MICHAEL MACY: Welcome to the Cornell sesquicentennial celebration, and to our program on six degrees of separation. Our program celebrates 150 years of Cornell's contributions to networks that bring people together across vast distances-- from the telegraph network of Ezra Cornell to the discovery of small world networks in recent years.
Small world research addresses a puzzling empirical finding, that each of us may be just six handshakes away from a firefighter in London or a rice farmer in the Yangtze Delta. And this finding has been confirmed in at least three independent studies, two of which were conducted by scientists who will be speaking to you this afternoon. Six degrees has also been made into an award-winning play and into a movie by John Guare, who will be speaking to you as well.
So have you ever had this experience of traveling in a distant land, and you meet a complete stranger? And having a little conversational small talking, you discover that you know someone in common. And someone will then remark, it sure is a small world. But this is really a puzzle. Because how can the world be so small if there are over 7 billion people?
Moreover, it's not just the number of people that makes the six degrees of separation such a puzzle. It's also the way that we're connected. Social networks tend to be highly clustered, which means that our friends are also friends with each other. So a chain that goes from friend to friend to friend to friend to friend may actually not go very far, no matter how many links in the chain. The chain of friends just wraps around itself in these dense clusters. So in a highly clustered world in which everyone is connected to their friends who are connected to each other, how is it possible that people are on average just six steps away from one another?
Well this puzzled social scientists for about three decades, until the mystery was finally solved right here at Cornell University by the two people who will be speaking with you today-- Duncan Watts and Steve Strogatz along with Jon Kleinberg and Lars Backstrom. But first we'd like to show you a clip from the movie version of Six Degrees, followed by a comment by its author John Guare, who made the video because he could not be with us today.
[MGM LION ROAR]
OUISA: I read somewhere that everybody on this planet is separated by only six other people. Six degrees of separation. The President of the United States or a gondolier in Venice. Just fill in the names.
OUISA: How everyone is a new door opening into other worlds. It isn't just big names. It's anyone.
OUISA: A native in a rainforest of Tierra del Fuego and an Eskimo. I am bound up, you are bound to everyone on this planet by a trail of six people.
OUISA: Six degrees of separation between us and everyone else on this planet. It's a profound thought. How Paul found us.
PAUL: I'm so sorry.
OUISA: Why does it mean so much to you?
Flan, how much of your life can you account for? Answer me! How much of your life?
FLAN: What can I account for? All of it!
OUISA: What did you want from us?
PAUL: Everlasting friendship.
OUISA: Six degrees.
Stockard Channing. Will Smith. Donald Sutherland. Six Degrees of Separation, directed by Fred Schepisi.
OUISA: You have to find the right six people to make the connection. Six degrees.
JOHN GUARE: This whole concept of how many degrees of separation between everybody on the planet. It started not as a philosophical concept, I believe it was a statistical concept in the early part of that last century, in the early 1900s, the turn of the century, the 1890s to 1900, of how many stations-- with the whole world was starting to be connected thanks to Marconi-- how many stations would you have to go to to find the person you were looking for? And I think it was around 5.82 or something like that.
And then I was at Yale in the early '60s. And there was a man there named Stanley Milgram, still very controversial person, doctor, who would devise tests to see how far people's cruelty level would go. There were shocking tests, where people would pretend to be shocked by an electrical current that he was pretending to put in them. And he would give people the power to turn on that electrical current. Stanley Milgram also brought up the concept of how many people there were-- these six degrees between you and me and everybody else on the planet, which again was some sort of sociological concept.
And again, I've always been fascinated by that, that I can find anybody on the planet. Except in America, if you're below a certain level, you fall between the cracks. And that was the ironic statement that I wanted to make.
But man, I remember a couple of weeks or something after the play had opened, there was the front page of the New York Times said, the lead story was that there are only-- it was 1990, and it says, there are only six degrees of separation between Bill Clinton and the presidency. Well I said that has nothing to do with the concept with my play. But I loved it. Suddenly it got sucked into the daily, into the conversation, and became a byword even in "Six Degrees of Kevin Bacon," games happened to be. There are bars called Six Degrees. And anyway it's just there.
But I'm very, very happy that the play, it was a great, great joy to write the play. It was very satisfying. And a great, great pleasure to make the movie. I was tremendously happy with Fred Schepisi's production of the movie. And I'm very happy that fact of the play and this concept of six degrees of separation gets me invited to be part of this panel on this sesquicentennial of Cornell University. Thank you very, very much on this beautiful spring day. Thank you.
MICHAEL MACY: I'd now like to introduce our panelists. Our first speaker-- I'm going to introduce them in the order that they'll be speaking. Our first speaker will be Duncan Watts, the author of the book Six Degrees: the Science of a Connected Age.
Duncan started working on the problem of small worlds as a graduate student in theoretical and applied mechanics at Cornell in the late 1990s. He went on to become a professor of sociology at Columbia University, and director of the human social dynamics group at Yahoo! Research. Today he is principal researcher at Microsoft Research, and a founding member of their New York City lab. He publishes regularly in the most prestigious scientific journals, and has been awarded academic honors in both the physical and social sciences. He is currently an AD White professor at Cornell, which means that he will be a regular visitor to campus for the next few years.
Following Duncan, our next speaker will be Jon Kleinberg Jon received his BA from Cornell, and his PhD from MIT, and then returned to Cornell, where he is now dean of computing and information science and Tisch University Professor. Jon is a winner of the MacArthur Genius Award, as well as the most important prize in computational mathematics. And in 2011, he was elected to the National Academy of Sciences. His research has addressed yet another side to the puzzle of six degrees-- how people are able to navigate social networks to find these short paths.
The next speaker will be Lars Backstrom, who earned his BA and PhD in computer science at Cornell. He then moved to Facebook, where he used his network science to improve your Facebook experience. He is currently head of the team that runs the news feed.
So if you ever wonder how Facebook decide what goes into your news feed, you can ask Lars. If you ever wonder how Facebook decides which friends to suggest that you connect with, Lars did that, too. So Facebook is changing the way a billion people interact both online and off. And Lars is changing the way Facebook does that.
Our final speaker will be Steve Strogatz. Steve is the Jacob Gould Schurman professor of applied math at Cornell, with degrees from Princeton, Cambridge, and Harvard. He has worked on yet another puzzle-- how interdependent actors spontaneously coordinate.
Steve is one of the world's most highly cited mathematicians. And at the same time, one of the world's most highly acclaimed science writers. His four books show how mathematics can capture not only the synchronization of crickets, but can also capture the undulation of a suspension bridge, the romantic pendulum of Romeo and Juliet, and perhaps most importantly, how mathematics can capture the popular imagination.
So I promised you that we would reveal the secret of six degrees of separation. And who better to do that to start us off than the person who made the discovery, Duncan Watts?
