Sandy Pentland: “Social Physics: How Good Ideas Spread” | Talks at Google

want to welcome you all. My name’s Bradley
Horowitz, I’m VP of Social for Google,
Social Product Management. And I’m here today to
welcome Sandy Pentland to come in and speak with us. I’m going to be brief. I encourage you all
to Google Sandy. You will pull up his
long list of credentials, which include academic
credentials, business credentials, across
many, many disciplines for many, many years. World Economic Forum,
Forbes’ Most Influential, it goes on and on. So I’m going to try to give you
a little anecdote of something that I don’t think
you’ll get on the web. Nothing too embarrassing. I’m a student, both former
and current, of Sandy’s. Former in the sense that I
was a media lab Ph.D. student. He was my adviser. And current in the
sense that I stay closely attuned to
everything that Sandy does. As we were walking over
here from building 1900, we were sort of doing
what old friends do, which is play the name
game and checking in on all the old connections
and friends that we share. How’s Roz doing? How Stan doing? How’s Ali doing? What about Fad? And we went through
them, and turns out everybody’s doing fine. You know, many of you are here,
actually in the second row, in front row. Many of you have gone
off to become professors at MIT or Berkeley
or Georgia Tech. And it was just so great. And thinking about
that for a moment, I recognized that
I was just walking through one of the vintages,
the sort of early ’90s vintage of Sandy’s students who have
all gone off to do great things. And Sandy has
consecutively piled on top of that,
round after round, of graduate students and
students that he has inspired. And in addition to all of
those lists of accomplishments, one of the things that
really touches me most about Sandy and his
work is that he’s such an inspirational educator. He not only has enthusiasm
for his own work, he’s able to impart
that to others and create generations of
people that care passionately about technology and science. And it’s just so great to be in
the company, which we will all get to share for
an hour right now, of a person who can
inspire and lead that way. And so with that, I’ll
hand it over to Sandy. Welcome. SANDY PENTLAND: Well, thank you. [CLAPPING] SANDY PENTLAND: Now, I’ll
have to inspire and cause passion, which is
actually part of what I’m going to talk about. So how does that happen? So maybe this is good. So I’m going to talk
about two things. One is basic science
about who people are, how we use electronic
media, how we use face to face media, how we evolved
as a social species. And then I want to move to
how we can use this knowledge to make things
better, to have a more sustainable digital ecology,
to make government work. Wouldn’t that be
amazing? [LAUGHS] And so on and so forth. And as Bradley mentioned,
I do a lot of things. I thought I’d stick this in. I love this picture. This is the boards
back in the 1990s. And that’s Thad in
the front there, Thad Starner, who I
think most of you know. And then two other things I
do that are of real relevancy here, maybe three. Is one is for the
last five years, I’ve run a group– helped
run a group at Davos– around personal data,
privacy, and big data. And that’s, of course, a very
relevant topic for this crowd, but particularly, going forward. And the group includes
people like the Chairman of the Federal Trade Commission,
the vice president of the EU, Politburo members from
China, et cetera, et cetera. So it’s a conversation between
CEOs of major companies and chief regulators
and advocacy. And I’ll talk about
that at then end and where I think
that things are going and what you might
want to do about it. And I just joined
the Google ATAP Board because it used to
be owned by Motorola, but when Motorola got
sold, they intelligently moved the really creative
interesting part over here to Google. And as Bradley mentioned,
that started a bunch of companies, which
are doing well. So the thing that I’m
really concerned about, the thing that’s passionate, is
making the world work better. And a sort of story for
this is about 15 years ago, I was setting up a series
of laboratories in India. And, you know, we had huge
government sponsorship. We had a board of
directors, which are some of the brightest most
successful people in the world. And it was a complete disaster. And it had to do
with a lot of things. It had to do with all
of the sort of macho, signaling charisma in the room
with the board of directors. But it also had to do with
the way the government failed to work, or did work. And looking back on
that, I can sort of see that premonitions of
the US Congress today. All right? So we went and visited
the Indian Congress, where we saw people
throwing shoes at each other and throwing cash in the air. And we look at the
US Congress today, and it’s somewhat
similar, unfortunately. So I want to make things better. And what occurs to
me is if we knew how to make our
organizations work, then we could really do things. Like we could solve
global warming tomorrow if we all knew how to sort
of talk about it rationally, come to a good decision,
and then carry that through. And the fact that that sounds
like ludicrous fantasy– oh, yeah, everybody agree. Sure. Not in our lifetime–
tells you just how profound the problem is. And that’s why I think one
of the most important things that’s happened in the last
decade, something you’ve all been part of, is
this era of big data, which is not about big at all. It’s about personal data. Detailed data about the behavior
of every person on Earth, where they go, what they
buy, what they say online, all sorts of things like that. Suddenly we could
watch people the way, say, you would watch an
ant hill or Jane Goodall would watch apes. We can do that, and
that has profound impact that is hard to appreciate. I made this little
graph, which comes out of– inspired by
Nadav’s thesis here, which is duration
of observation. These are social science
experiments, and the biggest medical experiments. So this is like the
Framingham heart study. 30 years, 30,000 people. But they only talked to people
like once every three years. So the bit rate was like
one number per month. So you had no idea what
these people were doing. They could have been eating
fried chicken all the time. You don’t know, right? Or most of the things
we know about psychology come from down here. This is a number of bits
per second, duration. This is a bunch of
freshman in Psych 101 filling out some surveys. And that’s what we take
to be social science, political science,
and medical science. But now we have these
new ways of doing things. And so what in my
group we’ve done is we’ve built little
badges, like you all have little name badges. And so we can actually
know where you go and who you talk to. And we do this
with organizations. We’ll track everybody
for a month. We never listen to
the words, but we do know the patterns
of communication. And I’ll show you a
little bit about that. Similarly, we put
software in phones, and we look at the
patterns of communication within a community. So I go into a community
and give everybody brand new phones. And some of the people here
have been integrally involved in these experiments. And we’ll look at their Facebook
activity, their credit card record, their sleep pattern,
their communication pattern, who they hang with,
who they call, and ask, what do all this
