0:00:00.738,0:00:02.735
If you remember that first decade of the web,

0:00:02.735,0:00:04.990
it was really a static place.

0:00:04.990,0:00:07.235
You could go online, you could look at pages,

0:00:07.235,0:00:09.748
and they were put up either by organizations

0:00:09.748,0:00:11.269
who had teams to do it

0:00:11.269,0:00:13.498
or by individuals who were really tech-savvy

0:00:13.498,0:00:15.235
for the time.

0:00:15.235,0:00:16.810
And with the rise of social media

0:00:16.810,0:00:19.209
and social networks in the early 2000s,

0:00:19.209,0:00:21.358
the web was completely changed

0:00:21.358,0:00:24.966
to a place where now the vast majority of content

0:00:24.966,0:00:28.278
we interact with is put up by average users,

0:00:28.278,0:00:30.975
either in YouTube videos or blog posts

0:00:30.975,0:00:34.290
or product reviews or social media postings.

0:00:34.290,0:00:36.637
And it's also become a much more interactive place,

0:00:36.637,0:00:39.274
where people are interacting with others,

0:00:39.274,0:00:40.970
they're commenting, they're sharing,

0:00:40.970,0:00:42.584
they're not just reading.

0:00:42.584,0:00:44.450
So Facebook is not the only place you can do this,

0:00:44.450,0:00:45.548
but it's the biggest,

0:00:45.548,0:00:47.332
and it serves to illustrate the numbers.

0:00:47.332,0:00:50.809
Facebook has 1.2 billion users per month.

0:00:50.809,0:00:52.739
So half the Earth's Internet population

0:00:52.739,0:00:54.392
is using Facebook.

0:00:54.392,0:00:56.324
They are a site, along with others,

0:00:56.324,0:00:59.543
that has allowed people to create an online persona

0:00:59.543,0:01:01.325
with very little technical skill,

0:01:01.325,0:01:03.801
and people responded by putting huge amounts

0:01:03.801,0:01:05.784
of personal data online.

0:01:05.784,0:01:08.327
So the result is that we have behavioral,

0:01:08.327,0:01:10.313
preference, demographic data

0:01:10.313,0:01:12.414
for hundreds of millions of people,

0:01:12.414,0:01:14.440
which is unprecedented in history.

0:01:14.440,0:01:17.000
And as a computer scientist, [br]what this means is that

0:01:17.000,0:01:18.664
I've been able to build models

0:01:18.664,0:01:20.986
that can predict all sorts of hidden attributes

0:01:20.986,0:01:23.270
for all of you that you don't even know

0:01:23.270,0:01:25.472
you're sharing information about.

0:01:25.472,0:01:27.854
As scientists, we use that to help

0:01:27.854,0:01:29.968
the way people interact online,

0:01:29.968,0:01:32.467
but there's less altruistic applications,

0:01:32.467,0:01:34.848
and there's a problem in that users don't really

0:01:34.848,0:01:37.318
understand these techniques and how they work,

0:01:37.318,0:01:40.446
and even if they did, they don't[br]have a lot of control over it.

0:01:40.446,0:01:41.936
So what I want to talk to you about today

0:01:41.936,0:01:44.638
is some of these things that we're able to do,

0:01:44.638,0:01:47.401
and then give us some ideas[br]of how we might go forward

0:01:47.401,0:01:50.170
to move some control back into the hands of users.

0:01:50.170,0:01:51.756
So this is Target, the company.

0:01:51.756,0:01:53.080
I didn't just put that logo

0:01:53.080,0:01:55.250
on this poor, pregnant woman's belly.

0:01:55.250,0:01:57.090
You may have seen this anecdote that was printed

0:01:57.090,0:01:59.151
in Forbes magazine where Target

0:01:59.151,0:02:01.512
sent a flyer to this 15-year-old girl

0:02:01.512,0:02:03.222
with advertisements and coupons

0:02:03.222,0:02:05.776
for baby bottles and diapers and cribs

0:02:05.776,0:02:07.460
two weeks before she told her parents

0:02:07.460,0:02:09.324
that she was pregnant.

0:02:09.324,0:02:12.028
Yeah, the dad was really upset.

0:02:12.028,0:02:13.744
He said, "How did Target figure out

0:02:13.744,0:02:15.568
that this high school girl was pregnant

0:02:15.568,0:02:17.528
before she told her parents?"

0:02:17.528,0:02:20.149
It turns out that they have the purchase history

0:02:20.149,0:02:22.450
for hundreds of thousands of customers

0:02:22.450,0:02:25.180
and they compute what they [br]call a pregnancy score,

0:02:25.180,0:02:27.512
which is not just whether or [br]not a woman's pregnant,

0:02:27.512,0:02:29.242
but what her due date is.

