According to Clifford Geertz, the purpose of anthropology is not to explain culture but to explicate it. That should cause us to rethink our relationship with machine learning. It is, we contend, perfectly possible that machine learning algorithms, which are unable to explain, and could even be unexplainable themselves, can still be of critical use in a process of explication. Thus, we report on an experiment with anthropological AI. From a dataset of 175K Facebook comments, we trained a neural network (...) to predict the emoji reaction associated with a comment and asked a group of human players to compete against the machine. We show that a) the machine can reach the same accuracy as the players, b) it fails in roughly the same ways as the players, and c) easily predictable emoji reactions tend to reflect unambiguous situations where interpretation is easy. We therefore repurpose the failures of the neural network to point us to deeper and more ambiguous situations where interpretation is hard and explication becomes both necessary and interesting. We use this experiment as a point of departure for discussing how experiences from anthropology, and in particular the tension between formalist ethnoscience and interpretive thick description, might contribute to debates about explainable AI. (shrink)

This article maps a controversy in network science over the last 15 years, dividing the field about the epistemic status of a central notion, scale-freeness. The article accounts for the two main disputes, in 2005 and in 2018, as they unfolded in academic publications and on social media. This article analyzes the conflict, and the reasons why it reignited in 2018, to the surprise of many. It is argued that the concept of complex networks is shared by the distinct subcultures (...) of theorists and experimentalists; and that these subcultures have incompatible approaches to knowledge: nomothetic and idiographic. Following Galison, this article contends that network science is a trading zone where theorists and experimentalists can trade knowledge across the epistemic divide. (shrink)

It is increasingly common in natural and social sciences to rely on network visualizations to explore relational datasets and illustrate findings. Such practices have been around long enough to prove that scholars find it useful to project networks in a two-dimensional space and to use their visual qualities as proxies for their topological features. Yet these practices remain based on intuition, and the foundations and limits of this type of exploration are still implicit. To fill this lack of formalization, this (...) paper offers explicit documentation for the kind of visual network analysis encouraged by force-directed layouts. Using the example of a network of Jazz performers, band and record labels extracted from Wikipedia, the paper provides guidelines on how to make networks readable and how to interpret their visual features. It discusses how the inherent ambiguity of network visualizations can be exploited for exploratory data analysis. Acknowledging that vagueness is a feature of many relational datasets in the humanities and social sciences, the paper contends that visual ambiguity, if properly interpreted, can be an asset for the analysis. Finally, we propose two attempts to distinguish the ambiguity inherited from the represented phenomenon from the distortions coming from fitting a multidimensional object in a two-dimensional space. We discuss why these attempts are only partially successful, and we propose further steps towards a metric of spatialization quality. (shrink)

In this article, we present a few lessons we learnt in the establishment of the Sciences Po médialab. As an interdisciplinary laboratory associating social scientists, code developers and information designers, the médialab is not one of a kind. In the last years, several of such initiatives have been established around the world to harness the potential of digital technologies for the study of collective life. If we narrate this particular story, it is because, having lived it from the inside, we (...) can provide an intimate account of the surprises and displacements of digital research. Founding the médialab in 2009, we knew that we were leaving the reassuring traditions of social sciences to venture in the unexplored territory of digital inscriptions. What we couldn't foresee was how much such encounter would change our research. Buying into gospel of Big Data, we imagined that the main novelty of digital research came from handling larger amounts of data. We soon realized that the interest of digital inscriptions comes instead from their proliferating diversity. Such diversity encouraged us to reshape our professional alliances, research practices and theoretical perspectives. It also led us to overcome several of the oppositions that used to characterize social sciences and to move in the direction of a more continuous sociology. (shrink)