DUNCAN WATTS: Thanks, Michael. So this story began almost 20 years ago just over there in the engineering quad in Steve Strogatz's office, when I was a graduate student. And as often happens with breakthroughs in science, we were thinking about a completely different problem. We didn't know about Stanley Milgram. We didn't know about John Guare. We were studying crickets, and we were trying to understand how populations of crickets that live right here on the Cornell campus in the late summer manage to synchronize with each other so perfectly, even though they're sort of spread out all over often quite large areas
And so while we were trying to understand this problem, I had a conversation with my dad on the phone one night. And he said this thing where he asked me, have you ever heard of this idea that everybody in the world is connected to the President of the United States by just six handshakes? And I hadn't heard of that idea. But I thought it sounded kind of curious.
And I wondered if it were true. And if it were true, would it be possible to prove it? And if it were possible to prove it, did it have anything to do with this cricket synchronization problem that Steve and I were working on? Because these crickets were all kind of listening to each other as well. But they're not all listening to each other equally. They're connected to each other by some kind of network. But it wasn't a network that we had a good language for describing in mathematics.
And so what Steve and I did was we came up with a language for trying to understand networks that were like the networks of crickets and other sort of complex systems in the world. And the way we did it was this sort of roundabout construction that you see up here on the screen.
If you were trying to sort of measure the number of links between everybody and everyone in the world, the obvious thing that you would do would be go out and map out that network and count. And that's something that Lars will tell us about later on. But back then we didn't have Facebook, and we didn't have any of these large networks. So we had to use math instead. And we did this sort of roundabout method of imagining what not just what the world looked like-- the network of the world-- but what all possible networks of the world could look like.
And so over here on the left-hand side, you have this world where everybody's sort of standing around a ring, and they're only friends with the people who are standing right next to them. And in that kind of world you have a lot of the clustering, that Michael just mentioned, where your friends are friends of each other. But you also have this problem that to get from anywhere on the ring to anywhere else in the ring, you have to do a lot of hops. So this is a big world where there's no six degrees of separation, but there are friends who are close to each other.
On the opposite extreme, you have this sort of randomly connected world, where everyone is choosing their friends randomly from the population of 7 billion people. And in that world, it's possible to very easily show that it does have this six degrees phenomenon. But it's a very strange world, because none of your friends are friends with each other.
And so these are the two kinds of extreme worlds that you could imagine in this little universe of possible networks. And what we did was just simply mathematically tune our way between them in this very sort of smooth way. And what we found was that even just introducing a tiny little bit of randomness into this sort of structured world on the left-hand side gave us these two properties that we were looking for. You still had the property that your friends were all friends with each other. And yet the path link between you and everyone else in the world had plunged dramatically, almost to the point where it was just like the random network on the right.
And so from this mathematical construction we were able to make a pretty big scientific climb, which is that not just social networks, but all kinds of networks ought to have this small world property. And we looked at a few of these early networks back in the late 1990s. And we were able to show that those networks-- networks of movie actors and power grids, power stations in a grid, and even neurons in the brain, and the nervous system of a worm called C. elegans all had this small world property.
And so we sort of convinced ourselves that this was something that we ought to find in the world, that it should have important consequences for how systems behave, whether it was systems of humans or systems of crickets.
But the one thing that we didn't get around to figuring out was the question that Ouisa-- Stockard Channing's character-- brings up in the movie, which is that even if it's true that people are six degrees away from each other, how do you find the right six people? And that's something that Jon cracked the next year.
JON KLEINBERG: Ah, right. So, right. So I had just come to Cornell actually when Duncan and Steve's paper was published. I had been out on the West Coast in the mid '90s thinking about web search, and how to make web search better, and this idea of trying to find things in a network.
And when I saw their paper, I went back and looked at some of the literature in all of this. I read Stanley Milgram's paper about the six degrees. And the amazing thing was he too had been really setting things up as a search problem, and trying to solve a search problem.
This is actually a hand-drawn picture from his paper in Psychology Today showing how his experiment worked. Yet the idea from this was right. I mean, Lars [? Volaire ?] tells about how you can do this when you have the entire Facebook network on a single set of hard disk drives. But in the 1960s, with pencil and paper and an experimental budget of $600, which is what Stanley Milgram had, what he did was he sent out letters to people in Nebraska. He said from his position at Harvard, Nebraska felt like sort of far enough out there that it was going to be a fair test.
And he hit 200 people. And he asked them, I have this friend who lives in Sharon, Massachusetts, a suburb of Boston. We're trying to test this idea that you're all six steps away from them. Not just a random person, not the president of the United States. I would like you to take this letter I have sent you, and I would like you to mail it to a friend of yours-- someone you know on a first-name basis, with the goal of advancing the letter as quickly as possible to this person.
And then his friends sat there, and letters started coming in. And actually a lot of letters came in. And then the median number of steps in these chains was six, which is the number that John Guare then discovered, and gave it this name, the six degrees of separation.
But I was looking at this as a computer scientist. And sort of what amazed me was that no one had asked the question, why in the world did this succeed? He had sort of asked them to do something very complicated, something that we ask the internet to do when it routes packets, but something you don't normally ask the population of the United States to do, which is to try to route these letters without the help of the postal system to this stranger that they've never met. How in the world is the network set to do this?
It didn't have to be that way. We look at the networks. The letter is zeroing its way in on the target. We could have lived in a world where the six degrees were there. But as Stockard Channing says at the end of the clip, it's sort of hopeless to find them.
But the world didn't turn out this way. The world turned out in this sort of much more tractable, friendly sort of way, in which you can actually route the letters to the place you're trying to go.
And so I played around with models for this. I'd try to take some of Steve and Duncan's models, which didn't really have this issue built into them. And what you could find actually is that some of the key is that the network has to be set up the way it sort of looks in this picture, which is that each of us has friends at many different scales of distance.
So we know a lot of people who live near us, typically. It's not true of everyone. If you just move to a new place, you don't know anyone. That may not be true. But most of us know a lot people who live very close to us. A lot of people who live, say, in our town or city, a lot of people who live in our county or state, a lot of people who live in our country, some people who live all the way around the world.
So these six degrees of separation are also sort of these six scales of resolution. And you expand out. And when the letter is trying to get to somebody, whenever you're trying to find someone, you sort of use them in reverse. You zero in through the country, through the city, the town, the neighborhood, down to the block. And it's crucial that all those steps be there.
And so as we think about human society, if we didn't have those long-range connections, we couldn't make these big jumps. But it's very easy to forget in the era of cellphone communication and the internet, we also need those last few steps. The letter could get all the way to Ithaca, New York and it somehow isn't going to find the person unless people in Ithaca, New York all know each other. And so it's really these scales of resolution that you need.
And so then, of course, the challenge was now we had a sequence of models. We had Steve and Duncan's. We had this proposal I had that you needed these different scales of resolution. And this is still the year 2000. And it just was completely unclear how we were ever going to test any of this. And I remember at the time saying to people to really test this, you would need to somehow build a system where you could convince a billion people to all sign up, say where they lived, and say who their friends are. And when is that ever going to happen?