0:02:29.242,0:02:30.546
And they compute that

0:02:30.546,0:02:32.314
not by looking at the obvious things,

0:02:32.314,0:02:34.826
like, she's buying a crib or baby clothes,

0:02:34.826,0:02:37.769
but things like, she bought more vitamins

0:02:37.769,0:02:39.486
than she normally had,

0:02:39.486,0:02:40.950
or she bought a handbag

0:02:40.950,0:02:42.661
that's big enough to hold diapers.

0:02:42.661,0:02:44.571
And by themselves, those purchases don't seem

0:02:44.571,0:02:47.040
like they might reveal a lot,

0:02:47.040,0:02:49.018
but it's a pattern of behavior that,

0:02:49.018,0:02:52.135
when you take it in the context [br]of thousands of other people,

0:02:52.135,0:02:54.892
starts to actually reveal some insights.

0:02:54.892,0:02:56.685
So that's the kind of thing that we do

0:02:56.685,0:02:59.252
when we're predicting stuff[br]about you on social media.

0:02:59.252,0:03:02.048
We're looking for little[br]patterns of behavior that,

0:03:02.048,0:03:04.730
when you detect them among millions of people,

0:03:04.730,0:03:07.436
lets us find out all kinds of things.

0:03:07.436,0:03:09.183
So in my lab and with colleagues,

0:03:09.183,0:03:10.960
we've developed mechanisms where we can

0:03:10.960,0:03:12.520
quite accurately predict things

0:03:12.520,0:03:14.245
like your political preference,

0:03:14.245,0:03:17.997
your personality score, gender, sexual orientation,

0:03:17.997,0:03:20.870
religion, age, intelligence,

0:03:20.870,0:03:22.264
along with things like

0:03:22.264,0:03:24.201
how much you trust the people you know

0:03:24.201,0:03:26.005
and how strong those relationships are.

0:03:26.005,0:03:27.790
We can do all of this really well.

0:03:27.790,0:03:29.987
And again, it doesn't come from what you might

0:03:29.987,0:03:32.089
think of as obvious information.

0:03:32.089,0:03:34.370
So my favorite example is from this study

0:03:34.370,0:03:35.610
that was published this year

0:03:35.610,0:03:37.405
in the Proceedings of the National Academies.

0:03:37.405,0:03:38.690
If you Google this, you'll find it.

0:03:38.690,0:03:40.562
It's four pages, easy to read.

0:03:40.562,0:03:43.565
And they looked at just people's Facebook likes,

0:03:43.565,0:03:45.485
so just the things you like on Facebook,

0:03:45.485,0:03:47.623
and used that to predict all these attributes,

0:03:47.623,0:03:49.268
along with some other ones.

0:03:49.268,0:03:52.229
And in their paper they listed the five likes

0:03:52.229,0:03:55.016
that were most indicative of high intelligence.

0:03:55.016,0:03:57.340
And among those was liking a page

0:03:57.340,0:03:59.245
for curly fries. (Laughter)

0:03:59.245,0:04:01.338
Curly fries are delicious,

0:04:01.338,0:04:03.868
but liking them does not necessarily mean

0:04:03.868,0:04:05.948
that you're smarter than the average person.

0:04:05.948,0:04:09.155
So how is it that one of the strongest indicators

0:04:09.155,0:04:10.725
of your intelligence

0:04:10.725,0:04:12.172
is liking this page

0:04:12.172,0:04:14.424
when the content is totally irrelevant

0:04:14.424,0:04:16.951
to the attribute that's being predicted?

0:04:16.951,0:04:18.535
And it turns out that we have to look at

0:04:18.535,0:04:20.153
a whole bunch of underlying theories

0:04:20.153,0:04:22.722
to see why we're able to do this.

0:04:22.722,0:04:25.635
One of them is a sociological[br]theory called homophily,

0:04:25.635,0:04:28.727
which basically says people are[br]friends with people like them.

0:04:28.727,0:04:30.741
So if you're smart, you tend to[br]be friends with smart people,

0:04:30.741,0:04:33.371
and if you're young, you tend[br]to be friends with young people,

0:04:33.371,0:04:34.998
and this is well established

0:04:34.998,0:04:36.743
for hundreds of years.

0:04:36.743,0:04:37.975
We also know a lot

0:04:37.975,0:04:40.525
about how information spreads through networks.

0:04:40.525,0:04:42.279
It turns out things like viral videos

0:04:42.279,0:04:44.685
or Facebook likes or other information

0:04:44.685,0:04:46.573
spreads in exactly the same way

0:04:46.573,0:04:49.027
that diseases spread through social networks.

