
There are half a million people on dialysis in the U.S.
There's 130,000 people who experience for the first time kidney failure each year.
But we only do about under 30,000 transplants a year.
So we have 130,000 new cases.
We do 30,000 transplants.
Most people who need a transplant die without getting one.
One of the ways you can get a transplant is someone...
healthy enough to give a kidney loves you enough to give you a kidney.
And there's lots of living donor transplants.
But often, you can't give a kidney to the person you love, even if you're healthy enough to give someone a kidney, because kidneys have to be well-matched to the patient.
So you could be in the situation where you want to give a kidney to one of your kids, but you can't.
And I want to give a kidney to one of my kids, but I can't.
But maybe I could give a kidney to your kid and you could give a kidney to my kid.
That's a kidney exchange.
And that increases the availability of living donor transplants.
And now we do about a third of living donor transplants through exchange, often more complicated exchanges than just between two pairs.
So that's a matching problem.
It doesn't use deferred acceptance algorithm for...
You know, for you matching guys out there, it doesn't use top trading cycles either, although that was what we originally proposed.
It uses integer programming to find cycles and chains that do the job.
Those are computationally hard problems.
And some of the instances are...
computationally demanding.
They can run a long time.
It's not that the worst case is computationally hard, but you never meet a hard problem.
But we've never met one that we couldn't solve.
So you computer scientists can help us out by enlarging the complexity literature to not just deal with worst cases.
I mean, Tim has done a lot of that, actually.
Looking at you, Tim.
Well, you know, the good news is, is the area I like to call beyond worst case analysis has been a really very major theme in computer science algorithms research over the past decade or so.
And, you know, in addition to more, quote unquote, traditional applications, you know, like energy programming, like you say, Al, I mean, if you can imagine, like, you know, the rise of...
success in machine learning, first with the deep learning wave last decade, but then, of course, also with the Gen.A.I.
this decade.
I mean, that in some sense throws the gauntlet to computer science analysts more than ever to explain how we're sort of solving problems that the traditional theory says should be unsolvable.
So lots of very smart people working hard on these kinds of questions.
I'm happy to report.
You know, I'd love to linger on kidney exchange a little bit longer because, you know, I know you have a deep...
book of amazing war stories from, you know, this work.
So one idea would just be some of the challenges that came up in getting your ideas accepted, right?
I mean, even the idea that you would have a quote unquote market for kidneys, right?
I mean, that has to take a lot of socialization to get people even comfortable with that idea in the first place.
It literally required legislation, right?
Well, we went ahead without the legislation, but then we got legislation.
The American law that says you can't pay for a kidney, is a little obscure.
It says you can't give valuable consideration to a donor of an organ for transplant.
So the question is, what's valuable consideration?
When I just told you about kidney exchange, something valuable is being exchanged where, you know, I'm giving a kidney to your patient, you're giving a kidney to my patient.
That's very valuable.
But the Department of Justice wouldn't write us a memo saying that that wasn't what they meant by that legislation, which is called the National Organ Transplant Act of 1984.
So there was a Southern congressman named Charlie Norwood who proposed an amendment to the National Organ Transplant Act that said those words, valuable consideration, don't apply to kidney exchange.
And we were already doing kidney exchange at that point, but we're doing it without clear legal.
guidance.
And one of the heroes of kidney exchange is my colleague, Mike Reese, who's got a lot of innovation in kidney exchange.
And he said something like, you know, one of my patients was a U.S.
attorney for the Southern District of Ohio, you know, something like that.
I could have him sue me.
And then we'd establish in court that this is legal.
You know, and everyone who loves Mike said, don't do that.
Don't do that.
I can never tell what would happen.
It doesn't seem like that.
But the amendment.
failed to pass in the first Congress.
It was introduced, and it failed to pass in the second Congress.
And in the third Congress, Charlie Norwood, who had had a transplant, he died of an immunosuppressive-related disease, and his colleagues named the act after him and passed it unanimously.
So kidney exchange, which doesn't involve any payments between patients and donors, is legal in the United States.
But just to be sure, in the legislation, it's not called kidney exchange.
It's called kidney paired donation, which is a term of art that keeps the word exchange out of the description.
And what made it take off was non-simultaneous chains, right?
So the exchange I described to you between two pairs, we always do those simultaneously.
Because one thing that not being allowed to give valuable consideration means, consideration is a contract term.
It means you can't write a contract, a legally enforceable contract that says, you guys give us a kidney on Monday and we promise to give you a kidney on Tuesday.
So when we do pairwise exchange, we do all four operations, two nephrectomies and two transplants simultaneously.
And that requires four operating rooms and four surgical teams.
So there's congestion in the market.
That is, it takes resources.
It's hard to do these.
So a lot of kidney exchanges now are done non-simultaneously in non-directed donor chains.
So last year we had 500 some odd non-directed.
kidney donors in the United States, people who wanted to give a kidney to someone and didn't have anyone particular in mind.
And so they allow us to organize a chain that doesn't have to loop back to the beginning.
And in that chain, every patient donor pair can get a kidney before they give one.
And so if the chain breaks, it's not a tragedy where some pair has lost a kidney but not gotten one and no longer has one to participate in exchange with.