And so remarkably quickly these systems [INAUDIBLE] became available. And that's something which, as Michael said, Lars is now really at the center of the Facebook.
LARS BACKSTROM: OK. So yeah, I came to Cornell in 2000. And I was an undergrad there. And then, as a grad student from 2005 to 2009, I worked with Jon on a bunch of these sort of social network questions. And we looked at a bunch of what are now in retrospect relatively small graphs of only a few million people.
And so when I was working with Jon, I learned about all of these kind of small worlds, and the theory behind it. And it was sort of clear that these things existed, that there were these paths that you could find, and that people could actually do it. But we'd never been able to sort of see at a large scale what did the entire social graph look like.
And so then in 2009, when I graduated and went to Facebook, one of the sort of obvious questions to start asking was, well, cool. Now I am at the center of a company that essentially has a large fraction of the social graph of the world.
And at the time, Facebook was much smaller than it is today. I think there were only a few hundred million people on it, whereas today there's over a billion. But even so, it's still a significant fraction of the world that was connected to the internet. And obviously it's sort of leaving out large segments of the populations that don't have internet access. But a lot of people were on. And you could sort of imagine that this was a good cut of the entire network.
And you can kind of see, I think, this is what it looked like a couple years ago, that you really do see people everywhere. That there's population density. And you see all of these kinds of long-range connections that Jon has talked about. They're obviously much more sparse. But you can kind of see how these packets would get across the ocean from one country to another.
And unfortunately, when I got the Facebook in 2009 even then the graph was much too large for us to kind of easily solve for this problem, easily look and say for any two random people what is the shortest path between them? You could kind of find pick two people and figure it out, and through this search algorithm, and for any two given people. If you really want to know what's the distance between all pairs of people, that was an intractable problem, maybe because there was hundreds of millions of people, hundreds of millions squared. A very difficult computation problem even at the scale of Facebook.
And so it happened that when I was in India, actually, for a conference in 2011, I ran into these guys who were from Italy. And they had just been working on how do you find all pairs' shortest paths on large-scale graphs. And of course they're not actually taking every single pair, because that would be computationally unfeasible. But they can approximate it. And when you're talking about something this large, you don't have to get the exact answer. The sort of statistical approximation works out.
And so in 2011 and 2012 we were able to go and take the entire Facebook graph, and look at how far are people actually apart. So we sort of knew the people could find paths that were six hops. What is the actual shortest distance between any two pairs of people?
And through that analysis, we found that it's actually six is pretty efficient. Most people are four or five hops apart. And so the fact that people could actually find a path that was almost the shortest in the entire world was quite impressive. And so we were able to go and look at that kind of macro level, what is the network, examine this thing that we'd sort of theoretically knew was there, but had never been able to study in such depth.
And since then, there's a lot of other kind of analyses that we've been able to do. Now that we have a sort of ever-growing sort of view of the social network, we're able to look not just at that kind of macro level of how far apart are people, but also start looking at some of the finer grained structures of what do relationships look like. What are the social networks of those two people? Can you kind of find people's relationship partners using just the social graph? And really study the social network of people in a way that was never previously possible. Because we have all of this data about all of these people from so many different walks of life. And you can now see in a way that was never previously possible what do social networks actually look like for people.
STEVEN STROGATZ: Well let me enlarge the discussion a little bit from networks of people to talking about all of science, and why I think all of us up here probably agree that networks may be the future for so many different branches of science, and for solving many of the major unsolved problems that confront us today in all disciplines. That is, this is much bigger than sociology, and much bigger than network theory, or mathematics. I mean this is really the future of science that we're talking about-- and technology.
So I want to begin by going backwards in time. What you see on the screen may look to you like a Christmas tree. But actually this is a naturally occurring phenomenon. Those are fireflies on a tree, in probably a mangrove tree near Bangkok.
So if you go along the tidal rivers in Thailand, every night of the year you'll see trees filled with fireflies-- male fireflies-- that are flashing on and off in perfect unison. This has been known for hundreds of years. I mean, of course, the people who live in Thailand have known it forever.
But as far as Western travelers to Thailand, the earliest report we have goes back to Sir Francis Drake in 1577. His ship's logs show that his crewmen noticed this unbelievable spectacle of mile after mile of riverbank filled with these trees that blink on and then blink off and on and off in perfect harmony.
Except how is this possible? I mean, what's causing this? And well, it's these little creatures, these beetles that are fireflies. And we have fireflies here in Ithaca and all through the United States. But they're rugged individualists. They do not particularly flash in unison. They can't seem to get their act together. And they're not interested in getting their act together. They're independent operators. But these Southeast Asian fireflies do somehow manage to fire in sync. And this is considered quite puzzling, because these are not the cleverest creatures on the planet.
Now the earliest theories about how it was possible, if you ask a group of people to get in sync, usually they have to follow a conductor. There has to be a maestro or a leader. And so there was the early theory that there was some maestro firefly. But no one has ever been able to find that firefly.
And you might also imagine that biologically it wouldn't make much sense for there to be one, because what if it got eaten by a bird? Then would the whole phenomenon fall apart that night? That's not how it works. It's not that there's a master firefly. It's also not due to the weather. Some people thought it had something to do with the humidity in the jungle.
In fact, the explanation we now know today is that fireflies organize themselves through some mysterious cooperative process. That is, they look at each other. In addition to flashing, they're responding to the flashes of others. And somehow through that give and take of flashing and responding to flashing, they spontaneously synchronize. And they do this, like I say, every night of the year.
So I bring them up as a metaphor for what I think is really the great problem of science today, which is self organization. Or to put it another way, how is it that you can have a group of individuals that somehow collectively do something fascinating and important?
I mean, if you think about, say, what the challenges are facing the geneticists working on the Human Genome Project-- maybe we should go to the next slide here. We now have these incredible diagrams of all the different-- this is actually not a gene diagram. But if we want to make sense of the human genome, we have something like 20,000 different genes coding for different proteins. And what I'm showing you here is just a little portion of the circuit diagram that controls, shows the main molecules that are responsible for cell division.
Now, you all know that cell division is the thing that goes haywire in cancer. Cells just keep dividing when they should stop. Cells keep growing when they shouldn't.
Now it's too small for you to see. But there's one little part of the diagram up in the top right. It says p53 box. p53, you may know, is a very famous tumor suppressor gene which, when it malfunctions because it's been mutated, this seems to be the case in about 50% of all cancers.
So understanding things like cancer is now not just a problem of biology, but the choreography of the activity of all these different enzymes and proteins and parts of a cell. That is, we're not going to solve cancer by-- maybe you just watched the show, actually, about cancer on PBS the other night, Emperor of All Maladies, which made the point that we hope that oncogenes would solve it, or that by understanding tumor suppressor genes we would solve it. But actually it's this intricate choreography going wrong in the cell.
So what I think here is that when we're talking about networks, yes, it's Facebook. Yes, it's Stanley Milgram and six degrees. But you should see this as the central challenge for understanding biology today, for thinking about the economy, for thinking obviously about terrorism networks. Pretty much every science today.
