The curly fry conundrum: Why social media “likes” say more than you might think
-
0:01 - 0:03If you remember that first decade of the web,
-
0:03 - 0:05it was really a static place.
-
0:05 - 0:07You could go online, you could look at pages,
-
0:07 - 0:10and they were put up either by organizations
-
0:10 - 0:11who had teams to do it
-
0:11 - 0:13or by individuals who were really tech-savvy
-
0:13 - 0:15for the time.
-
0:15 - 0:17And with the rise of social media
-
0:17 - 0:19and social networks in the early 2000s,
-
0:19 - 0:21the web was completely changed
-
0:21 - 0:25to a place where now the vast majority of content
-
0:25 - 0:28we interact with is put up by average users,
-
0:28 - 0:31either in YouTube videos or blog posts
-
0:31 - 0:34or product reviews or social media postings.
-
0:34 - 0:37And it's also become a much more interactive place,
-
0:37 - 0:39where people are interacting with others,
-
0:39 - 0:41they're commenting, they're sharing,
-
0:41 - 0:43they're not just reading.
-
0:43 - 0:44So Facebook is not the only place you can do this,
-
0:44 - 0:46but it's the biggest,
-
0:46 - 0:47and it serves to illustrate the numbers.
-
0:47 - 0:51Facebook has 1.2 billion users per month.
-
0:51 - 0:53So half the Earth's Internet population
-
0:53 - 0:54is using Facebook.
-
0:54 - 0:56They are a site, along with others,
-
0:56 - 1:00that has allowed people to create an online persona
-
1:00 - 1:01with very little technical skill,
-
1:01 - 1:04and people responded by putting huge amounts
-
1:04 - 1:06of personal data online.
-
1:06 - 1:08So the result is that we have behavioral,
-
1:08 - 1:10preference, demographic data
-
1:10 - 1:12for hundreds of millions of people,
-
1:12 - 1:14which is unprecedented in history.
-
1:14 - 1:17And as a computer scientist,
what this means is that -
1:17 - 1:19I've been able to build models
-
1:19 - 1:21that can predict all sorts of hidden attributes
-
1:21 - 1:23for all of you that you don't even know
-
1:23 - 1:25you're sharing information about.
-
1:25 - 1:28As scientists, we use that to help
-
1:28 - 1:30the way people interact online,
-
1:30 - 1:32but there's less altruistic applications,
-
1:32 - 1:35and there's a problem in that users don't really
-
1:35 - 1:37understand these techniques and how they work,
-
1:37 - 1:40and even if they did, they don't
have a lot of control over it. -
1:40 - 1:42So what I want to talk to you about today
-
1:42 - 1:45is some of these things that we're able to do,
-
1:45 - 1:47and then give us some ideas
of how we might go forward -
1:47 - 1:50to move some control back into the hands of users.
-
1:50 - 1:52So this is Target, the company.
-
1:52 - 1:53I didn't just put that logo
-
1:53 - 1:55on this poor, pregnant woman's belly.
-
1:55 - 1:57You may have seen this anecdote that was printed
-
1:57 - 1:59in Forbes magazine where Target
-
1:59 - 2:02sent a flyer to this 15-year-old girl
-
2:02 - 2:03with advertisements and coupons
-
2:03 - 2:06for baby bottles and diapers and cribs
-
2:06 - 2:07two weeks before she told her parents
-
2:07 - 2:09that she was pregnant.
-
2:09 - 2:12Yeah, the dad was really upset.
-
2:12 - 2:14He said, "How did Target figure out
-
2:14 - 2:16that this high school girl was pregnant
-
2:16 - 2:18before she told her parents?"
-
2:18 - 2:20It turns out that they have the purchase history
-
2:20 - 2:22for hundreds of thousands of customers
-
2:22 - 2:25and they compute what they
call a pregnancy score, -
2:25 - 2:28which is not just whether or
not a woman's pregnant, -
2:28 - 2:29but what her due date is.
-
2:29 - 2:31And they compute that
-
2:31 - 2:32not by looking at the obvious things,
-
2:32 - 2:35like, she's buying a crib or baby clothes,
-
2:35 - 2:38but things like, she bought more vitamins
-
2:38 - 2:39than she normally had,
-
2:39 - 2:41or she bought a handbag
-
2:41 - 2:43that's big enough to hold diapers.
