There’s a much-read copy of the Mueller Report on Sinan Aral’s kitchen table, and the part that’s especially dog-eared is the part about the Russian government’s “sweeping and systematic” interference in the 2016 U.S. presidential election. Aral is a professor of marketing and analytics at MIT, and on the principle that “you cannot manage what you do not measure,” he and colleague Dean Eckles have presented, in Science magazine, a kind of “send me in, coach!” plan for social scientists to help protect elections from being gamed and fiddled with.
But while Facebook says it will demand more openness from political advertisers, and the nation’s election officials are on notice that the attacks on their systems will only get worse, it’ll take concerted political and public will and muscle and cooperation for social science to get what they need to figure out how bad the meddling into democracy is, and therefore how democracy can try to counter it.
The Senate Intelligence Committee singled out Russian interference in social media in misinformation campaigns targeting the presidential elections. But nobody seemed to be able to conclude what the impact of that was: did it swing the election or did it not swing the election? Is your work trying to figure that out?
We have experts that believe that a combination of Russian hacking and social media manipulation tipped the election for Donald Trump. And we have other experts who believe that it was likely too small-scale to have changed the results of the election.
What we have right now is a host of armchair theories about what happened and what is happening. And we have no definitive answers, so it’s very easy for skeptics to be skeptical and for conspiracy theorists to be convinced that something happened, because there’s no real concrete scientific evidence or results that we can hang our hat on.
The point of our piece is to say that the answer to this question is knowable but it is not known, because no one has actually done the work of measuring the impact of social media manipulation on any of these elections.
The methodologies for measuring this exist — it just hasn’t been measured. And to the extent that it remains unmeasured and not well understood, elections and therefore democracies around the world remain vulnerable to this kind of attack.
When I was reading what you had written in Science magazine, the obstacles struck me as tremendous. You’d have to get access to Facebook and social media data. You would need people who run American elections to cooperate. When you wrote this, did you think, we have a technique for doing this — but how are we going to get people to let us do it?
It’s funny you say that, because if you were to have asked me that question 10 years ago or even six, seven years ago, I would have said to you that research is essentially happening now.
Access to voter turnout data, and access to precinct level and state level vote choice data is essentially publicly available. Access to the type of social media data that you would need was routinely happening prior to the Cambridge Analytica scandal, by authorized academics in collaboration with Facebook and Twitter. And it had happened since the beginning of the platforms.
And [then] the Cambridge Analytica scandal, which was a clear violation of the terms of service of Facebook by an academic at Cambridge — to share that [Facebook] data with Cambridge Analytica.
For every Aleksandr Kogan, the research associate at the University of Cambridge [who created the app that was at the heart of the scandal], there were hundreds of legitimate researchers conducting collaborative research with Facebook and Twitter, and not violating the terms of service, and publishing science that describe how social media was affecting us.
We have papers published on how social media messaging affects voting behavior, shopping behavior, dating behavior, exercise behavior — all of these are studies were done and were published prior to Cambridge Analytica.
Since Cambridge Analytica, I think we have all observed a dramatic chilling effect on research with the platforms that is going to reduce our ability to understand how Facebook, Twitter, YouTube and Instagram are affecting very real outcomes, like our elections.
There is a lot more at stake than what exercise shoes you buy.
Absolutely. I think that the impact on elections and democracy is front and center these days. Robert Mueller is possibly the most understated man in Washington. And during his testimony, he said that the Russian government’s effort to interfere in our election is among the most serious challenges to democracy that he’s ever seen in his entire career. He stressed that the threat deserves the attention of every American.
It has dramatic implications for society, and our ability to understand those implications is also at risk, because the platforms have retracted from an open and transparent approach to analyzing what happened on their platform.
Even though Mark Zuckerberg has pledged openness and cooperation and does not seem to want Facebook to be a vehicle for political manipulation, do you see any indication that he’s willing to really open his books for this?
Facebook and Twitter and YouTube and Instagram are facing what I call a transparency paradox. On one hand, they are facing tremendous pressure to open the kimono and to show us how the inner workings of Facebook are affecting society, and what it’s doing to our elections, and what it’s doing to our children’s behavior and so on. At the same time, however, they’re facing tremendous pressure to lock down the data, to secure the data, to protect people’s private information.