communication patterns have to do with outcomes? All right? Do they spend too much? What things do they choose
to buy, and so forth. And what you find
from the big data, and, of course, modern machine
learning sorts of things, is that you can build
quantitative predictive models of human behavior,
which you all know. But I think you
know the wrong part. And I’m going to tell you
about the other part that’s much stronger than what
you typically do, OK? And so you can predict behavior. And people go,
well, wait a second. What about free will? That may not have
occurred to you, but that’s a traditional
thing to ask. And I’ll tell you a little
bit about that along the way because it turns out
that a lot of our behavior is very habitual,
and that’s the part that we can model
mathematically. So the big picture
here, and this is part of the reason I
got off onto this research, is I go to places
like Davos, and you listen to the president of
this and the CEO of that. And when they talk about
changing policy, talk about doing anything, they
use economics metaphors. And the thing about
[INAUDIBLE] economic metaphors is that they’re all
about individuals. So you’ve heard about
rational individuals. And everybody rags
on the rational part. I’m not going to do that. I’m going to rag on the
individual part, OK? Because I don’t think
we are individuals. What we desire, the ways we
learn to go about doing it, what’s valuable, are
consensual things. So they actually are
captured by this sort of model, this
independent model. That matters because
those interactions are the sources, not only of
fads and economic bubbles, but they’re really the social
fabric that we live in. So everybody knows
about the invisible hand that are led by
the invisible hand to advance the
interest of society. What that means is that markets
are supposed to allocate things efficiently and fairly, right? If you’ve thought
about it, you know this doesn’t work in the
real world, [LAUGHS] OK? And the question is, why? So one of the things
that– there’s several things to
say about this. Most people think that
this statement is something that he made in “The
Wealth of Nations.” And I’m just going
to [INAUDIBLE] “The Wealth of Nations.” But it’s not. He made it in a book called
“Moral Sentiments,” which very few people read. And it went on to say
something very different. It went on to say that
“it’s human nature to exchange not only goods, but
ideas, assistance, and favors. And it’s these exchanges
that guide people to create solutions for
the good of the community.” So Adam Smith did not
believe that markets were socially efficient. He believed that it was the
social fabric of relationships that caused the
pressures of the market to be allocated
where they’re needed. And, in fact, a lot
of mathematicians believe now that this sort
of despite Nobel prizes about market
efficiency and markets being good for
governance, it’s not. It’s really– you have to have
the right sort of regulation replication mechanism. But this is another solution. Adam Smith way back when said,
if we could model the peer to peer relationships,
we could understand how market things
eventually resolve to be much more efficient. And that’s what we’re doing. We’re doing something
that you could call, sort of, Economics 2.0. Instead of the
individual approximation, we’re now modeling not
only individual behavior, but peer to peer behavior
at the same time. So that’s the sort of big
context for what’s up here. So let me give you an
example of what that means. So in a typical situation,
you have some people that influence each other. So, you know, their political
views, it’s what’s cool to do. You pick it up from
the people around you. And we have a lot of
evidence upon this from the experiments
we’ve done in our group, showing that people’s
attitudes about, you know, what music to listen to,
what apps to download, what spending behavior
to have, is largely predicted by their exposure
to what other people do. And you may not
want to believe this because it’s not the
rhetoric in our society. And you guys are the last
people to be saying this to because you guys are like the
best and smartest in the world. But it really is true that
about 50% of the variance comes from these peer
to peer relationships. And we know that when
we do incentives, when we try to these–
CEOs and governors try to set up governance schemes
to make people do things, they always talk about
individual incentives. That’s part of this mindset
that comes from the 1700s, is that we’re all
rational individuals. So we’ll give this guy
money to behave differently. But when you do that, what
happens is, of course, you’re putting that
incentive in opposition to the social pressure he
gets from other people. And if they’re aligned,
it works wonderfully. But if they’re not aligned,
which happens all the time, incentives don’t work. Individual incentives
don’t work. And the moment the incentive
goes away, you know, you start paying them,
they revert to what the social fabric does. So when you begin to think about
this mathematical framework that includes the social
fabric, an obvious thing occurs to you,
which is that, well, instead of giving one
person the incentive, maybe I can modify
the social fabric. What we have is
exchanges between people, and we have incentives being
applied to the individuals. And now what I’m
going to do is I’m going to modify the incentives
between the people, OK? And you can write down
these equations just the way you write down
economic equations. So this was in “Nature
Scientific Reports” just last year. So you could write it all
down, you know, with utilities and peer pressure
and externality cost. And you’d discover
something interesting, is when you add these
second order terms in, you find that incentives
that modify the interactions are generically more
than twice as efficient as incentives that
go to individuals. Generically that way. And in the data
that I’ll show you, it’s sort of four to
20 times more powerful. So this is the thing that I want
to really get across to you, is that this sort of power of
economics that has mathematics and prediction is very limited. It doesn’t include anything
that has strong peer to peer effects, like
bubbles, like fads. But you can now write those
down because we have enough data and enough math
be able to do it. So let me give you some
examples of what that means. These are simple examples. You can do much more
complicated ones. So this was, again, sort
of done in Nadav’s thesis. We got a community, where we
divided them into two pieces. It’s actually three
pieces, but I’ll just talk about two pieces. In one piece, we
had a little app that showed them how
active they were. And this was in the
winter in Boston, where people tend to just put
the blanket over their head and go away, right? So we wanted them to get
out and around and stuff. And so we can show them
their activity level, and we can give them money. So you can say, oh, if I was
more active than last week, I made $3. Wonderful, OK? But in the other part
of the community, we assign people buddies. And you would be his buddy. If he is more active,
you’d get rewarded. Not him, you. Your buddy gets rewarded
for you being active. And what you do
is you pick people that have a lot of interactions
with each other to do this. And it sounds a little
creepy, but actually, people got into it. Almost everybody