-
2:43 - 2:45And by themselves, those purchases don't seem
-
2:45 - 2:47like they might reveal a lot,
-
2:47 - 2:49but it's a pattern of behavior that,
-
2:49 - 2:52when you take it in the context
of thousands of other people, -
2:52 - 2:55starts to actually reveal some insights.
-
2:55 - 2:57So that's the kind of thing that we do
-
2:57 - 2:59when we're predicting stuff
about you on social media. -
2:59 - 3:02We're looking for little
patterns of behavior that, -
3:02 - 3:05when you detect them among millions of people,
-
3:05 - 3:07lets us find out all kinds of things.
-
3:07 - 3:09So in my lab and with colleagues,
-
3:09 - 3:11we've developed mechanisms where we can
-
3:11 - 3:13quite accurately predict things
-
3:13 - 3:14like your political preference,
-
3:14 - 3:18your personality score, gender, sexual orientation,
-
3:18 - 3:21religion, age, intelligence,
-
3:21 - 3:22along with things like
-
3:22 - 3:24how much you trust the people you know
-
3:24 - 3:26and how strong those relationships are.
-
3:26 - 3:28We can do all of this really well.
-
3:28 - 3:30And again, it doesn't come from what you might
-
3:30 - 3:32think of as obvious information.
-
3:32 - 3:34So my favorite example is from this study
-
3:34 - 3:36that was published this year
-
3:36 - 3:37in the Proceedings of the National Academies.
-
3:37 - 3:39If you Google this, you'll find it.
-
3:39 - 3:41It's four pages, easy to read.
-
3:41 - 3:44And they looked at just people's Facebook likes,
-
3:44 - 3:45so just the things you like on Facebook,
-
3:45 - 3:48and used that to predict all these attributes,
-
3:48 - 3:49along with some other ones.
-
3:49 - 3:52And in their paper they listed the five likes
-
3:52 - 3:55that were most indicative of high intelligence.
-
3:55 - 3:57And among those was liking a page
-
3:57 - 3:59for curly fries. (Laughter)
-
3:59 - 4:01Curly fries are delicious,
-
4:01 - 4:04but liking them does not necessarily mean
-
4:04 - 4:06that you're smarter than the average person.
-
4:06 - 4:09So how is it that one of the strongest indicators
-
4:09 - 4:11of your intelligence
-
4:11 - 4:12is liking this page
-
4:12 - 4:14when the content is totally irrelevant
-
4:14 - 4:17to the attribute that's being predicted?
-
4:17 - 4:19And it turns out that we have to look at
-
4:19 - 4:20a whole bunch of underlying theories
-
4:20 - 4:23to see why we're able to do this.
-
4:23 - 4:26One of them is a sociological
theory called homophily, -
4:26 - 4:29which basically says people are
friends with people like them. -
4:29 - 4:31So if you're smart, you tend to
be friends with smart people, -
4:31 - 4:33and if you're young, you tend
to be friends with young people, -
4:33 - 4:35and this is well established
-
4:35 - 4:37for hundreds of years.
-
4:37 - 4:38We also know a lot
-
4:38 - 4:41about how information spreads through networks.
-
4:41 - 4:42It turns out things like viral videos
-
4:42 - 4:45or Facebook likes or other information
-
4:45 - 4:47spreads in exactly the same way
-
4:47 - 4:49that diseases spread through social networks.
-
4:49 - 4:51So this is something we've studied for a long time.
-
4:51 - 4:52We have good models of it.
-
4:52 - 4:55And so you can put those things together
-
4:55 - 4:58and start seeing why things like this happen.
-
4:58 - 4:59So if I were to give you a hypothesis,
-
4:59 - 5:03it would be that a smart guy started this page,
-
5:03 - 5:05or maybe one of the first people who liked it
-
5:05 - 5:06would have scored high on that test.