0:04:49.027,0:04:50.818
So this is something we've studied for a long time.

0:04:50.818,0:04:52.394
We have good models of it.

0:04:52.394,0:04:54.551
And so you can put those things together

0:04:54.551,0:04:57.639
and start seeing why things like this happen.

0:04:57.639,0:04:59.453
So if I were to give you a hypothesis,

0:04:59.453,0:05:02.680
it would be that a smart guy started this page,

0:05:02.680,0:05:04.619
or maybe one of the first people who liked it

0:05:04.619,0:05:06.355
would have scored high on that test.

0:05:06.355,0:05:08.643
And they liked it, and their friends saw it,

0:05:08.643,0:05:11.765
and by homophily, we know that[br]he probably had smart friends,

0:05:11.765,0:05:14.821
and so it spread to them, [br]and some of them liked it,

0:05:14.821,0:05:16.010
and they had smart friends,

0:05:16.010,0:05:16.817
and so it spread to them,

0:05:16.817,0:05:18.790
and so it propagated through the network

0:05:18.790,0:05:21.359
to a host of smart people,

0:05:21.359,0:05:23.415
so that by the end, the action

0:05:23.415,0:05:25.959
of liking the curly fries page

0:05:25.959,0:05:27.574
is indicative of high intelligence,

0:05:27.574,0:05:29.377
not because of the content,

0:05:29.377,0:05:31.899
but because the actual action of liking

0:05:31.899,0:05:33.799
reflects back the common attributes

0:05:33.799,0:05:36.267
of other people who have done it.

0:05:36.267,0:05:39.164
So this is pretty complicated stuff, right?

0:05:39.164,0:05:41.363
It's a hard thing to sit down and explain

0:05:41.363,0:05:44.211
to an average user, and even if you do,

0:05:44.211,0:05:46.399
what can the average user do about it?

0:05:46.399,0:05:48.447
How do you know that [br]you've liked something

0:05:48.447,0:05:49.939
that indicates a trait for you

0:05:49.939,0:05:53.484
that's totally irrelevant to the[br]content of what you've liked?

0:05:53.484,0:05:56.030
There's a lot of power that users don't have

0:05:56.030,0:05:58.260
to control how this data is used.

0:05:58.260,0:06:01.372
And I see that as a real [br]problem going forward.

0:06:01.372,0:06:03.349
So I think there's a couple paths

0:06:03.349,0:06:04.350
that we want to look at

0:06:04.350,0:06:06.260
if we want to give users some control

0:06:06.260,0:06:08.000
over how this data is used,

0:06:08.000,0:06:09.940
because it's not always going to be used

0:06:09.940,0:06:11.321
for their benefit.

0:06:11.321,0:06:12.743
An example I often give is that,

0:06:12.743,0:06:14.389
if I ever get bored being a professor,

0:06:14.389,0:06:16.042
I'm going to go start a company

0:06:16.042,0:06:17.496
that predicts all of these attributes

0:06:17.496,0:06:19.098
and things like how well you work in teams

0:06:19.098,0:06:21.769
and if you're a drug user, if you're an alcoholic.

0:06:21.769,0:06:23.209
We know how to predict all that.

0:06:23.209,0:06:24.970
And I'm going to sell reports

0:06:24.970,0:06:27.070
to H.R. companies and big businesses

0:06:27.070,0:06:29.343
that want to hire you.

0:06:29.343,0:06:30.520
We totally can do that now.

0:06:30.520,0:06:32.308
I could start that business tomorrow,

0:06:32.308,0:06:34.360
and you would have[br]absolutely no control

0:06:34.360,0:06:36.498
over me using your data like that.

0:06:36.498,0:06:38.790
That seems to me to be a problem.

0:06:38.790,0:06:40.700
So one of the paths we can go down

0:06:40.700,0:06:42.732
is the policy and law path.

0:06:42.732,0:06:45.778
And in some respects, I think[br]that that would be most effective,

0:06:45.778,0:06:48.534
but the problem is we'd[br]actually have to do it.

0:06:48.534,0:06:51.314
Observing our political process in action

0:06:51.314,0:06:53.693
makes me think it's highly unlikely

0:06:53.693,0:06:55.290
that we're going to get a bunch of representatives

0:06:55.290,0:06:57.276
to sit down, learn about this,

0:06:57.276,0:06:59.382
and then enact sweeping changes

0:06:59.382,0:07:01.539
to intellectual property law in the U.S.

0:07:01.539,0:07:04.000
so users control their data.

0:07:04.000,0:07:05.304
We could go the policy route,

0:07:05.304,0:07:06.783
where social media companies say,

0:07:06.783,0:07:08.185
you know what? You own your data.