So It turns out you can make economic models in which that doesn't work at all because people renege on their agreement.
But this is one of those behavioral things.
It turns out we human beings are a lot nicer than economists often give us credit for.
And so that turns out to be a rare problem.
And consequently, non-directed donor chains are a very productive way to get...
functions of kidney transplants, including transplants for the hardest to match patients for graph theoretical reasons that you guys will understand.
If we're hard to match, and I don't know who listens to your blog, but think of sort of Erdo-Shrini kind of graphs going to infinity.
That's on topic.
Yep, that's what I thought.
I mean, almost all of your listeners will now know what I want to say.
There's a way of...
looking at graphs that says if graphs are closely connected, if there are lots of possible transplants for every pair, then it'll be easy to take care of everyone.
But if some people remain in sparse parts of the graph, that is, it's really hard for them to find a kidney, then it's especially hard for them to find a two-way exchange because it's only a rare pair that can give them a kidney and it's only rare that they can give one back.
But if you have chains, so suppose we're all hard to match pairs, then the chance that we can do exchanges with each other is hard.
It's very small because the chance that I can take a kidney from you is very small because my pair is hard to match and the chance that you can take a kidney from us is very small because your pair is hard to match.
But if we have a big data set, if we have a thick market, then the chance we can give someone a kidney is not so small.
The chance that they can give it back to us is very small, but the chance that they can give it to someone else isn't so small.
And so we can construct a chain that includes very hard to match pairs.
So that's been a big development.
I just came back from a conference in Egypt in which it turns out there's a longstanding dogma of the World Health Organization that says countries should be self-sufficient in transplantation.
And that essentially prevents small countries from having kidney exchange.
From having transplants, right?
Yeah, from having transplants.
It protects them from certain kinds of medical care because they're worried about black markets.
But I co-chaired a consensus working group that got a consensus statement, the conference voted on it, that said that countries should be able to cross borders in kidney exchange.
And, you know, in particular, small countries working with large countries has some advantages for everybody concerned.
We'll see if that happens.
We have to write a paper.
We have to, you know, this is part of market design on a sort of diplomatic scale.
Well, that sounds like very important work.
And just, yeah, it goes to show there's the theory and then there's the 100 steps involving humans after that of actually, you know, getting people comfortable with it and getting it done.
I'd love to circle back.
So, you know, you brought up Amazon, you brought up congestion and how technology can increase or decrease congestion.
So maybe I'd love to just get your thoughts.
on how technology has changed, you know, both the theory and practice of market design, right?
Because you got to witness, you know, the rise of the internet in the late 90s and then the 2000s.
You know, I know you've, you know, I think have done some work with eBay at that time.
So, yeah, just a little bit about the internet, the web, technology, market design.
What role has it played?
You know, computationally assisted markets are sort of something new.
And we've seen a little bit of that.
You know, when I was young, a cash register was a mechanical machine.
You know, you typed dollars and cents on it and a cash drawer opened up and you made change.
But now, long before the Internet, a cash register was a laptop that had...
you know, cash register inputs, and it also kept track of your inventory.
It remembered what you just sold so you could figure out that you needed to reorder it, things like that.
So that was already adding some computation to markets and marketplaces at the local level.
And then the internet came along and all of a sudden markets became global.
You know, eBay, you could sell stuff on eBay to people far across the world.
One of Scott's colleagues once sold his collection of Pokemon cards to a dealer in Finland that way when he was quite a bit younger than he was.
Wait, I'm not totally sure I know which colleague this is, and I want to know why he didn't sell me the Pokemon cards.
You were also a child at that time.
I see.
Sure enough.
So, you know, all of a sudden a kid could sell his collection of cards to a store in Finland.
You know, that's pretty remarkable.
So that's also computationally aided markets, right?
I mean, the internet allowed him to do that.
And of course, computationally aided markets can do all sorts of computations, including computationally difficult computations.
But the resident match, I mean, when it actually began, you know, as you said at the beginning, long before my involvement began in the early 1950s, and it worked on card sorting machines.
So it was doing something that could be done with card sorting, physical card sorting.
I guess it's still doing something that could be done with physical card sorting, but that wouldn't be the way to do it because now there are 45,000 applicants and thousands of jobs.
So doing it digitally happens very fast.
So medical students have to wait.
They submit their preferences.
The hospital submit their preferences.
And then there's a waiting time.
A lot of that waiting time is checking, making sure you're now a medical student.
If you have a funny preference list, funny in the sense that you seem to have applied to places that didn't.
interview you, didn't list you, they'll call you up and they'll say, you know, you sure you haven't clicked on the wrong things?
You know, we want to make sure that when we run the match, we're running them on your actual preferences.
So they do a lot of checking of that sort to make sure that the inputs are correct, which takes time.
But once they are, you know, the algorithm itself runs very fast.
Now, you know, those are smart markets.
They do computations.
And we're going to see even smarter markets.
That's what we're seeing with AI, right?
I mean, possibly some of the checking can be done automatically, including communicating with people.
You know, we noticed something funny about your preferences.
Would you confirm that you've submitted the correct preference list?
You know, we're going to see more smart markets.