signed up for it. And what you got is you
got everybody’s, like, looking at the other
guy and saying, well, are you being active? Because they’re reminded on
their phone all the time, and they’re getting a
couple bucks for it. It’s not a big thing, OK? But remember that that incentive
scheme, that social network scheme, if I’m not
incenting the individuals, I’m incenting the network, is
generically more than twice as efficient. In fact, in this experiment,
the way we did it, we found it was four
times more efficient. And if we’d done it the right
way, if we went back and did it again, as sort of a
post hoc analysis, we would have gotten eight to
10 times more efficient, OK? Pretty interesting. Oh, and one other thing. It stuck. When we ran out of
money, we turned off the money, no more incentives. People kept being more
active for the period that we could observe, OK? Because what we’d done is
changed the social fabric. We made being active a
topic of conversation, a topic of social pressure,
of prestige, of interest. And, of course, that
has momentum to it. Here’s another
example that’s online. We went to a canton
in Switzerland, which was trying to save power. Keep power below a
certain level because they had hydroelectric
power up to one level. And beyond that, they
had to turn on diesels. And noisy, expensive, polluting. And they tried educating
people, and they tried financial incentives,
and nothing really worked. And so we convinced them to
sign people up as buddies. So you would sign up
with him as your buddy, and if you saved energy,
he would get a reward. Now, the rewards
were, like, pathetic. The budget was
$0.50 per week, OK? And little dancing
bears on your website. It was, like, really stupid. But what happened is that for
the people that signed up, you got a 17% reduction
in energy usage. Now, that sounds good, but let
me give you the comparable. There have been some places
where people have raised prices to get 17% reduction. On average, you have to
double the price of energy to get that reduction
in energy use. That’s the price
elasticity curve. So for $0.50 a week, we could
get the same effect as doubling the price. This one is by a friend
of mine, James Fowler. Done with Facebook in 2010. He’s sent 61 million
people a message about getting out and
vote in the election. The sort of simple summary of
it is it had almost no effect. People didn’t do it. A few people did. He could go back
and sample people and say, how many people
did this have in effect. He also included a
“I Voted” button, which would then show your face
to your Facebook friends, OK? This also had no effect, except
with one particular class of people, which is the
people you had strong face to face relationships with. If you appeared in the same
images with this other person regularly, then
among that group, that “I Voted” button
would generate two to three more people voting. So greater than yield one,
a cascade of behavior. So, again, what
was happening there is social pressure, the
peer to peer things. The centralized
thing didn’t do it. It was peer to peer
pressure between people with strong ties. And, in fact, actually,
it’s not captured by the electronic
network at all. These are things that
were outside of that, OK? So that’s an
example that I think is really interesting
that you can build on. And somebody
mentioned, actually, how many of you know
the red balloon contest? [INAUDIBLE] knows it. [LAUGHTER] SANDY PENTLAND: [INAUDIBLE] OK. But so DARPA had this social
network grand challenge where we had to find 10 red
balloons somewhere in the US. And everybody tried using
the economic incentives. You know, you get
some money if you found a balloon a reported it. We used something
like this mechanism. We’re able to recruit
what we estimate to be about two million
people in 24 hours, and won. Again, not giving people
directly the incentive. In that case, it’s a
little more complicated, but giving people these
peer to peer things. So that’s cool. Why do you think
humans are this way? Well, let me give you an
example that I think really tells us why the action
is not between our ears. The action is in our
social networks, OK? We are a social species. We evolved that way. Why? Let me give you a really
graphic example of that. So this is a site called eToro. It’s a social network site. On this site, people buy
and sell dollars and euros and gold and silver,
and stuff like that, OK? And unlike almost every
other trading platform, it’s a social platform. So I can see what
every other person, what every of these 1.6
million people are doing. You can’t see the
dollar amount, but I can see that they’re
shorting euros, long dollar,
leveraged 25, right? One day contract, and I
can see how much money they made at the end. I could see their
return on investment. So here are people playing
with their own money, substantial amounts of their
own money, millions of them all over the world, doing
this very regularly. Sort of average maybe one a
day, that sort of transaction, right? A couple, maybe a couple days. And we can make a graph
of the social network. And that’s what this. This is 1.6 million people. These are the same
1.6 million people. And wherever there’s
a dot, this person decided to follow
this other person. And follow in eToro
has a different meaning than Facebook. Follow means I’m going
to take 10% of my money, and whatever that person does,
my 10% percent of the money will be invested
exactly the same way. So this is follow
with teeth, OK? And this is the
graph of following. So these are people
learning from each other, looking at each other
strategies, and trading. And you see some people
are going it alone. They read the newspaper,
they look at the wire, they browse the web,
then they trade. Other people are in this orgy
of social dialogue, right? All following each other. And, in fact, if
you look at it, you see that there are all
these loops in there. So, you know, I follow you,
you follow him, he follows me. Uh oh. [LAUGHS] And what happens in
this loop is that you get very few new ideas. It’s the same ideas
going around and around. And this is the
sort of thing that’s a precursor of an
economic bubble. That the question is, which of
these sort of social strategies gives greater return
on investment? Why are we a social species? The way that people
almost always analyze it is greater information. That these guys all read the
newspapers and everything. They have all the
information in the world, OK? These guys read the same
newspapers and everything, but they also look
at each other. What would you expect to happen? Well, you can write
down the equations here. And what you can
do is you can look at the propagation
of new strategies across this population. So when this guy comes up
with a new thing to do, how likely is it to propagate
throughout the social network? And in that way,
you can quantify the number and diversity of new
strategies any one person sees. These people will see
about no new strategies because they’re
all on their own. These people will see
very few new strategies because they’re all
listening to each other. It’s the same thing
around and around. And these people are
much more diverse. If we look at that
return on investment, we get a curve like this. So, again, this is a
mathematical function that has to do with the
number of new strategies. So the rate of
propagation of strategies through the medium,
the social network. If you look at the