0:07:08.185,0:07:10.674
You have total control over how it's used.

0:07:10.674,0:07:12.522
The problem is that the revenue models

0:07:12.522,0:07:14.246
for most social media companies

0:07:14.246,0:07:18.277
rely on sharing or exploiting [br]users' data in some way.

0:07:18.277,0:07:20.110
It's sometimes said of Facebook that the users

0:07:20.110,0:07:22.638
aren't the customer, they're the product.

0:07:22.638,0:07:25.352
And so how do you get a company

0:07:25.352,0:07:27.910
to cede control of their main asset

0:07:27.910,0:07:29.159
back to the users?

0:07:29.159,0:07:30.860
It's possible, but I don't think it's something

0:07:30.860,0:07:33.180
that we're going to see change quickly.

0:07:33.180,0:07:34.680
So I think the other path

0:07:34.680,0:07:36.968
that we can go down that's[br]going to be more effective

0:07:36.968,0:07:38.476
is one of more science.

0:07:38.476,0:07:40.986
It's doing science that allowed us to develop

0:07:40.986,0:07:42.736
all these mechanisms for computing

0:07:42.736,0:07:44.788
this personal data in the first place.

0:07:44.788,0:07:46.894
And it's actually very similar research

0:07:46.894,0:07:48.332
that we'd have to do

0:07:48.332,0:07:50.718
if we want to develop mechanisms

0:07:50.718,0:07:52.139
that can say to a user,

0:07:52.139,0:07:54.368
"Here's the risk of that action you just took."

0:07:54.368,0:07:56.448
By liking that Facebook page,

0:07:56.448,0:07:58.983
or by sharing this piece of personal information,

0:07:58.983,0:08:00.485
you've now improved my ability

0:08:00.485,0:08:02.571
to predict whether or not you're using drugs

0:08:02.571,0:08:05.433
or whether or not you get[br]along well in the workplace.

0:08:05.433,0:08:07.281
And that, I think, can affect whether or not

0:08:07.281,0:08:08.791
people want to share something,

0:08:08.791,0:08:12.030
keep it private, or just keep it offline altogether.

0:08:12.030,0:08:13.593
We can also look at things like

0:08:13.593,0:08:16.321
allowing people to encrypt data that they upload,

0:08:16.321,0:08:18.176
so it's kind of invisible and worthless

0:08:18.176,0:08:19.607
to sites like Facebook

0:08:19.607,0:08:22.236
or third party services that access it,

0:08:22.236,0:08:25.483
but that select users who the person who posted it

0:08:25.483,0:08:28.153
want to see it have access to see it.

0:08:28.153,0:08:30.319
This is all super exciting research

0:08:30.319,0:08:31.939
from an intellectual perspective,

0:08:31.939,0:08:33.798
and so scientists are going to be willing to do it.

0:08:33.798,0:08:37.408
So that gives us an advantage over the law side.

0:08:37.408,0:08:39.133
One of the problems that people bring up

0:08:39.133,0:08:40.728
when I talk about this is, they say,

0:08:40.728,0:08:43.374
you know, if people start[br]keeping all this data private,

0:08:43.374,0:08:45.487
all those methods that you've been developing

0:08:45.487,0:08:48.140
to predict their traits are going to fail.

0:08:48.140,0:08:51.660
And I say, absolutely, and for me, that's success,

0:08:51.660,0:08:53.446
because as a scientist,

0:08:53.446,0:08:57.134
my goal is not to infer information about users,

0:08:57.134,0:08:59.901
it's to improve the way people interact online.

0:08:59.901,0:09:03.119
And sometimes that involves[br]inferring things about them,

0:09:03.119,0:09:06.141
but if users don't want me to use that data,

0:09:06.141,0:09:08.179
I think they should have the right to do that.

0:09:08.179,0:09:10.830
I want users to be informed and consenting

0:09:10.830,0:09:12.942
users of the tools that we develop.

0:09:12.942,0:09:15.894
And so I think encouraging this kind of science

0:09:15.894,0:09:17.240
and supporting researchers

0:09:17.240,0:09:20.263
who want to cede some of that control back to users

0:09:20.263,0:09:22.574
and away from the social media companies

0:09:22.574,0:09:25.245
means that going forward, as these tools evolve

0:09:25.245,0:09:26.721
and advance,

0:09:26.721,0:09:28.135
means that we're going to have an educated

0:09:28.135,0:09:29.829
and empowered user base,

0:09:29.829,0:09:30.929
and I think all of us can agree

0:09:30.929,0:09:33.493
that that's a pretty ideal way to go forward.

0:09:33.493,0:09:35.677
Thank you.

0:09:35.677,0:09:38.757
(Applause)