-
5:06 - 5:09And they liked it, and their friends saw it,
-
5:09 - 5:12and by homophily, we know that
he probably had smart friends, -
5:12 - 5:15and so it spread to them,
and some of them liked it, -
5:15 - 5:16and they had smart friends,
-
5:16 - 5:17and so it spread to them,
-
5:17 - 5:19and so it propagated through the network
-
5:19 - 5:21to a host of smart people,
-
5:21 - 5:23so that by the end, the action
-
5:23 - 5:26of liking the curly fries page
-
5:26 - 5:28is indicative of high intelligence,
-
5:28 - 5:29not because of the content,
-
5:29 - 5:32but because the actual action of liking
-
5:32 - 5:34reflects back the common attributes
-
5:34 - 5:36of other people who have done it.
-
5:36 - 5:39So this is pretty complicated stuff, right?
-
5:39 - 5:41It's a hard thing to sit down and explain
-
5:41 - 5:44to an average user, and even if you do,
-
5:44 - 5:46what can the average user do about it?
-
5:46 - 5:48How do you know that
you've liked something -
5:48 - 5:50that indicates a trait for you
-
5:50 - 5:53that's totally irrelevant to the
content of what you've liked? -
5:53 - 5:56There's a lot of power that users don't have
-
5:56 - 5:58to control how this data is used.
-
5:58 - 6:01And I see that as a real
problem going forward. -
6:01 - 6:03So I think there's a couple paths
-
6:03 - 6:04that we want to look at
-
6:04 - 6:06if we want to give users some control
-
6:06 - 6:08over how this data is used,
-
6:08 - 6:10because it's not always going to be used
-
6:10 - 6:11for their benefit.
-
6:11 - 6:13An example I often give is that,
-
6:13 - 6:14if I ever get bored being a professor,
-
6:14 - 6:16I'm going to go start a company
-
6:16 - 6:17that predicts all of these attributes
-
6:17 - 6:19and things like how well you work in teams
-
6:19 - 6:22and if you're a drug user, if you're an alcoholic.
-
6:22 - 6:23We know how to predict all that.
-
6:23 - 6:25And I'm going to sell reports
-
6:25 - 6:27to H.R. companies and big businesses
-
6:27 - 6:29that want to hire you.
-
6:29 - 6:31We totally can do that now.
-
6:31 - 6:32I could start that business tomorrow,
-
6:32 - 6:34and you would have
absolutely no control -
6:34 - 6:36over me using your data like that.
-
6:36 - 6:39That seems to me to be a problem.
-
6:39 - 6:41So one of the paths we can go down
-
6:41 - 6:43is the policy and law path.
-
6:43 - 6:46And in some respects, I think
that that would be most effective, -
6:46 - 6:49but the problem is we'd
actually have to do it. -
6:49 - 6:51Observing our political process in action
-
6:51 - 6:54makes me think it's highly unlikely
-
6:54 - 6:55that we're going to get a bunch of representatives
-
6:55 - 6:57to sit down, learn about this,
-
6:57 - 6:59and then enact sweeping changes
-
6:59 - 7:02to intellectual property law in the U.S.
-
7:02 - 7:04so users control their data.
-
7:04 - 7:05We could go the policy route,
-
7:05 - 7:07where social media companies say,
-
7:07 - 7:08you know what? You own your data.
-
7:08 - 7:11You have total control over how it's used.
-
7:11 - 7:13The problem is that the revenue models
-
7:13 - 7:14for most social media companies
-
7:14 - 7:18rely on sharing or exploiting
users' data in some way. -
7:18 - 7:20It's sometimes said of Facebook that the users
-
7:20 - 7:23aren't the customer, they're the product.
-
7:23 - 7:25And so how do you get a company
-
7:25 - 7:28to cede control of their main asset
-
7:28 - 7:29back to the users?
-
7:29 - 7:31It's possible, but I don't think it's something
-
7:31 - 7:33that we're going to see change quickly.
-
7:33 - 7:35So I think the other path
-
7:35 - 7:37that we can do down that's
going to be more effective -
7:37 - 7:38is one of more science.
-
7:38 - 7:41It's doing science that allowed us to develop
-
7:41 - 7:43all these mechanisms for computing
-
7:43 - 7:45this personal data in the first place.
-
7:45 - 7:47And it's actually very similar research
-
7:47 - 7:48that we'd have to do
-
7:48 - 7:51if we want to develop mechanisms
-
7:51 - 7:52that can say to a user,
-
7:52 - 7:54"Here's the risk of that action you just took."
