4 The 1970s
It is clear that by the early 1970s the term data science had been in circulation in several contexts and referred to ideas and tools relating to computational data processing. Importantly, these usages were not obscure—the AFCRL was one of the premier research laboratories in the world and closely connected with Harvard and MIT (Altshuler 2013), an international cross-roads of intellectual life where many would have come into contact with the term. Similarly, IBM and UNIVAC, the sources of the founders of two self-proclaimed data science companies, were the two largest computer manufacturers at the time.1
4.1 Naur’s Datalogi
Although the AFCRL closed the Data Sciences Lab by 1970,2 the term continued to be used, most notably by the Danish computer scientist Peter Naur, who suggested that computer science, a relatively new field, be renamed to data science. His argument, consistent with previous usage, was that computer science is fundamentally concerned with data processing and not mere computation, i.e. what the AFCRL derided as numerical calculation. Earlier, in the 1960s, Naur had coined the term “datalogy” (Danish: datalogi) for this purpose, but later found the term data science to be a suitable synonym, perhaps due to its currency or to his familiarity with the DSL, which shared his research interest in developing programming languages (Naur 1966, 1968). In contrast to the AFCRL, Naur provided an explicit definition of data science:
The starting point is the concept of data, as defined in [0.7]: DATA: A representation of facts or ideas in a formalized manner capable of being communicated or manipulated by some process. Data science is the science of dealing with data, once they have been established, while the relation of data to what they represent is delegated to other fields and sciences.
The usefulness of data and data processes derives from their application in building and handling models of reality.
…
A basic principle of data science is this: The data representation must be chosen with due regard to the transformation to be achieved and the data processing tools available. This stresses the importance of concern for the characteristics of the data processing tools.
Limits on what may be achieved by data processing may arise both from the difficulty of establishing data that represent a field of interest in a relevant manner, and from the difficulty of formulating the data processing needed. Some of the difficulty of understanding these limits is caused by the ease with which certain data processing tasks are performed by humans [Naur (1974): 30-31; emphasis and citation in original; the reference “0.7” refers to Gould (1971)].
Clearly, Naur’s definition inherits the classical definition described above; it locates the meaning of the term in the series of practices associated with the larger activity of data processing. These practices include establishment, choice of representation, conversion and transformation, the modeling of reality, and the guiding of human actions. One difference is that Naur is keen to locate data science within a division of labor implied by this general process, separating data science per se from the work of data acquisition (establishment) and the domain knowledge required to acquire data effectively. In this view, data science is more specifically concerned with the formal representation of data (i.e. with data structures and models), a practice that must be done in light of how data are to be transformed downstream, and with which tools (i.e. algorithms and programming languages). As we shall see, the weighting that Naur assigns to this kind of work is not inherited by later theorists. However, the general image of a sequential process with distinct phases in the life cycle of data is. Here we see the appearance of the image of a pipeline, unnamed but implied by the concept of process, which dominates the mental representation of the field from its origins in the 1960s.
Far from being a fluke, Naur’s usage developed the classical definition of data science initiated by NASA and the Air Force, intentionally or not. The fact that his attempt to rename computer science failed outside of his native country (and Sweden) is not important; his understanding of computer science sheds light on how closely the concept of data was (and is) related to computation and process.
It is worth noting that Naur’s definition implies a familiarity with the real-world provenance of data processing in industry and government. Indeed, by this time computational data processing had penetrated all sectors of society, and the pressure to improve tools and methods to represent and process data had increased as well. As a result of this pressure, two important data standards were developed in this period: Codd’s relational model, which laid the foundation for SQL and commercially viable relational databases in the 1980s, and Goldfarb’s SGML, which would become a standard for encoding so-called unstructured textual data (such as legal documents) and later the basis for HTML and XML Goldfarb (1970). This focus on the human context of data processing is reflected in his later work; a volume of selections of his writing from 1951 to 1990, which includes his essay on data science, is entitled “Computing: A Human Activity” (Naur 1992).
4.2 Other Uses
The term appears in the Proceedings of the IFIP World Conference
(Computers in Education: Proceedings of the IFIP ... World Conference 1975) Computers in Education, Proceedings of the IFIP World Conference
“The case is easily obscured, alas, by an instinct to confuse mathematics and data science” (p. 756)
“Recommending data science and data technology as areas of intellectual endeavor, and data studies as a subject for school learning;” (p. 758)
4.3 Parzen
Interestingly, in 1977 a prefixed variant of the term appeared in the title of the technical report, “Non-Parametric Statistical Data Science: A Unified Approach Based on Density Estimation and Testing for ‘White Noise’” (Parzen 1977). However, Parzen later published a version of this work as “Nonparametric Statistical Data Modeling” (Parzen 1979), which indicates that his original word choice was mistaken. Yet the original choice may not have been entirely unmotivated: Parzen’s work attempted to unify parametric and non-parametric methods under one umbrella. Given the natural inclination of mathematical statistics for the former and data analysis for the latter, his choice of the term data science may have signaled an attempt to encompass both approaches to data. It is also worth noting that he later used to the term to introduce a “new culture of statistical science called LP MIXED DATA SCIENCE” (Parzen and Mukhopadhyay 2013), after the term became popular.3 Whether or not this unifying goal was his motivation, statisticians later became quite interested in the term for precisely this reason.
4.4 The 1980s
As evidence for the visibility of the AFCRL as this time, here is an add that appeared in a 1964 issue of the British weekly Nature:
↩︎A SYMPOSIUM on “Models for the Perception of Speech and Visual Form”, sponsored by the Data Sciences Laboratory of the Air Force Cambridge Research Laboratory, will be held in Boston during November 11-14. Further information can be obtained from Mr. G. A. Cushman, Wentworth Institute, 550 Huntington Avenue, Boston, Massachusetts 02115 (“Announcements” 1964).
Altshuler writes that the lab was “abolished” in June 1972 “in response to a large reduction in manpower authorizations” (Altshuler 2013: 27). However, the unit is not mentioned in the July 1970 to June 1972 research report (AFCRL 1973). The lab’s closure may have been a consequence of the passing of the Mansfield Amendment in 1969, which prohibited the military from carrying out “any research project or study unless such project or study has a direct and apparent relationship to a specific military function” (Appropriations 1970: 348). ↩︎
Parzen’s use of the term “culture” here echoes his comments on Breiman’s famous essay on two cultures of statistical models, where he suggested that there are in fact several cultures, including his own, to which he devoted the majority of his response (Breiman 2001: 224–226).↩︎