The world needs computational social science, and it is not only, or even primarily, about the AI revolution. The reasons follow below and I also cover what a computational social scientist should know, do, and a few words about our hairstyles.
Should it really be an academic discipline?
Maybe this is changing, but five or so years ago, when making money from the digital traces we leave behind was all the rage, computational social science (CSS) became, in the mind of many people, the academic counterpart of the information economy. There was a notable split between those who thought it was for educating future Google employees—social media data science—and those who believed it is, you know, social science that is computational.1
Another thing that confuses people around me is whether it is the topics or methods that are computational. Books and overview articles start with “new types of data sets need new methods,” and thus don’t separate these aspects. The educational side of CSS could and should contain all of the above, but where are its scientific merits?
Even if technology changes society or adds new scientific methods, it cannot change the big questions of social science: How do, or should, people organize? Why are there inequalities, social classes, etc.? CSS has to earn the “social science” part of its name, or it is simply not important. But if the traditional disciplines already address the big questions, why do we need CSS?
The methods lab
In a parallel universe with a different academia, we would not have needed computational social science as a division of science. However, because of the idiosyncrasies of the mainstream social science disciplines, we do. Economics, sociology, political science, anthropology, . . . they all have their modes of explanation, accepted set of methods, career paths, seminar culture, fashion (maybe even hairstyles)?
Norms and conventions facilitate communication and understanding; if everyone reinterpreted the scientific method with every paper, following the progress would be a real headache. Still, the norms and conformity of several social sciences are stifling. Maybe it’s a good idea for the traditional disciplines to outsource some of the methods R & D to a discipline of its own, and since the bleeding edge of social science methods development is computational, let’s identify that discipline with CSS.2 But then, shouldn’t there be a computational sociology different from computational economics, etc.? I think not. Even more, that might be the true raison d’être for CSS. The methodologically pioneering spirit of CSS is precisely what we need to lower the barriers between the traditional disciplines.
Epistemological anarchy
I’m not the biggest fan of Paul Feyerabend, but if I read Against Method3 with my thickest reconciliatory glasses, I can see he’s got a point. The oh-so-human aspects of science—to build conventions, jargon, rules, and other restrictive structures that are the Lego blocks of society—are in the way of fulfilling our potential as knowledge creators. Why don’t we just grab the knowledge out there with the best means available? Why do we care about p = 0.05, or existence-and-uniqueness, when humans in action don’t?
I do believe the most productive, disruptive, paradigm-shifting periods of science were characterized by what Feyerabend calls “epistemological anarchy” (but I read as “epistemological freedom” since it is about diminishing the power of conventions, not institutions). And as long as CSS embodies such creatively chaotic anarchy, it deserves its name as a discipline.
… but well-groomed
Computational social scientists have a duty not to let conventions and disciplinary boundaries stop them. But that doesn’t mean we should drift away into a world of our own; quite the opposite. Science only fulfills its objective—to generate and collect knowledge—if others can understand it. We need to work with and relate our results to the disciplinary sciences and their theories. After all, we are overlapping with the traditional disciplines. If our knowledge doesn’t reach them, then what’s the point? So, we also need to speak the language and understand the big picture and priorities of each of the various social sciences. We need disciplinary knowledge and not just from one discipline. Even though we are scientific rebels, we must blend in in many contexts. We’re social scientists and not mere methods guys. Ditch the (metaphorical) mohawk and get a proper hairstyle.
The ideal computational social science
In summary, CSS is, first and foremost, a division of social science whose goal is to generate knowledge about society and the social human and make it accessible to everyone. It should do that in total epistemological freedom, being informed by everything from the latest in ML to the oldest in sociology and not bound by conventions of what constitutes a scientific explanation, etc. It should disseminate results back to the traditional disciplines but without obligations to get them adopted by said disciplines.4
The ideal computational social scientist
Does the above sound like too much? Am I saying that aspiring computational social scientists should know Habermas as well as they know feature selection? Well, that’s the ideal, but becoming very versatile needn’t be so challenging. The traditional disciplines are so extraordinarily specialized that sacrificing only a little depth would free up time to become very broad.
Maybe the biggest challenge is that we need to embrace our yin and yang—being iconoclasts and friendly neighborhood soccer coaches at once. We’re not creating a beautiful synthesis of decades of diverse science;5 we’re not leaning on time-tested theories when epistemology storms outside;6 instead, we’re making something new, using material from the past or not. At the same time, we need our disciplinary colleagues and their knowledge—both for their input to our research and as critics. And we need them to need us. Nothing beats collaborations with people from different backgrounds. It’s profoundly rewarding when a group’s diversity-elevated collective intelligence becomes more than a cliché.
Notes
- H Wallach, 2018. Computational social science ≠ computer science + social data. Commun. ACM 61:42–44. ↩︎
- Two minor notes here: First, I don’t think one can decouple methods development and actual research. Most good methods come from trying to gain some specific knowledge or dealing with some particular data. Second, in sociology (but maybe also other disciplines), methodology is a vibrant, stand-alone research topic (which I always found a bit strange, but ok). ↩︎
- P Feyerabend, 1975. Against Method: Outline of an Anarchistic Theory of Knowledge. New Left Books, New York. ↩︎
- That would have been a different story if the social sciences were more oriented toward closing open problems. Cf.: DJ Watts, 2017. Should social science be more solution-oriented? Nat. Hum. Behav. 1:15. ↩︎
- L Henrickson, B McKelvey, 2021. Foundations of “new” social science: Institutional legitimacy from philosophy, complexity science, postmodernism, and agent-based modeling. Proc. Natl. Acad. Sci. USA 99:7288–7295. ↩︎
- AF Wise, DW Shaffer, 2015. Why theory matters more than ever in the age of big data. J Learn Anal 2:5–13. ↩︎