Mapping the brain. We have the project called the Connectome. President Obama says we're going to map everything in the brain, and figure out things about Parkinson's and Alzheimer's and the normal functioning of the brain. This, too, will be a giant network problem. So let me leave it at that. And I guess we're now ready to have a panel discussion. And then soon we'll invite you to participate.
MICHAEL MACY: Yeah. Thanks, Steve. So it's clear from all the things that you guys have said that Cornell has played a leading role in the development and advancement of network science, as well as computational social science. Is this just luck, or is there something about Cornell that encourages breakthroughs in this area, and the advances that have happened at Cornell? What's the secret?
JON KLEINBERG: It's a good question. I mean, one thing that's always drawn me to Cornell is just the kind of radically kind of cross-campus collaborations which you have going. And actually just before this panel [INAUDIBLE], in the preparations, we were trying to map out all the connections that we have among ourselves, among the organizers. And the number of ways in which-- joint advising of students, joint things across graduate fields, joint grant proposals, you know.
I think a lot of that, Cornell's really been set up to really make that possible. It's become a challenge, because when I first showed up, the idea that Cornell was interdisciplinary, it was actually sort of a novel concept. Now, of course, the word interdisciplinary has become so generic that people just sort of say it, and they don't even think about what it means. And I think a challenge for us here is to really how do we sort of keep that narrative going, where in a world where it's become such a cliche to talk about interdisciplinarity.
I think it is deeply built into Cornell, and its graduate field system, in Wegmans, where I've certainly had a lot of research ideas when I bump into people. And I think there's something to that. That when you talk to colleagues who are at universities in really big metropolitan areas, really the only time they see their colleagues is when they're sitting next to each other at faculty meetings or in the office.
And somehow I think there is a notion that we get along with each other here in part because you're going to see people on the weekend. You'll see them at the farmer's market. And these aren't just your colleagues. They're also your neighbors. They're really part of the community. And I think that's both good for the collegiality, and I think it's just a healthy for an institution.
STEVEN STROGATZ: Let me jump in on to reinforce Jon's point there, because maybe you think he's kidding about Wegmans. That's really the key thing, not just Wegmans and the farmer's market, but the idea that we get to know each other as friends, and come to trust each other when we see each other at day care-- whether it's for our kids or our dogs.
That is, what's the point about trust? The point is that if you want to do interdisciplinary work, you have to admit what a beginner you are. You have to admit how ignorant you are about certain fields.
So when I have worked with Jon, and I find that I know nothing about computer science, or with Michael, and I need to brush up on my sociology, it really helps to trust the person-- that is, to let your guard down and be vulnerable, and admit what you don't know-- so that you can actually learn something and make progress. In a place filled with egomaniacs, or with people who need to keep their academic armor plate up at all times, it's very hard to do the interdisciplinary work and to take the chances you need. And so I think that truly is a key to what makes Cornell successful. Something about Ithaca, something about Cornell.
The institution itself, as many of you will know, has a certain modesty to it, and humility that goes back to Ezra Cornell, I think. I mean, I feel like I can see it in his statue. I mean, if you look at him, he's standing in front of his great invention. He's standing in front of the telegraph machine. You can barely see the telegraph. This is how modest the guy is. It's right there. Go look at that statue when you're outside, and look at the back of the statute.
LARS BACKSTROM: So it's sort of like the network structure of Cornell and Ithaca has given rise to the success of the study of network structure.
DUNCAN WATTS: As a grad student, I always used to think that as I was sort of shivering through my first freezing cold winter here, I used to think the genius of Ezra Cornell was he put his university in a farm paddock in a totally inhospitable place. Because there's nothing else to do. So you just sit inside for nine months of the year and you think.
MICHAEL MACY: And you can't get in or out.
DUNCAN WATTS: And you can't get in or out.
MICHAEL MACY: It's centrally isolated.
DUNCAN WATTS: And it's a little bit of a bubble. People don't have normal conversations at Cornell. You don't sort of say, what do you do? You say, what major are you?
So there's a really a kind of exclusion of the rest of the world that you could see as a bad thing if you live in New York City, as I do, these sort of, you think, oh, it's so much more interesting to live here. And there's so much more diversity. And there's so much more connectedness with the rest of the world.
But there's also something about sort of isolating yourself in a bubble, and just kind of wrestling with hard problems for month after month after month. And it's hard to do that when there's a constant source of distraction, and a constant inflow of people, and all the things that you have in a big city. So I think it's both true that being here encourages the people who are here to interact with each other in a much more sort of extensive way than you would get with the Columbia faculty, for example, where everyone just goes home at night, and goes to whatever else they're doing But there's also this sense that, particularly as a student, where you're just here with your problem.
MICHAEL MACY: There's one more thing that I would add to what's been said. So six degrees of separation, and social networks. That's really a social science problem. And yet if you look around you on the stage here, I am actually the only card-carrying social scientist on this panel. And the interesting thing is that some of the very best social science being done at Cornell is being done by people who are not card-carrying social scientists, but people coming from computer science, from applied math, from physics, from information science.
And I think that's part of the culture of the institution, that people in areas that might otherwise be working on brains and crickets also are really fascinated by the motion of social particles-- not just those in the physical world-- genuinely interested in it-- reach out to social scientists to collaborate and work together on these problems. And I think Cornell is really fortunate to have this interest, this deep interest in social science-- deep both in terms of deep questions; deep also in terms that it goes way back in the institution. There's interest in social science from other disciplines.
DUNCAN WATTS: I would also say that something that's happened over the last 20 years is that the world has changed. And that the composition of people here on the stage reflects that. 20 years ago when you thought about networks, it was really a very theoretical exercise. That there wasn't a lot of data.
When Steve and I started looking around for examples of social networks to test our theory on, there were very few examples of networks that were larger than 100 people. Because it was just practically too difficult to go out. And back then, sociologists would go out, and they would hand out survey tools to people, and ask them, who are your friends? And who are your colleagues? And who do you ask for advice? And this sort of stuff. And it's very hard to do that for a very large group of people. And it's also hard to sort of sit and observe the interactions between people.
And so what has really sort of opened the floodgates to network data has been the web. And that's a technological revolution that has unfolded over the last 15 or so years, and of course, has transformed so many other things in the world that you're all familiar with, like e-commerce and communication, and in dating, and other forms of social interaction.
But might be less obvious to you, but certainly important to us, is that it's transformed how we do social science. That social science is becoming a computational field. And so a computer science-- just as biology did back in the 1990s-- so sociology is going through that revolution right now, and computer science is really a necessary partner in this.
JON KLEINBERG: I think these are problems that no one field can solve, at least as they're currently organized. And certainly all the things I've been doing, and I think everyone up here, we wouldn't be able to do them without being able to have really deep conversations with social scientists who are wrestling with these problems. And so I think it's really a case of several fields being stretched simultaneously outside their comfort zones, and trying to find the point where they can all meet and do something genuinely new.
