A few years ago, Paul Edwards and colleagues (2011) introduced a notion of “science friction”—the idea that scientific datasets do not magically fuse together into a readily accessible “open” stockpile, and instead must be communicated and reshaped in order for scientists to collaborate across them. While it is all too easy to imagine endlessly wired interoperable devices, and bodies thoroughly mediated by fluid streams of measurement, the reality is not that simple. The Data Friction panel at the American Anthropological Association (AAA) meetings this past year attempted to take the idea of science friction further, and ask what else can we see when we turn our attention to frictionful encounters with data. This panel considered what alternative forms of knowing become possible by paying attention occasions where data fails to be mobile, or to the ways data and bodies resist being bound by models, devices, and infrastructures. What we see when we pay attention to frictions are significant questions of ownership, the slipperiness social relations, and examples of how people inhabit more fundamental social, material and conceptual incommensurabilities that data often surfaces. These social formations open up broader questions of the work that underlying notions of what constitutes “data” are doing.
Across the panel, it became clear that the presence of friction obscures certain things from participants’ point of view. In some respects, these ethnographies were ones where the participants themselves were unable to tell stories about what’s going on. In Ian Lowrie’s work, we heard about data scientists who manage data pipelines—people who structure the technical workflows for data and algorithms—who can’t pin down what “data” or algorithms” actually are, precisely because they shape shift when they encounter frictions of various kinds. Each movement in the pipeline involves a fusion or transformation of numbers, datasets, and/or algorithms, making it very difficult to separate figure from ground. Similarly, in Yulia Grinberg’s work, we heard from the makers of a wearable device designed to detect stress and arousal who struggled to pinpoint what their sensor actually indexes. The opaque “arousal” sensor picks up on everything from fight or flight responses, to exercise, to amorous feelings, to a general propensity to sweat. They can discuss what it might indicate when referring to very specific contexts—i.e., when it encounters friction with some other phenomenon, but this was all too on-the-move for there to be any stand alone, one to one referent. Another example is Patricia Lange’s work on Youtube users, who grew surprised and troubled when the social relations they seek online turn out to do work they did not expect. They believe themselves to be in control of their online persona, and were perennially surprised to learn that they are not.
In Bryce Peake’s research, we saw the presence of friction more directly addressed by the participants themselves. Those frictions stemmed from both Edward’s notion of friction and the sorts of post-colonial frictions that Tsing (2011) describes. His work involved participating in the “field” by building an open source software tool called Tinn, which ended up in a frictionful encounter with participants over data ownership. Like many others, his participants live with technical infrastructures that struggle to offer good reasons to trust—a failure which hits underprivileged groups even harder as part of a longer history of domination and oppression. In Bryce’s case, it meant his attempts to technically and legally build in meaningful control for participants over their own data were met with partial suspicion, and a good deal of uncertainty. These histories of dominance and division that Bryce found hard to escape.
Ian Lowrie made the point that this is not a case where we can end the analysis by noting that datasets are always already social. That sociality in turn has incommensurabilities, both social and material, that sustain its own set of struggles and connections. The concept of friction draws our attention to the work of making commensurability emerge, and alternatives when it cannot. This work goes beyond questions of metadata. We can see the work of overcoming incommensurabilities in the wearables industry that Grinberg describes, where marketing mobilizes both notions of objectivity and technical authority, and the language of subjective experience. Grinberg notes that these are fused together to do the contradictory job of persuading people of the value of technologies for improving health, while dissuading regulators from taking them seriously as medical devices (see also Fiore-Gartland and Neff 2016).
Commensuration is most starkly at issue in Tara Mahfoud’s work on the Human Brain Project. There, incommensurability is not a productive difference but a threat to the production of science, notably (I would argue) in a context where notions of scientific integration stand in for faith in European integration. Questions then emerge of whose commensurations are going to do that unifying work–whose models would frame the nature of the problem. Tellingly, scientists were simultaneously grappling with not knowing what data would even be important to someone else, and worth contributing, let alone which models would win the day. In this way, questions of what the shared social arena even is to begin with were asked while maintaining belief in the broader project of commensuration.