People are going to depend on them for...
matches with more or less understanding of what's going on and with the market eliciting more or less information.
So just as example, we already talked about Airbnb.
Let's talk about Uber for a minute.
First, those come at very different technological stages, right?
Airbnb could come when the internet was invented.
Airbnb became possible.
But you couldn't run Uber on the internet, you know, on your desktop computer.
Uber needs a smartphone.
It needs to know where you are.
And you have to be able to communicate with it wherever you are.
You're typically not at your desk when you're on an Uber.
But their matching algorithm, your desires are pretty simple.
Once they know your location and maybe where you're going, they can pretty much guess what you want.
You want is a car that will come quickly to where you are now and will take you to where you're going.
That's a much simpler problem than Airbnb has.
They can't just ask what city you're going to, where's your convention.
They have to show you pictures because you might like the house that has a view of the bay or you might like an ocean view.
So Uber knows to a good approximation what you want.
You want somebody to play fast.
And the drivers, we can think about their preferences.
So Uber just doesn't ask you any questions except, you know, do you want a more expensive or less expensive?
Do you want it?
quicker or slower.
I mean, they have a couple of questions, but they really know what you want.
They know the shape of your preferences.
And quasi-conversely, the Uber competitor that did ask you a lot of details, right, that actually had you try and choose between different drivers, this was sidecar, like, you know, went out of business.
There you go.
That's congestion for you.
Now, Uber suffers from congestion, too.
That is, if you go to San Francisco airport, there's a bunch of Ubers waiting for the next call.
And like...
potential transplant recipients who have waited long enough to be high priority, they're going to get many offers.
So if you don't want to go to San Francisco, if you, for some obscure reason, want to go to Palo Alto from San Francisco Airport, the first Uber driver online might not want to take you there.
He might want to go back to San Francisco.
That's where he gets most of his business.
So you might have to wait a little bit, during which time you might look at Lyft and take a Lyft.
So one of the problems Uber has is a little bit like the problem of deceased donor transplants is the patients don't wait around, you know, the rides, the organs don't wait around, they get cold and become unviable.
So the problem that Uber has of congestion, it takes time for the match to be made, is also very similar mathematically to the problem we have in...
in deceased organ allocation.
But Airbnb has a very different problem.
They have to show you pictures.
They have to sort of tell you a story.
You know, this is where you'd be.
And they don't like to tell you exactly where you'd be because they don't want to be worked around.
Disintermediated, yeah.
Disintermediated, thank you.
But they have to tell you quite a bit.
Otherwise, you can't form an opinion about which apartment you want.
So those are two markets that will evolve differently as markets get smarter.
You know, maybe your AI agent will know what apartments you like eventually, but somebody has to know what apartments you like.
Whereas for Uber, we sort of know what you like.
So earlier, this idea of repugnant markets came up.
So I guess the question here would be, is it ever inappropriate to solve a problem with the market?
Well, people think it is.
Some people think some.
Problems are inappropriate to solve with market.
I already indicated that it's against the law in the United States to pay someone to give you a kidney.
So it doesn't surprise economists that kidneys are in short supply.
You know, you can't use prices to increase supply.
Kidneys have to be a gift.
The price has to be zero.
You mentioned Iran earlier as the one place where there is a market.
And I assume the wait times there are not so high or lower than in.
So I think that's right.
But it's very hard to tell because I know much less about and there's much less data about what's the incidence of diabetes and of kidney failure and things like that.
So I don't have all the comparative stuff.
It's controversial in Iran, too.
Before COVID, I once gave a talk in Berlin about controversial markets.
And I talked about kidney exchange and surrogacy.
and prostitution.
And the reason those were three good markets for me to talk about in Germany, three controversial markets, is that the German laws are exactly the opposite of the American laws.
In Germany, the only one of those three that's legal is prostitution.
And surrogacy and kidney exchange are not legal, although kidney exchange, there's legislation moving forward.
So there isn't kidney exchange in Germany, and there's not.
There's surrogacy in Germany, but of course, surrogacy is legal in California and in New York and in Massachusetts.
So German couples that need a surrogate can come to the United States.
And then the German courts have to figure out how to repatriate the baby who is not legally recognized as belonging to this couple.
But what the courts have had to do is figure out how to let the German couple adopt their own baby, which, you know, constitutional courts think of.
high-level things, you know, is this something we want to support as a society?
But family courts have to think about who takes home the baby.
And one good answer is, how about the baby's parents?
So it turns out, just like markets need social support, bans on markets need social support.
Otherwise, they'll be subverted by people who need surrogate help to start their family, for instance.
And, you know, remember, we're all...
descended from long lines of people who never in any generation forgot to have children.
So the urge can be very strong, right?
Making it against the law to have children the only way you might be able to doesn't always stop people from having children that, right?
So what my new book is about is about controversial markets and morally contested transactions.
You know, what happens when you start to try to design?
these markets and sometimes inadvertently design the black market that replaces them.
So let's talk about moral economics a little bit.
What would you say are some of the key themes or a favorite story from it that you think would resonate with the audience?
Most of the legislators, legislatures in Western Europe.
where surrogacy is widely illegal, that is, they don't recognize parentage through surrogacy, they worry that it exploits women and they want to prevent exploitation.