Certainly from the computer science side, the thing that the web really taught us starting about 20 years ago was that we're really building systems for human beings. I mean parts of computer science were more forward thinking, and they saw that in the 1970s, and 1980s. But I think the mainstream onrushing mass of computer science really woke up to that in a big way in the early '90s, when suddenly you've had millions and then billions of people putting all this information online, and engaging in this outpouring of self expression and creativity. We suddenly though this is what we're building these systems for. It's for people.
And if you're really building it for these enormous masses of people, then you need to understand something about how they think individually. You need to understand how they operate collectively. You need to understand how the incentives that you create cause them to engage in new behaviors-- in a sense, how the system rebounds against your best efforts to design it. Because the system is populated by people with their own motivations and their own reactions to incentives.
And so I think just as the social sciences are being stretched by the appearance of all this data, those of us in computer science are really being stretched quite dramatically by the fact that these systems, that we're-- think of any big computer science system in the last 10 years-- YouTube, Wikipedia, Facebook, Instagram. All of these are about assembling audiences and then allowing them to really sort of unlock something that's sort of bigger than any one of them individually.
MICHAEL MACY: This is really an exciting time to be a social scientist. Social science has struggled for a long time to be able to have data. It's very hard to observe human behavior. People don't want you to be observing them.
We give out surveys, but those are to random samples. And those samples don't have people's friends and family in them. So we can't find out about their peer relations with surveys very easily. And we do experiments with what, 20 college sophomores in the lab, and then try to generalize. It's really hard to do social science.
But that's changing now. And this is an incredibly exciting moment in social science, because now, suddenly when it rains, it pours. We now have terabytes of data, the digital records of people's online interactions that are thanks to things like Twitter, and Facebook, and other social media, Last.fm. That we have just incredible data on human interactions that I think is going to be transformative in the years ahead.
And just as social media platforms like Facebook are changing the way we do social science, I think it's also the case that these social media are changing the phenomena that we're studying. Do you think that there are ways in which the entity of social networks, what we're studying, is changing before our eyes because of people choosing to interact increasingly all over the world, choosing to interact via these online social media?
LARS BACKSTROM: It's one of the things that we looked at when we did this study to see how connected people were, is as the network was growing and evolving over time, was it becoming kind of tighter? So we could actually measure, pick two random people, and maybe in 2009 they were four hops apart. Were they still four hops apart in 2012, or had some new connection been made that kind of brought them closer together?
And you could actually see this in the data, that over time, as the network grew, and as more connections were made, the world was actually getting smaller according to sort of like distance measure. And I think that's continued. And you kind of feel that in your daily life also, that technology.
And not just Facebook, but if you think about video conferencing, or how easy it is to make international phone calls, and how cheap it is compared to 20 years ago, all of these sorts of things have enabled us I think to connect much more effectively with our sort of weaker ties. I'm much more connected to my friends and family who live halfway across the world than I possibly ever could have been 20 years ago when it was $2 a minute to even talk on the phone. And I could email them, but that was never very efficient. And things like that.
DUNCAN WATTS: I think this is really a fundamental point. Because we often, when we sort of get excited, we sort of wax lyrical about how the internet is to social science what the telescope was to astronomy. And that maybe there's this sort of glorious revolution in social science that we're about to see.
But we have to keep in mind that because social science is about people, it's very different from studying ants or atoms. The ants don't care that you understand them. They just keep doing whatever they were doing before you understood them. And the fireflies are flashing exactly the same way today as they were 1,000 years ago. And they're totally oblivious to whatever math Steve comes up with to understand why they're doing that.
But everything that we build to try to understand human behavior changes human behavior. And so there is this sort of constant moving of the goal posts. And I think that's actually a good thing. We wouldn't want to have a kind of mastery of knowledge of human behavior the way we seek to have a mastery of knowledge of electrons. That's not something that I think aligns with our sort of desire to be human.
So I think it's good that social science is never going to be like physics. It's good that we're going to constantly have this process of reflection between understanding and the phenomenon itself. But it still allows plenty of room for improvement, and for science to inform other things, like business and policy and things that we care about.
MICHAEL MACY: Another difference with studying ants, or brain cells for that matter. When people are studying ants, we're not tempted to try to use our intuition to imagine what those ants are thinking, and how they're feeling, and how are they motivated. Whereas when we're studying people, we have that opportunity. We are one of the things that we're studying. We can then use our intuition to imagine how does the world look. Is this an advantage or a disadvantage?
DUNCAN WATTS: It's one of my favorite things to talk about. So I think it's both, of course. I mean, I think if we did not have the ability to sort of imagine why other people do the things that they do by reflecting on our own motivations and behavior, most of human behavior would be total [INAUDIBLE], totally inscrutable. We would have no idea why people do what they do. So it's great for us to be able to generate hypotheses about each other, to have this powerful brain that can sort of simulate other people's behavior.
But it also comes at a cost, which is that we have so much intuition for why people do what they do that we often are convinced with our own stories. We're convinced of our own explanations. And when we study ants or electrons, we also come up with stories. We call them hypotheses.
But we don't trust our hypotheses. And so we use this thing called the scientific method to test those hypotheses. And over the last 300 years or so that's been an incredibly powerful engine of knowledge creation, the scientific method.
And one reason I think why social sciences has not progressed as quickly as some of the other sciences is that because we have so much intuition, we don't use the scientific method. Why do I need to do some experiment to test a theory when I already know why? I know why poor people are poor. Or I know what to do about the economy.
Someone just the other day was telling me how he would have prevented the financial crisis. Everyone's sort of totally convinced of their own theories of the world. And yet if we actually go and test them, most of the time we're wrong. And I think that what we're also seeing now is that as we get more and more data, it's becoming increasingly clear that we don't actually know what we're talking about. And we have to be more scientific. And that's a healthy thing.
STEVEN STROGATZ: Can I ask you something about that? I mean, when you bring up the scientific method for social sciences, I think there are a lot of people-- I mean this occurs to me-- I bet to some of you too-- that social sciences make it very difficult in principle to do some of the things that you associate with the scientific method, like controlled experiments, for instance. That is, it's hard to keep everything constant, except one variable when it's people, or societies. So are you in a way taking an unfair shot at the social sciences by saying it's been a failure to use the scientific method? Maybe it's impossible in principle.
MICHAEL MACY: Now I think that randomized trial controlled experiments really are the gold standard across the sciences, for especially if we're trying to figure out causal processes, to understand what are the mechanisms that are producing the patterns we observe. And it's always been difficult to run experiments in social science, in part because we don't want to hurt people. I mean, there's the ethical considerations. But also just scale. I mean, we run these experiments, as I was saying before, with a handful of college sophomores in the lab who are not necessarily acting the same way people outside off campus would.
But again this is also changing, and in part I think because of social media creating platforms where we can run enormous experiments with much larger numbers of participants. There are online labor markets-- for example how Amazon operates an online labor market. Another called CrowdFlower, where at very low cost you can have thousands of participants in an experiment from across the globe in all different walks of life.