number of new things, this is very low,
this is very low, this has many new
types of strategies. And this vertical axis
is return on investment. So this is, like, one of these
no BS types of majors, OK? Real people, their
own money, doing it on their own choice,
making money, or not. People who trade by
themselves are market neutral. You might expect
that on average. They hire the market. They lose a little bit of
money in trading costs. People who are in
these momentum echo chambers don’t do
very well either. And what isn’t shown
here is sometimes there are crashes
that blow them all up. So they actually do pretty
badly on the long term. But people in the middle
make 30% more money. So this is not something that
is in traditional economics. What we’re talking
about here is a blend of social strategies
for learning from other people, plus
individual information. It’s the peer to
peer interactions. And probably the reason that
we have a social species, this learning from
each other, is because it has this much
more efficient output. And there’s a big
literature about this. Don’t just believe me. This is a wonderful example
because I can quantify it. And every doc here
is all in the trades by millions of people
for a whole day. So this is, like, more
data than you know, right? And if I did just
one asset class, like dollars versus euros,
it wouldn’t have this spread that it does. It would be a nice band. So as you get more
diverse learning from your social environment,
your return on investment goes up until you begin
getting too many loops. And then it goes back down. Now I like this example because
I think this example applies to the government, it applies to
making decisions in companies. If you begin thinking
about it, we’re all living these
social networks, and what we’re trying to
do is make good decisions. Here, I’m showing you that
a mixture of social learning plus individual learning– I can
tell you a lot more about it, it’s in the book– gets
you better decisions. And not just better
decisions by one person, this is better decisions of
the entire 1.6 million people. Now, that’s a really
different thing. I should also mention
that one of the things we did with this platform is when
we discovered that they were in this echo chamber
state, that’s not good for the
platform or them, OK? Everyone’s going to lose money. So we looked at
the loop structure, and we figured out how
best was the optimal way to break it up was. And we gave coupons
to key people, small group of key people, that
would cause this echo chamber to break up in an
optimal manner. And that doubled the return
on investment of the people in the entire network. And that lasted for about
three days, four days, and they went back to
being stupid again. But [LAUGHS] that’s
their problem. We’ve done this
sort of repeatedly. We know it works. So you can actually
control the flow of ideas in a network
like this and improve the average function of
the people in the network. It’s a very different
way of thinking about things than the normal
way because you’re not concerned about individuals. You’re not concerned about their
education and their decision making. You’re concerned about
the pattern of learning and the performance
of an ensemble, rather than the individuals. So one of the other things
that this big data tells us is that this process can be
broken up into two pieces. And to illustrate that, I’ll
show this diagram that’s from Danny Kahneman’s
Nobel Prize lecture. He’s the father of
behavioral economics. And he makes the
point that people have two ways of thinking. There’s a slow way of thinking. [INAUDIBLE] probably
knows about thinking fast and slow, very popular. Slow way of thinking that’s the
serial reasoning that we do. And there’s this
fast way of thinking, which is associations. You take the
experience you had, you say, how is this situation
like my previous experiences? Maybe you interpolate
a little bit, and you make your
decision very, very fast. This is a very old mechanism,
100, 200 million years old. This is pretty much
unique to humans. Interestingly, this is the
much better mechanism by far if you have the right
set of experiences. If you don’t have the
right set of experiences, this is a disaster waiting
to happen because you’re going to associate the wrong
things with the wrong things, and follow right off the cliff. And when I look at the
learning that people have from each other in
these social networks, I see a qualitatively
different type of behavior. So when I look at
slow learning– so this is a learning
that people integrate into their conscious
representations. So the new song you
heard, the new fact, the new product that came out. People are very
promiscuous about this. It only takes one
exposure to integrate that into your ensemble of
things you know about. And this is a way
almost everything that you guys build is based on. Oh, we’re going to have
more information, right? But information is not behavior. It turns out that to get
behavior change, which is what I call idea flow,
you need something different. So this is an exploration. We are trying to find new
possibilities and new facts. But it’s relatively isolated
from behavior change. You could learn about
all sorts of things and never change your behavior. This is why it’s
hard to stop smoking, this is why it’s hard
to stop overeating, why all sorts of
things are hard is that our habits, our actual
behaviors, that reside here are largely independent of this. Now, there’s some leakage. If you concentrate real hard,
some early adopters, yes, it does happen. But as I showed with
the voting experiment, the transfer from here
to here is very weak. On the other hand,
what is the case is that if you see multiple
people experimenting with the same idea, people whom
you have strong relationships with, then you will
with very high certainty tend to adopt that behavior, OK? So what you’re doing is
this social learning. If I see for some
of my peers that doing this results
in a better outcome, then without even thinking about
it, I’ll begin to adopt that. If I hear about it, you know,
through email or on the web or something, it’s
very unlikely. We have a database
of all of– I can’t say the name of the
company, but a competitor bought it for $1 billion. The social network for
the inside companies. So we have the deployment for
over 1,000 companies in there, using that social network. And what we find can be summed
up in an interesting statistic. If you get invitations to
join this intracompany social network from as many as 12
people in a half an hour, you’re still
unlikely to sign up, unless those people
are people you already have strong relationships with. If they’re people you know
face to face or people you work with regularly, then
as few as three invitations makes it almost certain
that you’ll sign up. So that’s just like
the voting thing, it’s what I’m
talking about here. Behavior change, which is
what you usually care about, has to do with this social
reinforcement mechanism that I call engagement. It’s community vetting
of ideas and behaviors that results in the
adoption of a new behavior. It’s not the broadcast, in
fact, that we often think about. So let me show you
an example of this. So this is data
from a German bank. It has five departments,
managers, development, sales, support, customer
service is the last one. And this is all the