-
7:54 - 7:56By liking that Facebook page,
-
7:56 - 7:59or by sharing this piece of personal information,
-
7:59 - 8:00you've now improved my ability
-
8:00 - 8:03to predict whether or not you're using drugs
-
8:03 - 8:05or whether or not you get
along well in the workplace. -
8:05 - 8:07And that, I think, can affect whether or not
-
8:07 - 8:09people want to share something,
-
8:09 - 8:12keep it private, or just keep it offline altogether.
-
8:12 - 8:14We can also look at things like
-
8:14 - 8:16allowing people to encrypt data that they upload,
-
8:16 - 8:18so it's kind of invisible and worthless
-
8:18 - 8:20to sites like Facebook
-
8:20 - 8:22or third party services that access it,
-
8:22 - 8:25but that select users who the person who posted it
-
8:25 - 8:28want to see it have access to see it.
-
8:28 - 8:30This is all super exciting research
-
8:30 - 8:32from an intellectual perspective,
-
8:32 - 8:34and so scientists are going to be willing to do it.
-
8:34 - 8:37So that gives us an advantage over the law side.
-
8:37 - 8:39One of the problems that people bring up
-
8:39 - 8:41when I talk about this is, they say,
-
8:41 - 8:43you know, if people start
keeping all this data private, -
8:43 - 8:45all those methods that you've been developing
-
8:45 - 8:48to predict their traits are going to fail.
-
8:48 - 8:52And I say, absolutely, and for me, that's success,
-
8:52 - 8:53because as a scientist,
-
8:53 - 8:57my goal is not to infer information about users,
-
8:57 - 9:00it's to improve the way people interact online.
-
9:00 - 9:03And sometimes that involves
inferring things about them, -
9:03 - 9:06but if users don't want me to use that data,
-
9:06 - 9:08I think they should have the right to do that.
-
9:08 - 9:11I want users to be informed and consenting
-
9:11 - 9:13users of the tools that we develop.
-
9:13 - 9:16And so I think encouraging this kind of science
-
9:16 - 9:17and supporting researchers
-
9:17 - 9:20who want to cede some of that control back to users
-
9:20 - 9:23and away from the social media companies
-
9:23 - 9:25means that going forward, as these tools evolve
-
9:25 - 9:27and advance,
-
9:27 - 9:28means that we're going to have an educated
-
9:28 - 9:30and empowered user base,
-
9:30 - 9:31and I think all of us can agree
-
9:31 - 9:33that that's a pretty ideal way to go forward.
-
9:33 - 9:36Thank you.
-
9:36 - 9:39(Applause)
- Title:
- The curly fry conundrum: Why social media “likes” say more than you might think
- Speaker:
- Jennifer Golbeck
- Description:
-
Much can be done with online data. But did you know that computer wonks once determined that liking a Facebook page about curly fries means you're also intelligent? Really. Computer scientist Jennifer Golbeck explains how this came about, how some applications of the technology are not so benign — and why she thinks we should return the control of information to its rightful owners.
- Video Language:
- English
- Team:
- closed TED
- Project:
- TEDTalks
- Duration:
- 10:01
Morton Bast edited English subtitles for Your social media "likes" expose more than you think | ||
Yasushi Aoki commented on English subtitles for Your social media "likes" expose more than you think | ||
Yasushi Aoki commented on English subtitles for Your social media "likes" expose more than you think | ||
Yasushi Aoki commented on English subtitles for Your social media "likes" expose more than you think | ||
Morton Bast approved English subtitles for Your social media "likes" expose more than you think | ||
Morton Bast edited English subtitles for Your social media "likes" expose more than you think | ||
Morton Bast edited English subtitles for Your social media "likes" expose more than you think | ||
Morton Bast edited English subtitles for Your social media "likes" expose more than you think |
Yasushi Aoki
do down -> go down
Yasushi Aoki
The description is different from the one on the TED page.
http://www.ted.com/talks/jennifer_golbeck_the_curly_fry_conundrum_why_social_media_likes_say_more_than_you_might_think
Yasushi Aoki
He said, "How did Target figure out...
# This is not words of the father.
http://www.forbes.com/sites/kashmirhill/2012/02/16/how-target-figured-out-a-teen-girl-was-pregnant-before-her-father-did/