They want to protect vulnerable people.
On the other hand, when bans on surrogacy are permeable, as they are because you can come to the U.S., you used to be able to go to Ukraine.
That was a big market for surrogates.
When they're permeable, it means they are going to be babies and there's no one more vulnerable than a baby.
Italy is trying to criminalize coming to the United States to have a surrogate.
But these other places where surrogacy isn't legal, they're trying to figure out ways to get the babies home.
And in places like Britain and Canada, surrogacy is legal in the sense that they acknowledge parenthood, but you can't pay the surrogate.
In Britain and Canada, having a surrogate is like having a kidney donor.
You can't pay for it.
So there are surrogates in Britain and Canada, but not nearly as many as...
You would think.
Well, maybe just as many as you would think.
I mean, not as many as you need.
Maybe as many as you think, given the price is fixed at zero.
Yes.
So they come to the U.S.
And the British law is it's legal to come to the U.S.
It's legal in the U.S.
You can come home with your British baby.
So that's the kind of thing that I think about.
What can you accomplish?
What can you not accomplish by regulation?
I also think a little bit about maybe what we should want.
Should we be trying to prevent surrogacy?
Now, American surrogates, you know.
Right, might feel differently.
Yeah, they get paid around $50,000 for a surrogate birth.
They're all already mothers, so they know what they're getting into.
It turns out not to be a terrible job for a woman with small children in the house being pregnant.
So, you know, we think it's okay.
There are other markets like that, like the market for blood plasma, right?
Plasma is an essential source of pharmaceuticals, cheaply having to do with...
immunology, antibodies, things like that, but also clotting factor for people who don't have it who are hemophiliac and a bunch of things.
And it's on the World Health Organization's list of essential medicines.
If you want to provision a hospital anywhere in the world, your pharmacopoeia has to have plasma products because you can't run a modern hospital without it.
The World Health Organization and the European Union believe that you have to be self-sufficient in plasma from unpaid donors.
And it turns out no one has been able to do that.
And so fortunately, you don't have to do it.
You can believe it's immoral to pay donors in your country, and you don't have to because you can buy all you need from the United States.
Because we are the Saudi Arabia of blood plasma.
We pay plasma donors and we export 70% of the world's plasma.
And that's why there are millions of deaths in Europe due to shortage of plasma products.
And it's interesting that these different rules regimes can coexist at the same time in different locations.
Right.
So think about marijuana in American states, right?
Marijuana was once widely illegal.
But I'm guessing that when it was illegal everywhere, there's some chance that you guys knew someone who would know where you could get some.
That is, there was a pretty...
big black market in marijuana among people of certain age and socioeconomic status and education.
But now, if you're in the state of Idaho, where it's still completely illegal, you're surrounded by states where it's legal.
So I'm predicting that eventually it's going to be legal everywhere in the U.S., even though it's still a federal Schedule I drug.
It's treated like heroin as a very dangerous substance with no medical uses.
Of course, it starts to have medical uses.
But there's also, you know, becoming widely available legally for recreational purposes.
So if you were a state of Idaho highway patrolman, there was a time where you could have stopped a car on the highway and made an arrest for.
somebody who had marijuana in their car.
And when you got back to the police barracks, the state police barracks, they'd clap you on the back and say, you know, go to rest.
You know, you stopped someone bringing drugs into Idaho.
But now they're going to say to you, someone didn't finish their magic brownies that they bought legally in Washington.
And, you know, they're on their way east where they're going to eat them at dinner.
After dinner, can you arrest them?
You know, why can't you be chasing bank robbers and murderers?
You know, we have real problems in Idaho.
So I think that it's going to get harder and harder for Idaho juries to be presented with cases they want to convict when marijuana purchases are legal.
There's also the problem of crime.
You don't want your market to be just run by criminals.
And of course, in the United States, we had an experience like that in the 1920s and early 30s with prohibition of alcohol.
And we prohibited alcohol.
It didn't reduce alcohol consumption to zero, but it was far from that.
It created a lot of organized crime.
We created the prohibition with a constitutional amendment, and we ended it with another constitutional amendment.
So there are now legal markets for alcohol.
We haven't solved the problem of alcoholism.
It's not that alcohol is unproblematic, but you can't buy moonshine whiskey from gangsters anymore.
They're out-competed by the fancy wine and liquor stores.
It's not that wanting to ban alcohol was a crazy thing, right?
And we still deal with those problems.
Alcoholics Anonymous was founded around the time that Prohibition ended.
Now we have sports gambling and Gamblers Anonymous.
Marijuana is going to be problematic.
It always was, but we didn't succeed in preventing marijuana.
And we created criminal organizations that sometimes also deal in other drugs.
You know, I think it's going to go the way of prohibition once it's fully digested how we're dealing with it in the United States.
So repugnant transactions are important.
Let me sort of add with a note that we're not out of the woods for artificial intelligence.
Right.
At Stanford, we are not allowed to use artificial intelligence to vet admissions.
I don't know about, I mean, I'm sure it's true for undergraduate admissions, but for PhD admissions, we're not allowed to upload all the...
dossiers we get for students into a large language model and ask it questions.
