Jul 11, 2018 11:30 AM

Facebook Opens Its Private Servers to Scientists Studying Fake News

Social Science One, an independent research commission, will give social scientists unprecedented access to Facebook's data.

This much is obvious: What the world needs now is less fake news. In general, sure, but particularly on the planet's leading source of information: Facebook. The thing is, to separate the informational wheat from the disinformational chaffe, what you actually need is a better definition of fake news. And that's … well … less obvious.

"What does it mean, exactly? It's not always clear. We know, for instance that if somebody doesn't like a story they see on Facebook that they might classify it as false, when it actually just reflects an opposing ideology," says political scientist Gary King, director of Harvard University's Institute for Quantitative Social Science. But just because you find an ideological viewpoint abhorrent doesn't mean it's factually incorrect.

So King wants to know: Could some clever researcher develop rules capable of classifying news as true or false—rules that could be shared with other people, platforms, algorithms, and universally applied? "Is it possible? I don't know," King says. "But it would be a terrific development."

It's the kind of development King is optimistic could emerge from Social Science One, an independent research commission he cofounded to give social scientists unprecedented access to data inside Facebook and—one day, he hopes—other private companies with troves of scientifically valuable user data. The organization, and Facebook's participation, was first announced in April, but its official name was only announced today, along with its inaugural mission: To investigate the spread of information and misinformation on Facebook, and their impact on elections and democracy.

For years, accessing Facebook's private data came with a whopper of a caveat: Whatever findings your research turned up had to be preapproved by the company—before you made it public. But Social Science One, acting as an intermediary, removes that condition. The organization has insight into what kind of data Facebook has available and what kind of data researchers need. Now it's bridging the gap: Starting today, researchers from around the world can apply for funding and data access that Social Science One will approve—not Facebook. If researchers want to search for something in the platform's data that could make it look bad—or if they actually find something—Facebook won't be able to pump the brakes.

The first data set Facebook makes available will comprise roughly one petabyte—that's one million gigabytes—of privacy-protected data on public Facebook posts, including many links to intentionally false news stories. Contained in the data set will be anonymized information on things like the age, gender, and political views of the people who clicked those links; what kind of device they used to access them; which links they viewed, shared, and reshared (including the ones they shared without clicking on); along with the number of likes and loves and wows and so-ons that the posts received.

It's … a lot of data. But, hopefully, very useful data. "There's an enormous number of questions you can ask about what 2 billion people around the world are clicking and reading and sharing," King says.

Researchers will have a month to submit requests for money (up to $50,000 each) or data, before the initial review process kicks off. King says he's streamlined the process to make it easier for researchers to apply. (The description of the proposed investigation can be no longer than five pages, single spaced—a limit he says scientists accustomed to 20-page grant applications will appreciate.) The Social Science Research Council, an international social science nonprofit and a separate entity from Social Science One, will handle the money, which will flow from seven ideologically diverse organizations, from the Charles Koch Foundation to the James L. Knight foundation. SSRC will also oversee the peer-review process.

"We’ve got a broad, globally distributed, and diverse set of top-tier data scientists who are starting as peer reviewers," says SSRC president Alondra Nelson. The initial review process, if all goes as planned, with take roughly six weeks. After that comes a month of training researchers in how to access Facebook's data securely. Then come the actual studies, which could take anywhere from days to years to complete.

If the application process is easier than researchers are accustomed to, the review process will be tougher for King and the various subcommittees of Social Science One. With the help of the SSRC, the commission will conduct additional reviews for ethics and privacy, to avoid any Cambridge Analytica–style fiascos. Any scientist seeking access to money or data will need to pass not only the standard review protocols of their home institutions, but a second, special review conducted by specialists specifically appointed by the SSRC¹.

"We decided we needed a higher level of ethical and privacy standards, but it's impossible to pick a standard everyone agrees on," King says. So the SSRC is appointing experts on emerging concerns surrounding privacy and ethics¹—experts well versed in issues on which university review boards may not be up to date.

If balancing the interests of academics, the public, and a private company like Facebook sounds complicated, that's because it is. "One of the things I've developed an appreciation for, in conversations about accessing Facebook data, is how hard it is from a practical perspective," says Yale University psychologist David Rand. An expert in the dissemination of misinformation on Facebook, Rand is unaffiliated with Social Science One but has followed its development closely. Facebook’s data lives in a complicated set of storage formats that take a lot of work to unpack and stitch back together—something most academics are incapable of doing on their own. He says that even if one assumes that Facebook has the best of possible intentions, making the company's data available and useful to researchers will be no small feat—and will likely require a considerable investment of resources from the company.

"One of the key questions is going to be not just how willing Facebook is to do that, but, to be fair, how able they are to do it—and whether they'll continue to participate in the long term."

King, for his part, is bullish about the whole affair. You have to be, he says. "You can’t take on a project as incredibly complex as this unless you're wildly optimistic," King tells me with a laugh. "So of course I think it’s going to go perfectly. But I also recognize that’s probably not the case."

1. Correction at 10:45 a.m. on 7/12/2018: An earlier version of this story contained language that suggested Social Science One would appoint specialists in the research proposal review process and oversee privacy and ethics reviews. These duties actually lie in the domain of the SSRC.