Grinberg’s wearables company had the ambition to bring emotion research into the “wild”, which raised similar issues of incommensurability. In my own research with Rajiv Mehta, we did in fact bring that type of sensor into the “wilds” of family caregivers—that is, people who have a family member with a serious medical condition, and as a result experience extraordinary levels of chronic stress and burnout. Our results were exactly as Grinberg reports here: our “stress” sensors often did pick up on fight-or-flight moments, but did not really speak to the long term strain and exhaustion that these people were experiencing, nor could they capture the complex web of associations that made some people respond in a fight-or-flight sort of way, and others respond differently to situations that could rightly be called crises. The sensors’ presence “in the wild” pointed out to deep uncertainties about what data really refers to, and partial commensurabilities between what is sensed and what is lived. However, the use of sensors in this way did bring “the world into the lab,” as Grinberg put it in the session—our participants were able to narrate their caregiving experiences thorough the sensor readings in a new way. In that sense, if we take data to be not a straightforward “measurement,” but a material site of frictions, data friction might be more than a good concept to think with. It might also prove a productive aspect of our methodological repertoire, by materially surfacing what partial or uncertain connections (Strathern 2005) might be going on in ways more nuanced than mere measurement.
This turn to the notion of friction is promising for thinking anthropologically about data. Whether Youtubers and their expectations of the social world around them, or in calibrating a data or modelling pipeline, or the deeply troubling power relations that the Tinn project was trying to traverse, there is a wide range of social work going on here. As much as pointing out that data is never “raw” is both true and important, the injunction to remember its sociality still holds out the prospect that data, by virtue of its connective semiotic properties, might accrue seemingly on its own in a big pot resting in the cloud. Friction invites us to speak more concretely on these matters.
If everything can conceivably be data, perhaps we can also use the notion of friction to ask what work calling something “data” does. We could speculate that framing things as “data” per se might help people suspend their disbelief about the frictions they will no doubt face, and to hold out hope that the stuff might connect to someone or something else (or fear, to the extent those connections can become damaging in the wrong hands). This body of work shows signs that a suspension of disbelief about the possibility that data might not move, or might not mean, could be a productive stance for actors to take. Seeking higher level data science abstractions create numerical commensuration. Maintaining faith in Youtube as a kind of archive creates a hope for kinship and social connection that may or may not be realizable. Both require a faith in the connective properties of data. Sometimes the suspension of disbelief is not useful. Lowrie, for example, noted an instance where data scientists warned their colleagues to speak concretely at a math conference: the math might postulate elegant flows of connection but real datasets are less good at that. We see the suspension of disbelief at its most destructive in the Mahfoud’s account of a brain project, where the hope invested in simulations creates a lacuna in dealing with more down-to-earth problems missing data, annotation discrepancies, etc.. This important work is also, notably, undervalued, invisible work (Gray et al 2016, Irani 2015). The hope for commensuration might be one element that persistently keeps it out of view.
Edwards, P., Mayernik, M.S., Batcheller, A., Bowker, G. and Borgman, C., 2011. Science friction: Data, metadata, and collaboration. Social Studies of Science, 41 (5), p. 667-690.
Fiore-Gartland, B. and Neff, G.(2016). Disruption and the Political Economy of Biosensor Data. In Nafus, D. (ed). Quantified: Biosensors in Everyday Life. MIT Press.
Irani, L., 2015. Justice for ‘data janitors’. Public Culture, 15.
Gray, M.L., Suri, S., Ali, S.S. and Kulkarni, D., 2016, February. The crowd is a collaborative network. In Proceedings of the 19th ACM Conference on Computer-Supported Cooperative Work & Social Computing (pp. 134-147). ACM.
Strathern, M,. 2005. Partial Connections. Rowman Altamira.
Tsing, A.L., 2011. Friction: An ethnography of global connection. Princeton University Press.