Okay, so Stanford is somewhat concerned that maybe we'd be uploading something with implicit biases that we didn't know about, or maybe they're concerned about what the newspapers would say if they heard that that's what we were doing.
But you can imagine that we're not too many terrible incidents away from having regulations that will say a lot of the things that you think are going to be good about AI shouldn't be allowed to happen.
So we're not out of the woods.
Markets need social support.
That's something that economists haven't studied enough.
I wish I understood it better than I do.
And one reason I'm writing a book about this is to say this is something to study.
This is part of market design.
It's just like you can't expect the world to read Econometrica and figure out how to implement the papers in it.
Neither can you expect everyone to say, oh, well, if an economist thinks this is a good idea, it must be a good idea.
Let's do it.
Or to switch to an infrastructure layer, like, you know, sort of think again, Cryptoland, right?
We have all of these infrastructure tools.
We actually need to solve a coordination problem to get people to use them.
It's another version of that, right?
But so with your examples of, you know, kidney markets on one side and then contexts like marijuana on another and then maybe surrogacy somewhere in the middle, do you have a sense of when markets are ready to move into design space versus when they're sort of just...
impenetrable like you're going to be banging your head against a marketplace wall.
I should be careful what I say here, but I think there has to be widespread dysfunction that affects lots of people so that there's at least the hope of making a Pareto improvement.
You know, when markets are really working badly, there's some chance that a better design will make.
everyone or almost everyone better off.
But until that time, there's a distribution question and there's therefore, you know, entrenched interest.
So I think this business with self-sufficiency and transplants, for instance, the reason this conference I just went to was an international conference, but it was in Egypt.
And the reason it was in Egypt is these rules about self-sufficiency are working against.
low and middle income countries to get transplants.
They should be able to cooperate with each other and pool their resources and find ways to match hard to match patients, things like that.
It seemed like a great idea that countries should be self-sufficient in rich countries, even though no countries are self-sufficient in transplantation.
We just don't serve the need, right?
People who need transplants die without them.
But still, it made some sense for rich countries to say, you know, everyone should act like us.
We worry about bad things that could happen, so we refrain from allowing them to happen.
And at the same time, that means we refrain from caring for a lot of patients.
But those patients, many of them, many, many, many of them are not in rich countries.
So in Egypt, people understood that these rules are working against our patients.
And we'll see what we can do about cooperating and getting cooperation between poor countries and rich countries.
And, you know, in lots of global trade, that's what makes sense in trade.
But the idea is that the shortage is so bad in transplants that you can start to form a broad coalition.
You know, one thing people sometimes ask me is, why don't we have something like a medical clearinghouse or a New York City public schools clearinghouse for college admissions?
Because college admissions is complicated and it's congested.
So there's problems in college admissions for both for applicants and for colleges.
But I think the colleges aren't yet feeling the pain enough.
that they'd be really open to coordinating with each other for a centralized market.
Not to mention there, I guess the costs of preference formation are really high, right?
So the coordination problem is there's both little demand absolute, but maybe also little demand relative to the cost of the switch.
Right.
Although there's lots of information about schools and there'd still be visits and things like that.
So one of the questions is how many places do you have to apply to?
And that's sort of a prisoner's dilemma problem.
The more high school graduates- Yeah, it's true.
The congestion actually supports a need to apply to many.
Yeah.
And so one of the things we're seeing in these congested markets is the growth of preference signaling, which we pioneered in economics.
That is, there's lots of signaling in labor markets, but it's signaling, often it's spent signaling.
It's signaling of your quality.
That is, you go to college to show that you can learn the kind of things that college can teach you.
And then you go get a PhD to show that you're the kind of person who can learn the kind of things that you need to learn to get a PhD.
And you write a dissertation and you get letters of reference.
You're sending lots of signals that say, I'm smart.
But when people are making lots of applications, it turns out it also helps to send signals that say, I'm interested in you.
So if you were to take the trouble to interview me, you'd have a chance of hiring me.
And, you know, when you're getting hundreds of applications for every position and can only have the time to interview a fraction of those people, then you value those signals that say, here's someone who looks interested in us.
And that's one reason why colleges look for expressions of interest.
You know, if you're ever going to take your kid to Look at college campuses.
You should register at the admissions office and take the tour so that they'll know that you took the trouble of visiting them and you might be interested in their college.
You shouldn't just park your car in a visitor lot and look around.
But there are now more formal ways of sending signals and you have to think about how to design those so that the signal is informative.
So you have to put a restriction on the number of signals you can send, things like that.
You see that in dating sites too.
Would you like to speak briefly about how it works in the economic sort of professor market?
My understanding is you.
were one of the designers of the way it works to this day.
Is that right?
I was.
So the American Economic Association allows new PhDs in economics who, well, they allow anyone, but they allow new PhDs in economics who are applying for jobs listed in the job openings for economists bulletins.
They allow them to send signals of particular interest to two of those jobs before interviews are scheduled.
So what does it mean to jobs?
You might apply like other people do.
You might apply to 100 jobs.
And you shouldn't signal Harvard or Stanford because we're fancy places.
We assume that if we would offer you a job, you'd be at least interested in considering it.
