Let's send Facebook some election observers | Opinion

Facebook Privacy Scandal
Facebook CEO Mark Zuckerberg makes the keynote speech at F8, Facebook’s developer conference in San Jose.

Facebook insists that people can trust it to manage its giant social network in a way that doesn’t undermine democracy. I say, verify: Let’s send in an army of scientists to study its role in the upcoming 2018 elections.

There’s been too little research into how Facebook influenced the last election, even though there’s plenty of reason to think it did. The “I voted” button was designed to boost turnout. Search-engine rankings can affect voting preferences. Fake news was pervasive, though we still don’t know who controlled it and how much it mattered. As I’ve suggested, Facebook likely has all the data it needs to study the effect of Russian propaganda. So far, though, we haven’t heard from Facebook.

What to do? This is a scientific issue, not an ideological one, so we need to treat it scientifically. It’s not enough to look at correlations: We need to set up randomized experiments and actually test for causation. Such “A/B tests” — which, say, expose a treatment group to one headline and a control group to another — are standard practice in data science. In the hands of researchers with a strong sense of their academic, intellectual and patriotic duties, they could provide valuable insight into the state of democracy, both in the U.S. and elsewhere.

What I’m proposing isn’t entirely new. In April, Facebook announced that it would invite scholars to study the impact of social media on elections. Yet the details are sparse, and the ambitions probably are, too. True, some people worry that academics could get too much access, particularly to private data. That’s not unreasonable, but it’s also surmountable. It shouldn’t excuse us from demanding a reckoning.

Facebook’s efforts to provide more information on political ads won’t solve the problem either. For one, it might not catch every political ad — plenty aren’t overtly political but make the associated emotional nudges. Also, the data are insufficient: it’s important to know which people viewed an ad, not just how many. Then there’s the question of whether an ad changed minds: Die-hard Trump voters sharing die-hard Trump conspiracies is less of a threat than other kinds of propaganda.

Only scientific tests can offer useful answers. It doesn’t even matter whether we know in advance which ads to consider “political.” A well-designed study would allow us to measure the effects after the fact, thanks to the millions of individuals who use Facebook. It’s a scientific Garden of Eden if we can just make use of it.

Considering its recent scandals, Facebook should welcome the scientists. The company has pretty much zero claim to our trust in light of its failure to fulfill promises about the business model of WhatsApp and user privacy. It’s conflicted in policing political ads, given how much money it makes on them. It should be ready to make concessions.

We’re long past the quaint days when people objected to Facebook manipulating their emotions with news feeds. Social media platforms have real influence, and we should measure it instead of trying to pretend it doesn’t exist. The danger is that we demand too few studies, not too many. Let’s err on the side of being more informed.

Cathy O’Neil is a Bloomberg Opinion columnist. She is a mathematician who has worked as a professor, hedge-fund analyst and data scientist. She founded ORCAA, an algorithmic auditing company. @mathbabedotorg