These two pressures are at odds. To solve the transparency paradox, we have to thread this needle very carefully. We have to figure out a way for these platforms to become more secure and more transparent at the same time.
Another privacy obstacle is about voting itself, the secret ballot. How do you get information that would link, say, a misinformation campaign on social media to how someone ends up voting, without compromising those votes?
There are two avenues for a manipulation campaign to affect the results of the election. The first is to affect voter turnout. And the second is to affect vote choice — if people vote, and how people vote.
It is entirely conceivable that a sophisticated manipulation campaign on social media could affect the result of an election without changing the direction of one vote, just by changing the turnout numbers for one side or another.
Voter turnout data is much more readily available, and in fact public, and can be collected at the individual level, and has been for years and years in political science research.
Now, vote choice data is secret, and nobody is arguing that that should somehow be compromised in any way. But there are statistical techniques that could help us understand what the likely patterns of voting are in districts and in states in aggregate, and, where social media manipulation was targeted, to understand whether the outcome in a particular district or state was contrary to what we would expect — and whether those are related to or correlated with exposure to manipulative messages.
You’re talking about randomizing and aggregating, but does what you’re trying to find out come down to thinking, here’s a voter in Wisconsin who may be seeing Facebook messages that appear to be posted from friends of friends, and you want to know what he’s going to do when he gets into the voting booth — compared to what he might have done without this?
Yes. The best way to describe it is, what would the voters’ behavior have been had they not been exposed to Russian manipulation? What would the alternate universe of no Russia manipulation have looked like in terms of voting?
We do this every day when we are assessing the impact of advertising on product sales, when we try to calculate return on investment in the advertising industry. These techniques are known and have been used extensively in the past.
The stakes are so much higher, though.
The stakes are higher than return on an advertising campaign. It is ironic that the former is easier to do than the latter, that we’re better able to run experiments to measure the effect of our advertising on selling shoes than we are to do the same thing to understand whether our democracy is at stake.
Another obstacle is political will. The alarm bell has been sounded on Capitol Hill for not only manipulation but for voting interference. And nothing has happened, at least so far.
Let me just say upfront that I’m a registered independent, but at the same time I am hearing FBI Director Christopher Wray warn us that the threat just keeps escalating. Not only are Russia’s attacks likely to intensify in 2020 but he’s saying that other countries are entertaining whether to take a page out of that same book.
Robert Mueller has warned us of the same types of thing and yet we have potential legislation that is being blocked. And honestly, I think that this is essentially legislative malpractice. It is very difficult to understand how one could think, as [Senate majority leader] Mitch McConnell has said, that we have already taken action on this problem because there is no doubt that we have not.
Can you quickly run through the four steps that you’d like to do if Santa Claus were to say tomorrow, you can do anything you want and get the material you want?
There are four basic steps to answering what the effect of social media manipulation or Russian interference is in an election.
The first is to assess the message content and reach: how many messages were spread? And try to understand who is exposed to these messages.
The second step is to understand targeting, to understand which messages went to which people in which districts in which state?
The third step is to assess causal behavior change — what would have happened had people not been exposed to social media manipulation and what did happen when they were exposed to social media manipulation?
And the final step is, given the difference in what would have happened and what did happen, let’s aggregate that up and calculate what the effect of that is on voting behavior in terms of turnout and in terms of vote choice.
If we do those four steps, we not only have an understanding of how social media manipulation can affect elections that have happened in the past, but we have a blueprint for understanding how such manipulation can affect the 2020 presidential election and other elections as well.
Who would be your audience for these findings?
I think that everyone in the United States has a right to know. I also think that obviously Congress has a duty to understand what has happened, what is happening, and what will happen in upcoming elections.
The audience is actually beyond the United States, and targeted at citizens and legislatures of liberal democracies around the world. Research has estimated that a full third of the news that people consumed in Sweden before their general election was fake news.
There are fears that manipulation might have altered the effect of the Brexit vote in the United Kingdom. There are similar concerns about Indian elections and the last presidential election in Brazil.
This goes well beyond any one given election. It goes well beyond any one country. And I think that it’s not going away.