A colleague once described my work as “interweaving warm-blooded humans and the screens“, and I was delighted with that description. At that confluence of the social and the technical, I am becoming more and more mindful of the unequal power dynamics between disciplines in the face of new methodological developments such as computational social sciences, biosocial research, and digital humanities. This post is intended to be a little space here for my own continuous reflection.
Recently at a workshop on digital tools for the humanities, a Stanford graduate student rather poignantly noted that oftentimes collaboration with computer scientists felt more like colonization by computer scientists. This statement, even if not true, is far too sharp to ignore. Frankly, I think it is true. Not long after that workshop, I attended a THATCamp, where I spent my time teaching folks how to use Gephi, and I tried to spend some time telling them that the network they create is the result of an interpretive act. I don’t think they cared, I think they just wanted to know how to make node sizes change dynamically in tandem with partition filters. This is an issue that has concerned me for some time: the way wholesale importation of digital tools, techniques and objects into humanities scholarship tends to foster a situation where rich, sophisticated problems are contracted to fit conveniently into software.
If we are creating a mess by generating so many haystacks of big data that we are losing all the needles, then we need to figure out a different kind of way of doing things, as we cannot sew new cloth without any needles. Whatever else we make of the ‘big data’ hype, it cannot and must not be the path we take to answer all our big global problems. On the contrary, it is great for small questions, but may not so good for big social questions. Social scientists need to find a way not to be complicit in the new wave of struggle over the politics of method that is intrinsic to what big data brings.
But as Hanna Wallach writes in this great article in Communications of the ACM, the methodology is often instead: “Why not use these large-scale, social datasets in combination with the powerful predictive models developed by computer scientists”… and see what we get? (I guess you can replace “powerful predictive models” by any (for social scientists) non-standard method.) So “computational social science” has come to mean something slightly different from what it sounds like.
Within computing we have generally only focused on the wondrous and have ignored the terrifying or delegated its reporting to other disciplines. Now, with algorithmic governance replacing legal codes, with Web platform enabled surveillance capitalism transforming economics, with machine learning automating more of the labor market, and with unexplainable, non-transparent algorithms challenging the very possibility of human agency, computing has never been more deinon. The consequences of these changes will not be fully faced by us but will be by our children and our students in the decades to come. We must be willing to face the realities of the future and embrace our responsibility as computing professionals and academics to change and renew our computing curricula (and the worldview it propagates). This is the task we have been given by history and for which the future will judge us.
We reject the vague conceptualization of the discipline of ML as value-neutral. Instead, we investigate the ways that the discipline of ML is inherently value-laden. Our analysis of highly influential papers in the discipline finds that they not only favor the needs of research communities and large firms over broader social needs, but also that they take this favoritism for granted. The favoritism manifests in the choice of projects, the lack of consideration of potential negative impacts, and the prioritization and operationalization of values such as performance, generalization, efficiency, and novelty. These values are operationalized in ways that disfavor societal needs, usually without discussion or acknowledgment. Moreover, we uncover an overwhelming and increasing presence of big tech and elite universities in highly cited papers, which is consistent with a system of power-centralizing value-commitments. The upshot is that the discipline of ML is not value-neutral. We find that it is socially and politically loaded, frequently neglecting societal needs and harms, while prioritizing and promoting the concentration of power in the hands of already powerful actors.