email, and the red stuff is all the face to face. We get this off of little
badges we put on people. So you probably can
track this stuff, but you’ve never tracked
the face to face stuff. Nobody does. And what we find is that
the sort of punch line is that the email,
the pattern of email, has very little to
do with productivity or creative output. But the pattern of rich
channels of communication has a huge amount. So I’ll show you a slightly
distracting thing first, and then I’ll tell you
the real punchline here. So these guys are going
to do an ad campaign. They’re starting now where the
boss sends out lots of email to have lots of meetings
to figure out how to do it. During that time, nobody
talks to customer service. They deploy the thing, it’s a
disaster, and as a consequence, they deploy it now. And then they have all day
meetings with customer service to figure out how to fix it, OK? So the real punchline,
because we’ve done some dozens
of companies now, is that you can see the pattern
of rich channel communication, and that predicts typically
30%, and sometimes 40% of the variation in
productivity of work groups. 30% to 40% is bigger by far
than anything that you look at. I mean, you’d have to like
kill the people to get that big of an effect. And the mathematical
formulation of it is basically a probability
that if I talk to you and I talk to you, what’s
the likelihood that you two also talk to each other? It’s those loops. And it’s this learning
from each other, keeping people in the loop,
nice little mathematical relationship, that
predicts this productivity. And there’s another
thing, so that’s that engagement I
was talking about. There’s another
[INAUDIBLE] exploration, and that’s the stuff
that your boss tells you is not in your
job description. That’s going to talk
to the people in sales, or the janitors, or the
people at the other company. Just picking up new ideas,
new facts, new observations, and bringing them back
to your work group to bang against each
other and see if they make sense to do that social
learning process, OK? I wrote a paper for
Harvard Business Review that lays this out. It’s called the “New Science
of Building Great Teams.” And it won Paper of the
Year award, which is nice. But it also won
Paper of the Year from the Academy of Management,
which is the academic side. And that’s the first
time, I believe, that Harvard Business Review
and the academic business guys have ever agreed. So maybe it’s worth
taking a look at. Anyway, so that’s companies. Let’s look at the real world. So this is a company I
helped found in 2006. Sold to AT&T’s
mobile advertising that’s been out recently. People moving around
in San Francisco. Big dots are the
most popular places. Maybe some of you guys
have seen this before. I like it. I show it often. Looks like a nicely
mixed city, but actually, if you analyze it, if you
cluster people by their paths and by their exposure to
each other, what you find is you find that it’s
a very segregated city. There’s these groups of
people that hardly ever are exposed to each other. And then there’s other
groups within the group, they don’t know each other,
but they go to the same places, they see the same things, and
they have the same habits. So in other words, they
have very strong engagement within the groups and they learn
habits of behavior from that. Now, sometimes that’s good. So, for instance, sometimes
it’s sort of trivial. It’s like you might discover
that one group here, you get a fad for red dresses. No particular reason. It’s just what people
in this group do, OK? In another group,
though, what you find is you find that they
have a different attitude about paying back credit
cards than maybe you do. And so they don’t have
such good risk scores. Again, it’s not anything
that they thought about. It’s just what people
in their group do. They learn from each other. [INAUDIBLE] George just, like,
threw it away, got a new one. Nobody came after him. It’s the smart
thing to do, right? And then the other
thing that you find, which is very important,
is chronic diseases vary by group
because of behavior. Chronic diseases are mostly
a function of your behavior. You eat too much,
you drink too much, you don’t exercise
enough, all those things. And a lot of other
things we don’t know. But we don’t know why
that particular group is susceptible to
diabetes, but we know they’re very much more
susceptible than other people. It seems to be that they learned
bad habits from each other, OK? So I’m going to give a
TED talk in a little bit, so I thought I’d put this in. So what TED does– and his
idea’s worth spreading, right? Make his wonderful videos
blast him out to everybody. And what that’s doing
is it’s increasing people’s exploration. You got a million views of
this little movie, da da da da, right? But it doesn’t change behavior. That’s my prediction. That’s what I see
in all of this data. What changes behavior
is all the peer to peer communication
that comes afterwards, where you say, what
do you think of that? Another guy says, oh,
yeah, that’s awesome. And then the third
guy says, well, OK. And you get this validation
among your peer group, and that’s what leads you to
actually change your behavior. So if you like what I’m saying,
if you think it’s interesting, talk to your peer
group, [LAUGHS] right? Maybe it’ll change
your behavior. So you can actually do
something interesting today, which is you can use
these ideas to map stuff in the entire cities. So this is mixing of different
communities in Mexico. And I should start
it over again. So the red stuff is where
the most mixing happens, and the yellow stuff is where
very little mixing happens. And if it’s blank, we
don’t have the data. So you can see on Sundays,
there’s very little mixing. People stay home. Monday, a lot of
people come out. So according to the
things that I’ve told you, it’s this mixing
between communities that’s the source of a
lot of the innovation and creative output
in a community. It’s the banging
together of ideas that causes innovation and
better social outcomes. And you can now do this–
and this is just cell tower activity. This is not any
personal data at all. This is just at the
level of cell towers. You say, well, which
parts of the community do the people at this
cell tower come from, OK? You don’t know the
individual people. But you can use this
in interesting ways. If I use that same
method– this is something we did
in the Ivory Coast. I helped convince
the carrier, Orange, release all of their cell
tower data for the Ivory Coast. Ivory Coast is a
very poor country that also had a civil war. So the government can’t
go in the northern half of the country. What they do is they
have poverty statistics for the lower half. And using this method,
you can fit the statistics for poverty in the lower
half and extrapolate them to the upper half. And the poverty that your
measuring is interesting. So this has to do
with two factors. It’s this exploration outside
the community and engagement within the community, OK? And so this MPI is a
multi-factor thing. And it’s a combination
of poverty, but also life expectancy,
crime, and infant mortality because they all co-vary
with one another. So that’s an example. So you saw it with Mexico City. You see it here. This is something
a former student of mine, Nathan [INAUDIBLE] did. He took all the data
from councils in the UK. So there are
neighborhoods that are administrative units in the UK. And he looked at their