Facebook and other social media are running experiments 24/7 to improve the user experience. But sometimes these dovetail with academic interests-- a collaboration between Cornell and Facebook to study emotions, and how emotions are communicated from person to person. That was last spring. Now some of you may have heard about this. So I think that's a change that's happening for the better.
DUNCAN WATTS: Things that's happening may give a slightly different answer. I think you're absolutely right. There are experiments that are, of course, impossible in principle. You can't go to war with half of Iraq and not with the other half, and see which one works out better. So there's lots of limits to the scientific method.
But there are also limits to the scientific method in physics as well. We can't split the universe in two and change the gravitational constant in each half, and see how that changes the overall expansion or contraction of the universe. So there's plenty of things in all disciplines where it's not possible to run controlled experiments.
And I think that what's happening is that the gap is closing a bit, that there are going to be plenty of examples of things that we care about in social science where we either continue to not be able to gather the data that we're interested, or we continue to not be able to run experiments. But there are lots of things where we can now, that we couldn't before, and I think that's sort of good.
STEVEN STROGATZ: And also it makes me wonder if when you mentioned that the internet may be the telescope for social science, maybe that's very good as an analogy in the same way that, as you say, in astronomy you can't do controlled experiments. So astronomy is a great observational science, but hard to. So is that really the analogy?
JON KLEINBERG: So when Lars and his team learn something about the network, and put that in the system, they change the network. And that's very different. And I think it's also an argument really for how we're trying to educate students, say, here at Cornell. You're going to have people at these company.
So you're at some startup which is recommending streaming music to people. And suddenly overnight you've amassed an audience of 25 million people, 50 million people. And it's you and your four friends who just got BAs and whatever you got BAs in.
And you're sitting there and you're making decisions day after day. So you say, I have to recommend some songs. And should I show a friend who liked it? Should I show how many people liked it? Should I show three friends who liked it? These are all decisions. And they have psychological consequences. They have market consequences.
And you know the reaction, sort of the kind of stodgy economic reaction would be, well let's assemble a team. Let's get the psychologist here, and the economist, and let's assemble a focus group. But there's no time. You're writing code. And this thing is hurtling forward like a rocket. You're going to make some decisions. And you're going to draw on not just your intuitions, but hopefully on the courses you took. You're going to draw on your education.
And I think having that mix of backgrounds so that you understand how to build the systems. You understand the statistics of designing experiments. You understand the social, psychological principles. You understand the privacy implications. It's going to be one or two or three people in a lot of those cases.
Sometimes you have time to assemble the whole team. Great. But the number of these decisions that are getting made is several orders of magnitude bigger than the number for which you can really convene some kind of process like that. And you're relying on people with their undergrad educations that they got here or they got somewhere else to sort of have that kind of balance of backgrounds I think.
LARS BACKSTROM: One thing I think is interesting is a lot of the challenges you mentioned would apply to, say, medicine or something, too, where you have to be very careful. And they've obviously that's done pretty well.
I think one of the kind of unique challenges to a lot of the kind of network science is that you're measuring effects not just on one person, but you're trying to understand the entire network. And you have all these different connections. And the things that you do to a small number of people are going to potentially affect everybody.
And now it makes it actually very difficult to measure this. And you might think I'm in a pretty good position to run these kinds of experiments. And we do that a lot, as Michael mentioned, to try to make the user experience better. We're always trying to change the UI, or the way we do ranking, or things like that, in order to improve Facebook.
But even in Facebook, one of the things that we really struggle with is we build some new feature which we think is great. We're all really excited about it. And of course we're trying to be scientific. And so we run a test. It's a randomly controlled experiment. And you see that this feature that has some impact on the fraction of the people that you tested on. But it's very hard to predict how that's going to go both over time, as how are people going to use it in the longer term. But then also, how [INAUDIBLE] is it going to change when everybody uses it?
And the example that I always like to use is imagine that you built a chat service. So imagine that Facebook had already existed at some point long ago. There was no chat on Facebook. And imagine that some engineer built chat for Facebook, and they wanted to run a test. They wanted to run, like, let's see what happens if I give 1% of the people chat.
And then, of course, that would be a miserable failure. Because I would have chat. I'd be in the test group. And I'd be able to chat with 1% of my friends. And I'd think, well, this isn't a great experience. I'm probably going to go to use some other chat service. And then those kind of challenges.
That's obviously a very extreme example. But you can kind of imagine how that plays out, and how it actually presents somewhat unique challenges to the scientific method. Because you're trying to imagine what would happen if you were to expose everybody to this
JON KLEINBERG: I think a lot of the features of Facebook, because, I mean, the chat's a great example, because right, if you have no one to chat to, then it looks like.
But even you give someone a better photo uploading experience. I'm just going to make up an example, because I'm not building these things. But a better photo uploading experience to 0.1% of the population. Now that's going to have spillover effects, because you upload photos more fluently. So your output stream now has more photos. So your friends see more photos. They go, hey, that's kind of a good idea. Maybe I'll upload more photos.
You didn't even give them a new photo uploader. They just saw you using it, and they just modeled your behavior. And so you can have things leak out of your test anytime you're trying to create a social experience, even if it looks like an individual.
MICHAEL MACY: But there's actually, let me share with you a somewhat bigger problem. So as a social science, I'm sort of living in both these worlds, the social science world, and the computer information science applied math sociophysics world. Let me show you a little bit of the pushback that I hear from social scientists against these interlopers coming in from outside.
And if I could just use the analogy of Duncan's wonderful quote, we finally have our telescope with social media. OK. We have the telescope. But do you know where to point it? And the criticism is that people doing social science who were not card-carrying members are data mining. That's the thing that I hear the most. Just collecting data sort of mindlessly. Running some statistical tests to look for patterns in the data. And that's the end of it. Where's the theory? Any thoughts on that?
JON KLEINBERG: So first of all, I think in any science there's the phase where you just to figure out what's there. And so when you look at network science-- whatever you want to call it-- network science, this new merger of computational and social sciences, you should judge it as a science that's 10 years old, and you should ask what physics looked like when it was 10 years old, and you should ask what economics looked like when it was 10 years old.
And I think all of these engage in a kind of there's a phase where you're sort of groping around, and you haven't found the light switch yet. And so you're just sort of feeling, trying to figure out what's going on. And you're sort of mapping out the terrain. And I think all sciences go through that. And it's sort of a necessary step. You don't know what you're building theories of otherwise. And in a sense, sciences that sort of start building theories too quickly are building theories of not very much.
But I think it's a huge challenge. It's something that we need to start bringing theory. And I think actually what we started with here in this panel of six degrees is an interesting case of something where, in fact, there was a lot of theory building. That in fact a lot of it got validated.