But I used to work at the University of Pittsburgh.
And that's a great American university.
It doesn't have the same cachet as Harvard or Stanford.
And so we understood that we couldn't earn our living by just interviewing the same people who Harvard and Stanford interviewed because we like the same people, but we needed people who might not get jobs at Harvard and Stanford, but would still be exciting and would be excited about us.
So allowing someone to send a signal to Pittsburgh says, I know that Pittsburgh is a fine place.
And if I don't get that job at Harvard that I've applied for.
I'd be interested in Pittsburgh.
And that distinguishes them from the people who are otherwise similar, but would be interested in Ohio State instead of Pittsburgh.
So it would help us organize our interviews and include some interviews that had more probability of turning into hires than ones that didn't have signals.
And among the places that like signals, for instance, are universities in England.
Because the thing about English universities, they get lots of North American applications and they hire lots of North American PhDs.
But every North American PhD could teach in England because of the language of instruction there.
And so the job for Oxford or for LSE or for Royal Holloway is to figure out, is this a person who knows that we're an excellent university?
And if they don't get a top job in North America, they might like to come here.
Or is this a person who really wants a job in North America?
But since it's so easy to add another application, they've added LSE, even though they prefer an undistinguished job in the U.S.
And if they can distinguish those, then they can start to figure out which Americans to invite for interviews.
There's this question of how do you signal interest?
You know, it's not just the peacock feathers which say, I'm a healthy male.
I can carry around this fan and not get eaten by foxes.
It's not just you should be interested in me.
That's the conventional signal.
It's I'm interested in you.
Given that you're facing congestion, you don't have time to interview everyone who applies to you, it might pay to interview me.
So why two?
Why two signals?
That's a good question.
The deep theoretical answer is we knew that one would help and a hundred would hurt.
If we had a hundred signals, then the fact that I didn't send you a signal would be really informative.
You'd say you had a hundred signals and you didn't send me one, you wouldn't even think about me.
So we picked two because it was bigger than one and smaller than a hundred and closer to one.
But we're seeing a natural experiment play out in medicine because The medical specialties, which are facing an influx of lots of applications for residencies, they all have written position papers saying, we're going to do what the economists do.
We're going to have signals of interest.
And some of them do what the economists have done and have relatively few signals of interest, which in a medical context means a handful, five or six.
But some of them, obstetrics and gynecology, orthopedic surgery, otolaryndology, some of them allow 30 signals.
And 30 signals, we're starting to see, are acting as a soft cap on the number of applications.
Because indeed, if I don't send you a signal, you won't interview me because I had 30.
So that's going to play out in a different way.
Cutting down congestion in a different way, maybe.
Yes.
So we're going to see why two and not 30, or maybe 30 would be better.
But that's going to play out in similar labor markets.
Marketplaces have to help markets be thick.
Once they're thick, they're going to have to deal with congestion.
You know, once you have a lot of transactions you can look at, and then you have to make them safe and trustworthy and reliable.
So, you know, these are just principles of marketplace design that are pretty general, I think, even though marketplaces and markets are very different one from another.
And maybe just to connect a couple of the other dots from the conversation so far.
So I loved what you said a few minutes ago about, you know, the big opportunities for...
New market design is where there's a plausible possibility of a Pareto improvement, meaning everybody's worse off.
Everyone's worse off under the current situation.
They're better off after the design.
Exactly.
So everybody should sort of be in favor of the new design, and it's largely a coordination problem.
And as you were telling that story, I, again, I sort of circled back to the example of doctors and hospitals, because actually I learned from your first book, Who Gets What and Why?
But I remember you talked about what things were like.
before stable matching was implemented.
And you talked about doctors in the beginning of their second year of med school, before you even really know, they know what specialty they want, right?
They're already getting sort of offers at that point.
So all of this just sort of connects and makes so much sense to hear you say this.
No, no, unraveling is a big problem.
You know, economists over the years have spent a lot of time thinking about the price of transactions without always thinking about the other dimensions of transactions, like their timing.
Incidentally, I mean, you know, matching wasn't always a big topic in economics.
There was this idea that you could sort of model everything like a commodity market, including labor markets.
It's just that the trouble with that in labor markets is instead of having some small vector of prices, one for each commodity, you had all these personalized prices.
You know, the wage that you'll offer to hire me is different from the wage that you would offer and the wage that I would demand to work in one job is different from.
in another job.
So all of a sudden the prices don't do the work that we think of prices as doing, of giving us a low dimensional vector that coordinates lots of actions.
So matching, you know, you care who you're matched to.
And so you can't just choose what you want.
You also have to be chosen.
So Scott, as a former student of ours, you've gotten lots of advice from him.
I know you wanted to maybe ask a little bit about the next generation of students.
Yeah, look, I mean, let me just also.
Again, shout out the blog.
I should shout out Al, too.
He's just an absolutely incredible mentor and advisor throughout, kind of like Tim has been at A16Z Crypto, I will add.
As you're thinking about students today, both people who are going to be scholars of marketplaces and marketplace design, and also practitioners, right?
People who are going to go out and build marketplaces in the field and maybe coordinate and collaborate with economists and other scholars.
What would you...
push them to think about and how would you, you know, how would you advise them?