socioeconomic outcome index, which is, again, poverty,
crime, infant mortality stuff, and compared it to the
land line phone records, and measured two things,
which are very similar to what I call exploration
and engagement and generated this graph. So when you get a community
that doesn’t talk to itself and doesn’t talk much outside
of itself, all the babies die. Well, not all the babies. A lot of babies die. When you get a community where
they’re richly integrating into the rest of society and
they talk among themselves, very few babies die. And, of course, this
is a richer community, that’s a poorer
community, that has less crime, that has more crime. So we’ve seen this
now in several places. We’ve seen it in England,
we’ve seen it in Ivory Coast, we’ve seen it in Cambridge. And you can begin
doing things with this. So for instance,
you can use this to be able to predict
GDP in cities. So we took the GDP for 150
cities in Europe and 150 cities in the US, and we measured
the amount of banging together of ideas that you got in rich
channels of communication, face to face primarily. And we did that by using
things like Foursquare to ask, how often do
people come together from different communities
and at what distance? You can print a nice
mathematical function. It’s the same form in
Europe as it is in the US. It varies by the density of
the city and the transportation infrastructure. So if it’s a very dense
setting with a really good transportation
infrastructure, you run into a lot of
different people, there’s a lot of ideas
banging together, and you get a lot of innovation. And so what this is
showing is a measure of this face to face
engaging and exploration. And this is GDP per head. I put kilometer, I think
actually that one is. And you can see that it
accounts for around 90% of the variance, which in social
science, is like a, you know, law from above or
something like that. In other words, if you
tell me the density of the city and the
transportation infrastructure, and actually just need to tell
me the average commute time, I can tell you the
GDP almost perfectly. Similarly, if I tell you that
mobility or the call pattern in a neighborhood,
I can tell you the GDP, the number of infant
mortality, and the crime rate. Again, r squared of
about 0.85 is like a law. It’s amazing, OK? And that opens up
the possibility of doing cool things. You could, for instance,
change the infrastructure to make more ideas
bang together, right? You could make it so
it’s easy to get around than a place like New
York or San Francisco, as opposed to an incredible
pain in the butt, all right? And so, you know, we,
at MIT, have done things like this shared
car, shared scooter. I’m on the advisory
board for Nissan and helping them
build what we hope to be the first autonomous
vehicle– commercial autonomous vehicle, all right? Because we actually built one. I helped to build one
almost 20 years ago, but it was never
deployed commercially. And so now we’re going– the
CEO says we’re going to do it. So that’s [INAUDIBLE]. But the thing I think is
most– so let me back up. So this is a tool we
built at MIT– this is Ken Larson and
Ryan Chin primarily– which allows you
to simulate things. So this is actually the
area around MIT– Media Lab. That’s the Media Lab there. So they’re built out of LEGOs. And they have a
laser range finder, which scans the 3D thing. And then you do
some computation, and you project back
what ought to happen. So, for instance,
you can project back how the wind will go,
or traffic patterns. But you could also
project back things like anticipated
creative output. How many different ideas
are going to bang together? Are all the communities
little silos, or are they actually mixing? So this is something
I’m sure people who plan these buildings
talked about all the time, but you never measure it. And yet, all the
data say, that’s the source of real
creative output. It’s also the source of getting
everybody on the same page. You have to mix
those two things. So where are we getting to build
the interactive tools to do that. So another way which probably
resonates more with this group is this, which is trading ideas
with each other electronically, rather than face to face. Rich channels, like face
to face, are important. But you can supplement them, you
can extend them in various ways by better data sharing. And to a certain
degree, that’s why people call personal data
the new oil of the internet because all those personal
experiences are not only good for
learning peer to peer, they’re also good for
advertising and lots of other things. And as I think I’ve
shown you, it also is something that
worries a lot of people and could be used
in very bad ways. And so about five years
ago, I helped start group at Davos, which included
people like the Chairman of the Federal Trade Commission,
Justice Commissioner of the EU, people from the Politburo
in China, CEOs of things like Vodafone,
Microsoft, et cetera, to be able to talk
about these problems. And so what we were looking
for was a win-win-win solution, where citizens would
feel protected and be able to see more
value from their data, from trading information,
where companies would be able to make money and
where governments would be able to provide
public goods, be able to make a more resilient
society, cyber resistance, and so forth. And the nice thing– this
is the sort of diagram they do at Davos. While people speak, some
incredibly talented artist draws the discussion. You can’t really interpret
it, but it’s just amazing to see them do it. But the bottom line was is
that the ideas that came out of that, which are now enshrined
in the Consumer Privacy Bill of Rights in this country
and the privacy directives in the EU, and are being
considered in China, have to do with changing
the way we treat data, personal data, the
sort of personal stories, in a very fundamental way. And that’s to put
much more control in the hands of individual
through notification that people are collecting data
about you, informed consent. That means you show them what
the data is, you describe to them what the value they’re
going to get from sharing is, and they opt in to
that particular use. And then they can opt out
if they don’t like it. Auditing, to make sure
that you did what you said you were going to do. And that’s the retraction
I’ve always talked about. So that’s where
things are going. In anticipation of that, I got
DARPA to give us a lot of money to build an infrastructure
like this because they wouldn’t do much, except
small experiments. But in this country, I
got [INAUDIBLE] in Italy to be a living lab, to
try to live in the future, by giving citizens more
control over their data, using this infrastructure
in conjunction with Telecom Italia, Telefonica,
the local government, and so on. And the experiment
is to be able to say, if people have a repository,
a copy of all the data that’s about them, and that
risk reward ratio is different because they can
opt into things, and they know what they’re
opting into, they can opt out, they could audit it. You’ve changed the
risk reward ratio. Will the sharing go up? Will companies make more money? Will people do more sharing? Will you get greater
innovation through that sharing of ideas and information? Will government be
able to do a better job at providing public goods? And so we’ve deployed this
for the last year and a half. It has many sort of
