2  A Note on Method

It is recognized that to trace the history of a term is insufficient to represent the full history of what that term indexes. In this case, the term indexes a complex assemblage of concepts, tools, and practices that characterize data science today, many of which clearly precede or parallel the use of the term. Nevertheless, the exercise serves as a valuable starting point from which to develop a complete historical account, since finding textual examples for a phrase’s string is relatively easy using textual databases, and because any related fields, such as operations research or data analysis or computational statistics, will be found to intersect with the term and may be pursed separately.

More important, although phrases like “data science” are, like all linguistic signs, arbitrary, they acquire motivation when they function as banners or brands under which allegiances are formed, catalyzing potential affiliations into actual ones. Such phrases are socially embedded speech acts with perlocutionary effects—they do not merely describe things in the world, they also instantiate them through their usage by agents, who influence the formation of their referents. This helps to explain why, once the phrase began to trend after 2008, many who previously would not affiliate themselves with the term began doing so, initiating a preferential attachment process to the term and thereby complexifying its definition. It also explains the purpose of efforts to define the field of data science, or to explain it away: each definition has a prescriptive dimension, since by proposing a “correct” definition, it attempts to influence usage and the field it denotes. The present essay is no exception.

Another reason for beginning with a history of words, via their traces as character strings, is that in historical research it is much easier to study words than the things they stand for, although we often (conveniently) forget this relationship and conflate the two, believing we can easily move from language to the world. In our perceptions of the world beyond the ken of our immediate experience—and even there—we are enmeshed in language to a greater degree than we may like to admit. Words in the form of written documents (texts) constitute the primary source of data on which the construction of historical understanding depends. So, even though one may wish to get past words and study things as they are, the fact that these things are in the past, and mainly represented to us through documents and other material traces, means that one must begin with these. Ultimately, however, the purpose of working with written records is to get at the things they stand for and index, much as quantitative data \(\mathcal{D}\) are used to construct an hypothesis \(\theta\) to explain their existence. We may regard this phase of work as similar to the Bayesian task of establishing likelihoods and priors on the way to estimating posteriors.¹

It is helpful to understand the larger theoretical lens through which these methods are applied and this history is presented and interpreted. The primary assumption is that science and all forms of knowledge production are cultural systems, in the sense proposed by Clifford Geertz and others (Geertz 2017; Martin 1998). To locate science within culture is not to affirm or deny the objectivity of science, or its effectiveness relative to other ways of knowing, but simply to assert that science, like all human endeavors, is made possible in and through social interaction over time and space through the media forms that make such interaction possible. Among these media forms the most significant is language. Given this, I adopt a discourse-centered approach to understanding culture (Urban 1993). Further, as a means to theorize the causal relationship between language and world, I adopt the view that discourse—spoken and written language, as opposed to generative grammar—is best understood as situated action (Mills 1940; Suchman 1987; Norman 1993). In this view, language does not simply refer to the world, but participates in it, as a resource that enables distributed cognition and produces effects through its use in concrete situations. From an interpretive perspective, the meanings of words index the work they perform. This essentially causal conception of language use allows us to make sense of the relationship between phrases like “data science” and the human endeavors with which they are associated, the posterior relationship encoded in the Bayesian framework described above.

Finally, no claims are made for having discovered the first actual usages of the term in question, neither at its origin nor during any of its transformations. Instead, the documentary record that comprises the sum of databases and documents, both digital and material, available to the author is regarded as a kind of film, or an archaeological settlement pattern, on which collective verbal behavior impinges and leaves its marks. It is likely to be incomplete, but also comprehensive enough to capture patterns to a degree of resolution high enough to support the claims being made.

Geertz, Clifford. 2017. The Interpretation of Cultures. Basic Books.

Martin, E. 1998. “Anthropology and the Cultural Study of Science.” Science, Technology & Human Values 23 (1): 2444. https://doi.org/10.1177/016224399802300102.

Mills, C. Wright. 1940. “Situated Actions and Vocabularies of Motive.” American Sociological Review 5 (6): 904–13. https://doi.org/10.2307/2084524.

Norman, Donald A. 1993. “Cognition in the Head and in the World: An Introduction to the Special Issue on Situated Action.” Cognitive Science 17 (1): 1–6. https://doi.org/10.1207/s15516709cog1701_1.

Suchman, Lucy A. 1987. Plans and Situated Actions: The Problem of Human-Machine Communication. 2nd ed. Cambridge University Press.

Urban, Greg. 1993. A Discourse-Centered Approach to Culture: Native South American Myths and Rituals. University of Texas Press.

---

1. This is more than an analogy. If we think of the work of textual interpretation on which historical research depends in a probabilistic framework, we may express the hermeneutic relationship between words \(Sr\) and meanings \(Sd\) as follows:

   \[P\left( Sd \middle| Sr \right) = \frac{P(Sd)P\left( Sr \middle| Sd \right)}{P(Sr)}\]

   Inasmuch as \(Sr\) and \(Sd\)—Saussure’s signifier and signifed—represent what Schleiermacher called the “linguistic” and “psychological” aspects of interpretation, the formula also expresses the logic of the hermeneutic circle as a matter of updating the prior and recomputing the posterior (Palmer 1969). The analogy between the Bayesian approach to causality and the hermeneutic approach to meaning has been noted by others (Groves 2018; Friston and Frith 2015; Ma 2015; Reason and Rutten 2018; Frank and Goodman 2012).