By Gordon Hull
Surprise! Facebook is back in the news and the doghouse, this time for allowing vast amounts of user data to find its way to Cambridge Analytica, which then used it to try to elect Donald Trump. The only surprise is that anyone is surprised. I’ll review why that is first, then offer a term that gets at what I take to be the real problem: we live in a deliberately-created hostile information architecture. Let’s first rehearse why the FB-Cambridge Analytica nexus is not a surprise.
(1) FB users perhaps didn’t generally know this, but the fact that apps had access to users’ friends’ data was known by academics. Back in 2009, two colleagues in the College of Computing and Informatics and I published a paper that complained about how Facebook’s interface and architecture were the source of some of its privacy problems. We used as an example that apps had access to not just your personal information, but that of all your friends. I claim no originality on this point: we cited previously published research.
(2) That FB could influence an election is old news, too. A 2012 study showed that if FB put a “go vote” reminder on users’ pages, they could increase turnout by a small, but measurable amount. Zeynep Tufekci noted the implications in 2014:
“A biased platform could decide to use its own store of big data to model voters and to target voters of a candidate favorable to the economic or other interests of the platform owners. For example, a study published in Nature found that civic “go vote” messages that were targeted in Facebook through users’ social networks (thanks to a voting encouragement app deployed by Facebook) resulted in a statistically significant increase in voter turnout among those targeted, compared with a similar “go vote” message that came without such embedding in social ties (Bond, et al., 2012). A platform that wanted to manipulate election results could, for example, model voters who were more likely to support a candidate it preferred and then target a preponderance of such voters with a “civic” message narrowcast so that most of the targets were in the desired target group, with just enough thrown in from other groups to make the targeting less obvious. Such a platform could help tilt an election without ever asking the voters whom they preferred (gleaning that information instead through modeling, which research shows is quite feasible) and without openly supporting any candidate. Such a program would be easy to implement, practically undetectable to observers (since each individual only sees a portion of the social media stream directed and nobody sees the totality of messages in the whole platform except the platform owners), easily deniable (since the algorithms that go into things like Facebook’s news feed are proprietary and closely guarded secrets), and practically unconfirmable.”
As Tufekci notes, Jonathan Zittrain posted a similar worry at about the same time.
(3) FB itself published on the ways it could manipulate users. For example, FB published a paper in PNAS that proved that you could manipulate people’s emotional state over networks. It also published a somewhat desperate study purporting to prove that it didn’t contribute to ideological echo chambers, at least not much, relying on the ideological construct that data is neutral.
(4) Cambridge Analytica’s role in the Trump campaign wasn’t even secret. A frightening article on Motherboard broke the story a year ago (recall here).
(5) It’s not hard to get to the idea that the problem is FB’s business model. As I said the last time I blogged about FB and Russian election meddling not being a surprise, “it is time to start properly regulating companies like Facebook. The public needs to know something about the convoluted methods by which its access to the world is curated. Platform companies like FB are making billions of dollars pretending to be neutral conduits for public conversation, even when they most assuredly are not.” The current disclosures about Cambridge Analytica underscore that the real issue is that the entire business model for companies like Facebook is predicated on doing… exactly what they did. They sell advertisers access to their users, access which can be tailored by the collection of vast amounts of data about those users. That an academic researcher might have misled FB as to what he was going to do with data he collected isn’t the point. The point is that companies like Facebook have vast amounts of social power, and that their interests do not naturally align with the interests of either their users or the idea of a national community. The sooner regulatory agencies wake up to this fact, and do something about it, the better.
***
Fine, but what do we do? As Colin Koopman points out in a recent NYT piece, the problems with FB are part of a larger problem, which is that our data-driven society, which is supposedly going to magically, wonderfully change everything about every aspect of life, is being built out without any care or concern for what might happen as a result. Even conceding the obvious point that the “change everything about everything” language is hype of the sort that always accompanies new technologies, data is likely to have a significant impact in a lot of areas.
As I emphasized above, and as Koopman also notes, we know that the incentives of those who make money on data generally do not align with the incentives of those whose data is used, often unknowingly. This problem is not confined to the private sector, either: as Andrew Selbst argues, the use of data in policing has built-upon and reinforced racial inequality, even when its adoption is well-intentioned. Indeed, policing even relies on some of the same data that Facebook uses. Thanks to the “third party doctrine,” once data is voluntarily handed over to a third party (say, a FB app, or even FB itself), there is essentially nothing that an individual can expect in the way of restrictions on its sharing. Helen Nissenbaum provided the conceptual vocabulary to explain this: privacy needs to recognize that information always exists in contexts, and the norms which govern how it uses in one context are not necessarily the same in other contexts. So privacy theory needs to pay attention not just to the release of information, but to the flow of it from one context to another. Selbst proposes one regulatory solution: the required use of “algorithmic impact statements” (along the lines of environmental assessments) accompanying the adoption of data technologies.
Along these lines, Koopman makes a general point:
“What we need is for an ethics of data to be engineered right into the information skyscrapers being built today. We need data ethics by design. Any good building must comply with a complex array of codes, standards and detailed studies of patterns of use by its eventual inhabitants. But technical systems are today being built with a minimal concern for compliance and a total disregard for the downstream consequences of decades of identifiable data being collected on the babies being born into the most complicated information ecology that has ever existed.”
That seems right; what I want to emphasize here is first that the language of technological neutrality, and of data objectivity, which underpins the idea that we can build out a data-driven economy without regulation, is pure ideology. But – and this is the real point – as such, it is itself a regulatory strategy. As Lawrence Lessig famously argued, we are regulated in a general sense by at least four factors: law, markets, social norms, and architecture (this is why Koopman’s building analogy is so apt). Each of these is to an extent malleable; as Lessig emphasized, when you’re talking about computers, the architecture portion – ‘code’ – is far more malleable than offline.
In other words, one way to present the current regulatory situation is that we’ve defaulted our way to a very specific regulatory structure. Following the work of people like Robert Rosenberger, I propose to call it “hostile information architecture.” As Rosenberger points out, when you see things like armrests in the middle of benches at bus stops, you should assume that they are there to prevent homeless people from lying down on the benches. When you see little leaf decorations over all the stair railings and retaining walls, as you do on my campus, you should assume that they are there to stop skateboarders. Hostile urban architecture is designed to select for some uses in urban spaces at the expense of others. It generally favors property owners and uses they condone. The current regulatory environment surrounding Facebook does the same thing to the infosphere, especially as it forms part of the civic space where we interact as a polity. As such, it’s an example of the neoliberal strategy of the complete subsumption of all of social life into the market and of the political consequences of that subsumption.
It’s not the only example of course; the FCC’s decision to rollback net neutrality is another excellent example of hostile information architecture. Deregulating in one way is regulating in another (in other words, “deregulation” is itself an ideology. The question is not whether to have regulation, but what kind to have. “Free Market” is an illusion that does a lot of ideological work). But the permissiveness of things like the third party doctrine or lack of oversight of FB is also a good way to understand in very concrete terms what that neoliberal strategy might look like and how it might work, as well as how directly hostile it is to the demos that constitutes democracy (recall here). Each of the five reasons why we shouldn’t be surprised by the Cambridge Analytica debacle and the ways it exploited social media to destroy public space exemplifies the problems of our hostile information architecture.
That is what needs regulatory attention, and creating a more hospitable information architecture should structure the specific regulatory strategies employed to rein in Facebook.
Recent Comments