technical elements. Some of the ones that
I’m proud about, as we’ve talked MITRE, which is
a government defense contractor, to
releasing something called OpenID Connect
as an open source. Identity mechanism, that’s
really quite state of the art. It’s now being supported
by MIT Kerberos Consortium. Their intent is to make
it a basic element, the internet security. If you are interested
in this stuff, you should take a look at it. And also trust
network technologies, whereby it’s a combination
of long computer code that give people more
control and audit ability over personal data. An example of this, as you
might be familiar with, is the SWIFT network, which
is for interbank transfer. So the SWIFT network handles
$3 trillion of money a day. And as far as we know,
it’s never been hacked, despite operating I think
at 164 different countries and with a lot of dodgy banks. And it has to do with a
marriage between the sharing protocols and the
legal contract, which is a consumer contract. It’s not special regulations. It’s just contract law with
close match between that. So the communications
begin with the offer of a contract, the signing of
a contract, it’s auditable, there’s joint liability so
everybody’s watching out for everybody else,
and it seems to work. The Visa network is a
similar sort of thing. Some of this is used in
the federal government for medical data sharing. And one of the things that’s
particular about our solution– and I know this is too quick,
but I’m hoping that there’s people that are
interested in this– is that like the
SWIFT network, we don’t provide sharing of
data, except in extreme. So generally, there’s
no reason to share data. What you want to do is you want
to share answers to questions, and you want to share stories. You want to say, oh, yes,
I’m in San Francisco, not, I’m at this lat long. Or I’m in San Francisco today,
not lat long at 3:15 PM. So by providing
the most, sort of, general story
elements possible, you get the commercial
opportunities, the public good opportunities,
with much, much less exposure in terms of raw data
sharing and unintended uses. It’s not a perfect answer,
but it’s a good answer. So we’ve deployed
this in [INAUDIBLE], we’re deploying it in MIT. We could do it other places. I’m going to tell
you just a little bit about the basic
architecture, but then I see I’m running too long, right? I am, yeah. So I won’t tell you too much. There is an architecture. The cool thing about this is
once people have personal data stories, you can spin up
applications trivially. There’s no collecting the
data to get bootstrapped. The data is already there. So we’re doing this with
Mass General Hospital. Were doing it with lots of
other people where, you know, when you opt in, day one, it
has a whole history of you because you’ve been
storing all that data. Anyway, so taking
too long, sorry. Big data, better life. That’s what we’re about. The book describes
all this in detail. Thank you. [CLAPPING] AUDIENCE: So you show these
beautiful correlations between some outcome for society
and the number of interactions, right? And I’m wondering, is
there strong evidence of causality there? But for instance, if we just
tweak how much interaction is going on in a given society,
would that, in and of itself, escalate it? SANDY PENTLAND: So we know that
it’s causal in small groups and in groups of 100
people because we’ve done interventions. We don’t know that it’s
causal in big groups. But you can look
at, for instance, the architecture of lots and
lots of different cities, and it makes a certain
amount of sense. You see the same pattern. Unless, of course,
[INAUDIBLE] fits so well. Basically, what
you’re talking about is sort of the Jane
Jacob solution, which is small communities with
very good transportation infrastructure between them. A small community where
you could walk around gives you the strong engagement
and culture and social support. And the very good
transportation infrastructure lets communities
interact with each other. That’s the way the design
shakes out, basically. So we think it’s causal. We don’t know. We’re trying to work with cities
to show that it is, all right? AUDIENCE: So I work in privacy,
and I liked your remarks on modification and
firm consent, auditing, and the rest. What do you think about actual
automatic expiration of such– SANDY PENTLAND: I think
that’s a great idea– AUDIENCE: Would it increase
the value over a long time, or would it have
a negative effect to the value of society
over a long time? SANDY PENTLAND: I think
it’s one of these things that you have to
experiment with, but I would expect
it would increase it. I mean, you know,
the fundamental thing is risk reward, right? You want to know
what’s going on, so you don’t want
to be spied on. You want to have
control over it. And you want to be able to
share an expectation of a return without a lot of downside. So expiration means that
it’s less likely to spread. Auditing means that it’s
less likely to get stolen. It’s still will sometimes,
but what it is really is it’s a framework that’s a
lot like our financial network, our banking. You know, you have
these strings of numbers that are the amount of
money that you have, and you give it to this
thing called a bank. And then you could look at
it, and the federal government comes and audits
them, and you could take it out if
you don’t like it. And so we’re talking about
that with personal data, where I put it in a bank, and
I say, I will give it to these people in return
for these sorts of things. And if I don’t like what you
do with it, I’ll take it back. And then the next objection is,
wasn’t this too complicated? And yes, it is too complicated. That’s why we have things
like mutual funds and 401ks, and junk like that is because
it’s just way too complicated for a human. But you’d have the same sort
of thing with personal data. The AARP would have a standard
way for elderly people to share data that
is deemed safe. AUDIENCE: Specifically
what I mean is I opt in into
something that that opt in is not treated
as indefinite. SANDY PENTLAND: It
should be absolutely. The opt in should be
part of the contract that it expires, right? AUDIENCE: Yes. SANDY PENTLAND: Yeah. AUDIENCE: Thank you. AUDIENCE: I had a quick question
about the trying to break up the investment trading circles. Is there a reason you chose
an individual incentive to try to break up
the social networks, or was that just the easiest
way to try to break those up? SANDY PENTLAND: So we tried
several different things. One was just giving–
first of all, they’re not
individual incentives. What it is, is saying,
here’s a coupon if you follow that person. So it’s saying, build a
link in the social graph. It’s not like you think about
it more or something like that. So we tried several things. One was to give
people random coupons. So just pay attention to a
random person that did nothing. We gave people coupons
to pay attention to the highest
performing people. That did something. That returns by about 2%. And then we took people that
were targeted to break up the feedback loops,
and that was the thing that had this much
larger effect, OK? But notice that it
wasn’t an incentive for any particular person
to do well, all right? Some of the people we gave
coupons did less well, OK? But I don’t really care. What it did is it
broke up the loops, and that the average
performance went up higher. [CLAPPING]