So in fact the number of hops, and the kind of balance of clustering and path length that Steve and Duncan saw shows up in Facebook. The kind of distribution of scales, of resolution, that are necessary for searchability, which originally came out of a theorem I was trying to prove, also shows up in Facebook. And so I think we are starting to see the beginnings of things where the theory actually, in some cases, [INAUDIBLE]. But we certainly need more of those as the field gets older.
STEVEN STROGATZ: Michael, if I could suggest, I think the audience may want to either comment or ask questions. We have microphones for you up here in the front. So we'd be happy to react to whatever is on your mind.
DUNCAN WATTS: While people are coming up I'll [INTERPOSING VOICES] this the question that you asked, that I think it's absolutely true. There are plenty of examples of sloppy research that have been done by non-social scientists. But I think that the other comment is also true, for if you look back over the last 100 years of social science, there's a enormous number of theories about why people do what they do. And many of these theories are inconsistent with each other, or even sort of flagrantly in conflict with each other. And yet they sort of all continue to sort of co-exist, because we haven't had the data to be able to sort of reject them. And so I think that really what we're looking for is a disciplining process where you keep the theory, but you also have the data, and they talk to each other in such a way that both of them get better. And I think that's sort of what we sort of hope to see as the field matures.
SPEAKER 1: So I have two questions which may or may not be related to one another. In human societies, would the number have been six 400 years ago? And maybe related, has the internet strengthened weak ties and weakened strong ties?
JON KLEINBERG: So it's a great question. I think the number went down, certainly. I mean, I say with complete authority. I don't know.
I'll say this. In a world where very few people ever travel more than a few miles from their birthplace, it's going to be hard to bring the number down in a world that's 25,000 miles across. We certainly know from the very first book Cornell ever did in the freshman reading project, Guns, Germs, and Steel was in effect about how if you looked back 20,000, 10,000 years ago, the world was actually quite segmented. And that allowed, between the Americas and the Europe/Asia land mass, for example, diseases to evolve independently. And that meant that when those two groups were brought together, it had incredibly destructive consequences. And so we know that the shrinking of the world can be quite cataclysmic in some cases.
But thanks for bringing up the point about weak ties. Because I think one thing that social media like Facebook have done that's very subtle is on the weak tie part of this. The strong tie part is a question, too.
So Glenn is referring to strong and weak ties, the balance between your very close friends, and your very distant acquaintances-- the so-called weak ties. It's changed things dramatically, but subtly, where, say, when I graduated from high school-- and I know I've used this anecdote on my fellow panelists previously. But when I graduated from high school, I was in a gym with 200 people at our graduation ceremony. And on that day, I permanently lost touch with probably 160 of them. Sure, if I worked really hard, I could find them. But I don't know, are they married? Do they have kids? What jobs they have? Do they still live in upstate New York? I don't know these things.
And I thought that actually has changed, that people graduated from high school now have this expectation that they're going to be kept in touch. They're going to be able to remain in touch with people that they sort of knew growing up, but not very well. And that's because you have this computational prosthetic that's helping you, that kind of lightweight connection where someone that you don't know very well, you're not going to hear much. But every few years, when they get married, and they change jobs, they have a baby, you're going to hear about it.
Facebook enables to do that, because that's a connection that's too lightweight, I think, for humans cognitively to be able to maintain. It's either above that level, or it's zero. But Facebook can maintain that forever on your path. And it's going to tell you those things.
And I think that's really its sort of an amazing change to your picture of what does your life look like, and what is your whole lifespan reaching back to childhood look like, that I don't have. There are ways in which I'm cut off my childhood that I think people now aren't. And that's really thanks to having this sort of assistance in Facebook that makes that possible.
DUNCAN WATTS: Can I just make the opposite point, though, which is that-- and this is also somewhat surprising-- is that back in the early 1990s some sociologists tried to answer the question how many friends do people have? And the way they did this was a very clever trick. Because if you ask people how many friends they have, they have no idea. They tell you the wrong number. They usually underestimate wildly.
And so they used this trick where they asked you how many people do know called Michael? How many people do you know called Stephanie? And how many people do you know who are airline pilots. And using this trick, they were able to back out an estimate of how many total friends people had. And the answer was surprisingly similar to what we see on Facebook today-- that the average was a few hundred, and the maximum was about 1,500.
And so another sort of under-appreciated phenomenon that's happening with Facebook is it may not be that we have more friends now than we used to. It's just that we have Facebook counting them for us. And so it feels like we have more.
Jon knew the other 160 people in that high school gymnasium. But if you asked him two years later to name his friends, he wouldn't have thought to list any of them. So there's a certain amount of just a measurement effect.
JON KLEINBERG: I think there is a notion that I don't know what's going on with them. And I think now it's possible [INTERPOSING VOICES].
SPEAKER 2: So if political science is not an oxymoron, I wonder what your network research tells us about whether the political network, in terms of political views and communications, is in fact less connected than the social networks you deal with in general.
JON KLEINBERG: So Michael, anything about polarization [INAUDIBLE].
MICHAEL MACY: Yeah. We're studying political polarization. We're actually using Twitter to collect data, looking for the alignments of political alignments of people, also using Amazon co-purchases. So we've been looking at the ways in which the people who read, for instance, science, what do they read in politics? And surprisingly, we found that sociology is not the most left-wing of the sciences from the standpoint of the readers, the consumers on Amazon. It was engineering.
STEVEN STROGATZ: Engineering is the most left-wing?
MICHAEL MACY: Engineering was the most left-wing.
STEVEN STROGATZ: I think engineering is the most right-wing.
MICHAEL MACY: In terms of popular science. So that is to say, that people buying popular science in the engineering category on Amazon are buying also left-wing books. So we're trying to use these data from online sources to map the ideological alignments of various phenomena.
Also lifestyles, we've done the Fortune 500. Most liberal of the Fortune 500, Facebook. And in the past we would have used surveys. But by using data, use digital traces of people's interactions, and their purchasing habits, we could get, I think, a richer understanding of how political alignments are shifting and changing, and in some cases fissures forming.
SPEAKER 2: What about the dynamics of it, though? Is it changing?
MICHAEL MACY: Well that's one of the limitations, is that we can't go back farther than our data, which doesn't go back too far, because social media's fairly recent. There is evidence out there from using more traditional sources that the polarization is greatest at the top in Congress. Not as much in the population as it is among the elites. And also that it's been deepening since about the mid 1970s.
And also, one of the things, too, that our research has shown is that polarization, it's not just people taking more extreme positions on issues. It's that their positions on issues come to be correlated with each other. So that I can predict your views on gay marriage by knowing your views on gun control. Or your views on the war in Iraq, or on abortion rights. So the fact that issues come to be correlated is that we are seeing at the population level. Even though we don't see people taking more extreme positions we see people taking more correlated positions.
Yeah. One last question.
SPEAKER 3: You mentioned Stanley Milgram at the outset of this discussion. His best known experiment, of course, was highly unethical, involved deception and lack of informed consent, violated the Nuremberg codes. You also mentioned the Facebook emotional contagion survey, which also involved deception and lack of informed consent.