Well, I mean, you have to think about markets in general and also in particular, right?
So different marketplaces are different.
One of the nice things about market design coffee at Stanford is, you know, we have people from Uber or Airbnb or Carta or Scale AI, you know, come join us at coffee and they all have different market design problems.
But...
There are some principles of design that you can think of, and one of them is that incentives matter.
Many companies don't really think about that.
Also, the thing about doing experiments, how to evaluate experiments.
There's a temptation on the web to do experiments, sort of A-B comparisons, and you do them on a tiny fraction of your transactions.
And if treatment A is better than treatment B, someone in marketing is going to say, why don't we go all to treatment A?
But if you have an economist on the team, they'll say, wait a minute, you know, no one noticed we were doing this experiment.
But once we give treatment aid to everyone, they're going to know that that's how we're organizing the market.
And they're going to think about how to react.
And we should think about that, too.
So lots of people don't think about equilibrium.
And economists think a lot about equilibrium.
But sometimes we don't think enough about equilibration.
How are we going to get to the equilibrium?
You know, can we fix this thing that we're worried about without breaking it?
So there's lots of things to think about.
academic discipline is going to evolve.
You know, my first half dozen papers on kidney exchange were in economics journals.
But my most recent half dozen papers on kidney exchange were in medical journals because we're trying to convince surgeons to do things the way we think they should do them.
And I think that's a natural evolution.
You know, I'm happy to have a platform in medical journals.
You know, you think about engineering.
Engineering is just applications of physics.
But...
They have their own journals.
You know, we were at a conference, a matching conference in Sicily this summer, and we visited before, and we visited Syracuse, where Archimedes lived.
And when Archimedes lived there, you know, the physicists and the engineers, they were the same guy.
But now, engineering is still about physics, but the civil engineers don't publish about bridges in physics journals, and the aeronautical engineers don't publish in civil engineering journals, and they're both using the same...
physics, which, you know, God created the heavens and the earth.
But they have to emphasize different things.
If you're building a bridge and there's some problem, you can often solve it by making the bridge heavier and stronger.
But that's not the way to solve problems for airplanes.
So I'm not sure that when, as we educate economists in market design, that we'll be just sending them off to economics departments.
And indeed, I have a student, Kurt Sweat from two years ago, who's in a medical school now.
He studied heart transplant.
And conversely, you know, economics departments are sometimes hiring computer scientists because they have the, you know, adjacent market design toolkit.
Same story.
We certainly should.
We're a little slow on that.
But it happened in my generation with game theorists, right?
A lot of the game theorists of my generation don't have PhDs in economics, but economics swept over us and was very welcoming.
And I'm still waiting for that to happen with computer science.
I'm just pleased we've made a dent at all so far, given how sort of traditional, I think, especially top economics departments are.
So I'm going to focus on the positive.
I'm optimistic.
It's directionally going the right way.
I don't see why it would stop now.
Yeah, exactly.
And maybe one thing I take away from what you just said, Al, was kind of the importance.
If you want to have impact, if you want your work to have impact, like...
Know who your audience is and go to them, like meet them where they are.
And if that means publishing somewhere where you're not accustomed to, go do it.
If that means spending time, you know, in your case, like in doctor's offices or, you know, in New York City Mayor's office, whatever it is.
Yeah, exactly.
So, yeah, no, absolutely.
We economists speak to each other in models.
No one else speaks in models, or at least in our kind of models.
So you have to learn their language so you can speak to them.
And you often have, I mean, one reason I wrote my previous book, Who Gets What and Why, was market design is an outward-facing part of economics.
So I've had many conversations in my life where I explain to people who are in a marketplace that's having some problem why they should be listening to an economist.
So I thought writing a book about that might lower the barriers to entry.
Scott, I'm sure this resonates with you even more than me, but the work we do at A16Z Crypto, we face that similar issue every day.
I mean, look, this has been utterly incredible.
Maybe if we could spend a minute on your outlook for the future, like what are you most excited about in the future we're going to see or that you hope we see?
Well, I'm excited about AI.
I remain excited about democracy.
I hope that we're going to keep having it.
I'm excited about the young generation of economists and computer scientists who more and more talk to each other.
I used to never go to computer science talks, but now the reason I don't go to them is we have competing economics talks and medical talks at the same time at Stanford.
So I think this is an exciting time for scientists interested in markets.
And of course, some of that work.
for various reasons, is moving out of universities into giant tech companies, the parts that involve training large language models, but other parts as well.
Nevertheless, I'm cautiously optimistic that American universities are going to remain the center of research for a lot of these things around the world.
I'm a little worried about that too.
You know, we've been a big exporter of education, which involves having lots of international students come to American universities.
So we'll see how that develops.
One of the matching markets that I think we need to pay more attention to is resettling refugees and other kinds of migrants.
Right now, that's clearly a broken market.
You know, when I go to Europe, I land at a European airport having flown across the water and I get off with my shoes dry.
A lot of people are trying to get to Europe in little boats over the Mediterranean and trying to get to the U.S.
across the land border.
And, you know, it's a matching problem.
They can't just choose where they'll be.
given asylum, and we can't just choose who will cross the borders.