21 thoughts on “Sandy Pentland: “Social Physics: How Good Ideas Spread” | Talks at Google”

  1. We dont need high science to understand that there is no free will and that nothing important happens in any of our heads. It just needs some good will to go against heavy programming we were all subjected to. Everything that is in each of our heads came and is coming from outside and is part of our multicellular organism (called humanity).
    Ideas of independence and free will are lies used as excuse to grab and keep power against powerless.

  2. Sandy's work is fantastic. I am also reading his book "social physics". That book is one of the best I have read. 

  3. behavioral modification? this is not  new thing. tyrants and slave masters have strived to do this for centuries with their chattel and livestock? Why do exactly the same thing and expect different results? Clockwork orange all over again?

  4. We need new social experiments which offer incentives in a subconscious and spiritual level rather than just on a monetary one. I have written a book which focuses on reorganizing our social relations and eliminating top down pressures to create a more just and socially mobile society. The book is called THE PRINCIPLE PHILOSOPHY OF PURPOSE and its on AMAZON. It's time for a new social revolution!

  5. The potential for misuse of this work needs to be discussed. Think of how many organizations there are that would like to influence your thoughts. They could do it without announcing themselves, but instead work through agents with nice personalities who could maintain engaging web personas. These "leaders" need not reveal that they are working to change your thought to align with some agenda or other.
    Imagine large numbers of paid shills, working hours every day, using these techniques to sway public opinion without ever showing their true colors.
    Despite the likeability of this researcher, and his seeming altruistic optimism, what he suggests doing here is immoral and unethical, no matter what the motive.

  6. Superb communicator and wise too: superb attention filtering

    Admitting to disaster 4:40 which is good, but that the india to-down experiment was even attempted is curious, given his level of experience and awareness.

    Tracking *bodies* 7:00 to attempt to predict behaviour by modelling habitual behaviour. 9:00 we are not 'individual', more like parts of a social fabric. Modelling peer-to-peer behaviour 10:30 sourcing in Adam Smith. 12:30 individual incentives don't work. 14:00 relational incentives are x4 to x20 more powerful. 15:00 rewarding neighbour's activity, sticky too 15:50 continued when money was taken away. 21:00 toro? following with teeth, ie following financial behaviour [ie money as vector, flocking]. 24:00 between isolation and echo-chamber, sweet spot 'in between', applicable across government, academia, etc. 26:30 performance of the ensemble not individual [my experience in classes]. 34:00 not broadcasting, even TED talks, but p2p engagement changes behaviour.

    Basically, a lot of social graph analysis. Meta-analysis. [Eco^2 is direct.]
    Sandy is applying slow thinking to fast processes. [My experience is limited to fast processes in a live class, extrapolated to global scale.]
    Sandy is still using broadcast media, in auditorium, or via youtube/TEDtalks. [Eco^2 starts with p2p, eg @Colin Kilburn passing this video on to me via the app.]
    19,869 views since March 7 2014, with only 137 likes at time of viewing. [Good indicator of… lack of interaction.]

  7. How the f$% can I learn everything he's saying. Teach me please!!!…yes I know about his book and papers. Not the same as implementing everything of course

  8. So longer commute time equals better GDP? Because people have more time on the way to talk to each other? I didn't really get the part about transportation and how it's related to GDP. Anyone please help me.

  9. Ha! The graphic on the right @32:15 is my network neighborhood on an early online social network — Ryze.

  10. If you watch the same video 100 times, then it might change your behaviour. i.e: If you exposed to the same information ALOT then it does influence you. But I do agree, we do tend to run our ideas by our peers before adopting them.

  11. AI pointed at humans is ass backwards to use ways to predict actions does not pinpoint intentions. Use AI in conjunction with the human heart to search for truth that is when you open statgats

  12. Incenting a network is more efficient than incenting an individual. Yeah, no shit sherlock; it also pisses people off and builds a corrosive social fabric in plenty of contexts.

Leave a Reply

Your email address will not be published. Required fields are marked *