My purpose here is not to lambaste anybody, but to ask, as unwitting or witting participants in these social networks-- Facebook, LinkedIn, Twitter or whatever their successes will be-- what can we do to protect ourselves against this sort of manipulation? How can we recognize it, and whether it's done by the companies themselves, by government actors, by non-government actors, what can be done about that?
MICHAEL MACY: You're not from the IRB, are you?
DUNCAN WATTS: So I think there are two answers to the question. One is that we need to sort of adapt the procedures that were introduced about 30 years ago, that have sort of morphed into university institutional review boards. So there are procedures for reviewing human subjects research that need to be, I think adapted to this new digital age, where we're able to, sort of, the boundary between experiments and observation is sort of blurring, and the differences between surveys and experiments is blurring.
So there are a number of people here at Cornell, and also at even where I work at Microsoft Research, who are sort of actively thinking about how we should be structuring sort of ethical review of all human subjects research. And I think that's something that's very much in our interests as the research community to kind of get out in front of to make sure that we're being transparent, and we're able to build trust among the people who use all of these different products.
But I think there's another, larger point that that answer will not address, which is just that the world-- and it's not just the web, but in particular the web-- is moving in a direction where experimentation is just a ubiquitous activity. That all companies, if they're not already, will be sort of constantly tweaking things, and showing one set of people one thing and another set of people another thing, and then measuring to see which one works better. And you know, it is going to be sort of a universal fact of life that you're going to be in hundreds or thousands of these experiments on any given day. And I think that that's something that we need to again to make sure that people understand, and so that they can approach the life online with that knowledge.
And I think that if we go back to the 1920s, when public relations, was just being invented, they were pretty sneaky about stuff. The famous example is smoking, that the tobacco companies targeted women in particular. And they sort of covered this as a sort of civil rights issues, that women should express their freedom by smoking. And so that's sort of something that today it probably wouldn't work, because I think that we're much more savvy about marketing now as a society. And I think that going forward, this will sort of all become normalized.
JON KLEINBERG: No, I mean, I'm glad you brought up the question, because it's I think implicit in a lot of this. And I mean it's a very hard question, which a lot of us, wrestle with. I guess I would say three things.
So one, on the point of manipulation, the sorts of A/B tests that companies do on products is an example, and we should watch for that. But you know, I think one thing the internet has taught us is to be sort of able to be more savvy about manipulation in all dimensions-- through advertising, like Duncan said; through the news media, which draws a lot of its revenue through manipulation; through politics. All of this is about us becoming more educated about the many, many ways in which, I think, the system is designed to manipulate things. And so that educating people to watch for those things is very important.
On the subject of experimentation, since that's what we're talking about, I think I would also distinguish-- I think all four of us-- all five of us up here, if I could count correctly, would have different takes on this in a nuanced sense. But I would distinguish between principles and implementation. So the Belmont Report, which laid out the ethical framework for experimentation, I think says a lot of things that are absolutely the case, and which we actually have to adhere to, and I think everyone agrees on adhering to it.
Then there's this question of how do we specifically implement. Because a lot of those recommendations were expressed in general terms, and a lot of the IRB apparatus That we have, then, an attempt at an implementation of those principles. And I think you always want to try to make sure the implementation matches the systems which are available. So it's always been a work in progress. You articulate principles, and then you try to match the changing world with them.
And then I think the final point is that with, say, with Facebook or Amazon or Google, this notion that you're constantly testing your product, there is no other way to do it. i mean, the point is Google wants to know what color blue to show for its links. It's not like there's some default answer. This notion that well, why don't they just use the regular color? What is that? So every single part of the page, every single part of what they show you, every single-- those are all decisions that have no default answer. And so in some sense, the only way to do that is by showing some people one version, some people another version, and we'll see which one's better.
STEVEN STROGATZ: But Jon, sorry. Why can't some people just opt out of being participants? Why couldn't it be that I don't want any experiments done on me online. And other people, like you, maybe would say, I don't care if this is the way it has to be. Experiment on me.
JON KLEINBERG: First of all, there is no default. I mean, you can opt out of using Facebook.
STEVEN STROGATZ: No. I don't care. Whatever default they get by experimenting on you, I'll accept.
I'm just saying don't experiment on me. Leave me alone. Experiment on you.
JON KLEINBERG: This has actually worked quite badly. So there are these things where we got our beta tester group to do something. And who are the beta testers? They're these sort of techies who sort of have complete disregard for certain things that most people care about. And you end up with a product that's useless to most people. So in some sense you end up with a bad experience doing that.
Now you could say that's just the cost of doing business maybe. But one has to understand the costs of that, which are quite extensive. i think it's sort of, people say this thing like it's uncontroversial, that I believe in an evidence-based approach to policy. Most people in this room would say, of course. We want an evidence-based approach to educational policy, economic policy whatever. That evidence comes from trying different things, and comparing them. You don't get the evidence-based approach without being able to try things out and see what works.
DUNCAN WATTS: I think there's actually a really profound question here, which is that-- and I've heard a number of people say this over the years-- that they don't want to participate in experiments, or they don't want their data to be used for anything unless they get a direct benefit back, and that that is explained to them. And actually many companies would be happy to sign up for that. Because as far as they're concerned, like sure, we'll only gather your data to improve this product.
But for science, that's really a terrible outcome. Because really, it's this problem of public goods, that you're contributing something to the understanding of the world. Then, if everybody says what you say, which is like well, fine, I don't mind you experimenting on other people, but just not me, and everybody says that, then nobody learns anything. And then we're back in the world where it's impossible to learn anything.
MICHAEL MACY: But Duncan, social scientists have been running experiments for years with college students, and we don't force them to be in the experience. They get to choose. And I don't see why that can't be extended to online populations as well.
I do think we are out of time. So I get to have the last word on this.
Just in closing, I want to thank our panelists. I especially want to thank David Easley, the Henry Scarborough Professor of Economics, who organized this panel. I'd like to ask you all to give David a big hand. And thank all of you for coming.
We've received your request
You will be notified by email when the transcript and captions are available. The process may take up to 5 business days. Please contact email@example.com if you have any questions about this request.
Over the past decade there has been a growing public fascination with the complex connectedness of modern society. At the heart of this fascination is the idea of a social network. Understanding networks, the incentives they create, and the aggregate behavior they facilitate is vital to understanding topics as varied as how Google does web search and how Facebook became so popular. Part of Cornell's Sesquicentennial celebration, April 24-27, 2015.
Moderator: Michael Macy, Goldwin Smith Professor of Arts and Sciences.
Panelists: John Guare, playwright, "Six Degrees of Separation"; Duncan Watts PhD '97, author of "Six Degrees: The Science of a Connected Age" and principal researcher at Microsoft; Steven Strogatz, Jacob Gould Schurman Professor of Applied Mathematics; Jon Kleinberg '93, Tisch University Professor of Computer Science; and Lars Backstrom, Engineering Director, News Feed and Infrastructure, Facebook.