So we have to do that a lot better.
It's obviously a real mess around the world, not just in the U.S.
on our southern border.
So that's a matching problem we have to deal with, because if the water continues to rise, there are going to be lots of human migrants around the world, internal migrants and external.
I mean, those are political questions.
But we're going to have to think about how to deal with people moving around from where they are to where they'd like to be and whether we want them to come.
So this whole question of what markets have social supports is going to be very big there.
It's going to be big on AI.
Think about things like facial recognition.
The city of San Francisco doesn't let the police use facial recognition.
You could imagine that if AI somehow makes an early wrong step of a highly publicized sort, it will have regulations that say you can't use AI for things.
Or you could imagine that AI will be beneficent and you know, augment jobs rather than replacing them and, you know, all sorts of good things will happen.
It's going to be fun to watch.
And let me just note one of the things that I learned from you very early, like one of the very first insights of your class, like when I took it, gosh, a decade and a half ago, was, you know, is that that outcome is endogenous, right?
It's not that these things are just, you know, good or like these markets are good or bad, right?
It's that...
through proper and social welfarely, you know, sort of maximizing or attuned, you know, marketplace design, you can actually get yourself into the equilibrium that has the better outcome.
You can certainly try.
You can work at it.
And that's part of market design.
Exactly.
It's not static, right?
It's not like the answer is yes or no.
These things are useful or harmful.
It's that we can think about the rules that govern the architecture of marketplaces to make them work better.
And we have to think about the transition, too.
Not just about the equilibrium, but about equilibrium.
So you guys are taking on a big responsibility.
We stand on the shoulders of giants, Al.
There you go.
Exactly.
You give an afternoon dinner talk at the Nobel ceremony, and mine had a metaphor of a pyramid of acrobats, each standing on each other's shoulders and lifting each other up.
That's a beautiful picture.
That's really excellent.
Fantastic.
Thank you so much, Al.
As always, it's super inspiring to talk to you.
Incredible, yeah.
Nice to talk to you guys.
Wow.
Well, that was...
That was quite the honor, getting to spend time with Al Roth.
Totally inspiring conversation.
Toward the end, we started talking about this notion of repugnant goods, repugnant markets, and moral economics, a sort of new book.
So going back to when you were there at Harvard, was that already kind of something he was thinking about at that time?
Yeah, it's funny.
He has this very well-known paper in the Journal of Economic Perspectives called Repugnance as a Constraint on Markets.
He's been thinking about repugnance for years and years and years.
And like the example he talked about with having to amend the National Organ Transplantation Act in order to enable kidney exchange, right?
Like, you know, it comes up in so many of the contexts that he's been working on because when you're going in to change the way a market...
functions, you know, you're going to go in and build a marketplace of one form or another, you need buy-in from participants.
There are lots of things that economists would like to say, oh, you know, we should just solve this, you know, design a market, like, let it go, we'll be done.
But actually, like, people are unwilling to do that.
By the way, crypto listeners, like, listen up.
This is like a huge part of how we have to think about building marketplace design in crypto land, right?
Like, part of the challenge when you're designing a market or a marketplace is It doesn't just have to work, but it has to work in a way that appeals to the prospective participants, right?
It has to, you know, enter into the sort of existing system in a way that participants are willing to adapt around your solution.
And so Al has all these stories, and they would come up in like almost every session of like something that made sense to do or would have in principle made sense to do and was repugnant.
Right.
It's just for whatever reason, the market participants said, no, no, no, you don't, you can't do that.
One thing I'm really excited about with this book is that it's something he's been talking about and running into for years and years, but has never really had a concrete framework.
Right.
There's never been a way to think about like how you interact with repugnance.
And it sounds like this book is going to give like the meat to the question of how do we reason about what is repugnant?
How do we reason about.
what you need to do to like work around it and like when it's a hard constraint versus when it's a soft constraint.
And again, you know, sort of, as I say, like this is something we need to do a lot of in crypto land.
I mean, one of those amazing, just, you know, his original theory work was in sort of the classic models, right?
So you talked about stable matching, which for example, did not have this idea of like constraints on couples.
And then I kind of feel like a lot of his journey has been the back and forth between the theory and the practice.
And you go out to practice, you're like, oh, couples are really important.
Solve the problem in practice.
But then don't stop there.
Right.
Don't stop there.
Write the theory paper that explains what's going on.
Yep.
Enrich the theory, take it to the next level.
And so perhaps like with repugnance, this is again what we're seeing.
Like you first encounter it out in the wild where markets that the papers tell you you should design, you can't.
And then you come back and you say, how can we update the theory so that we can sort of think clearly about these issues?
Absolutely.
No, totally.
I often describe it as like there's a diagram theory to practice to evaluation.
And then the evaluation sort of like points back to practice, but also points back to theory.
And you get, it's kind of hard to see in this visual, but it looks like a Nautilus, right?
Like you get this like sort of like golden section type, like spiral that constantly makes the marketplace designs deeper and sort of both more impactful and more intellectually fascinating.
All right.
Well, I think that's a great note to wrap up on.
So Scott, thanks so much for the co-host.
Fantastic conversation.
Tim, thanks so much.
This was awesome.