1 Origin of the story

In many ways, neither big data nor artificial intelligence is new. As quantitative techniques required by (neo)liberal governance and capitalist logic, both technologies are shaped by a long history of manipulating statistics for 58 million businesses. C. Elish and D. Boyd’s Profit, Population Control and Governance (Foucault, 2009-2010; Ross, 1991). More specifically, technologies including big data and artificial intelligence reflect a long history of slow technological development aimed at achieving discrete outcomes (Jones, 2016). Nonetheless, these terms are far from precise in public parlance, instead giving rise to a series of myths about automation and data-driven technologies (boyd & Crawford, 2012; Gillespie, 2014). In order to fully understand the innovations and imaginations that underpin big data and artificial intelligence, it is necessary to historically and socially locate the relevant technological directions and practices. To do so requires an in-depth study of how those invested in the phenomenon have taken on cultural fantasies and encoded specific agendas into the technology. In this article, we often purposefully integrate big data and artificial intelligence into a single concept to focus on the phenomenon we are questioning. Nonetheless, these terms have different roots, and many practitioners would dispute how they are used in public discourse. Furthermore, the boundaries of “AI” as a field of study and an ideal goal are fuzzy and often controversial. Rather than relying on rigid definitions, we first provide a brief history of these terms to elicit the social context and research culture in which they emerged.

2 Big Data

Big data was born in big business. The specific technology of big data dates back to at least the 1990s, but the term entered the business discussion through a 2001 Gartner report that defined big data as the “3Vs”: volume, velocity, and variety (Laney, 2001). It is worth pointing out that while the term big data is a new term, the collection of data and the use of statistics to measure and manage populations can be traced back centuries (Hacking, 1982; Igo, 2007).

However, since the early 21st century, many scholars and experts have attempted to provide alternative definitions that cover the technologies and practices that underpin the big data phenomenon. For example, after listing 10 different operational definitions used in different contexts, Gil Press (2014) offers two of his own: (1) “the belief that the more data you have, the more insights and answers you can automatically derive from ones and zeros” and (2) “the new attitude among businesses, nonprofits, government agencies, and individuals is that combining data from multiple sources leads to better decisions.”1 Press’s definitions are more useful than the familiar business rhetoric of “volume, velocity, and variety” because they highlight aspects of the myth that underpins big data.

By 2010, technology companies and other businesses were beginning to embrace big data as a new business model (Manyika et al., 2014). al., 2011). Consulting firms emerged to help companies process data, while technology companies focused on selling their “cloud” servers and “software as a service” products to help companies store and manage data. Nonprofits and government agencies began to feel that they, too, needed to use data to “get smart.” To meet these needs, educational institutions and funding agencies began to rebrand statistics and computer science work as “data science” (Lohr, 2015). At the same time, companies with less good intentions began to prey on anxious organizations by selling big data solutions that were little more than evaporating software.

As more organizations and people embraced big data, critics began to question the value and purpose of such analysis. In particular, journalists, civil society advocates, and academics began to question data-driven approaches in areas such as communications, criminal justice, education, and employment. They also turned to questioning long-standing but growing data-centric practices in credit, insurance, and advertising (Poon, 2016; Trow, 2011). The Obama White House’s changing response to these technologies provides an example of how attitudes toward big data are slowly shifting. Initially enthusiastic about the economic potential of big data, the White House convened a series of experts in 2014 who ultimately highlighted the opportunities and concerns associated with this emerging technology area. In 2014, the White House issued a statement on “Seizing Opportunities, Preserving Values” that highlighted the opportunities and concerns associated with this emerging technology area. The first report on “Big Data” was quite optimistic (Podesta, Pritzker, Moniz, Holdren, & Zients, 2014). However, by 2016, their second Big Data report focused on “Algorithmic Systems, Opportunity, and Civil Rights” and provided a more worrying description of the potential for data discrimination (Muñoz, Smith, & Patil, 2016). Concerns focused on the amount of personal data collected and sold, and the potential for these technologies to be misused to exacerbate inequality and cause harm.

3 New and Old AI Technologies

While “AI” is new to those looking to repackage Big Data, AI itself has been around for decades. Contemporary The concept of artificial intelligence in the sense of the word was first used in the 1950s and crystallized in the Dartmouth Summer Research Project on Artificial Intelligence (McCorduck, 2004). 3 Bringing together recent advances in the “systems sciences” (Mindell, 1998), including cybernetics, information theory, systems theory, and cognitive science, researchers during this period predicted rapid progress in solving the “AI problem” Indeed, the current optimism surrounding machine learning and new AI technologies is reminiscent of the early decades of AI research, when predictions of AI’s future surpassing human capabilities dominated public discussion of these technologies (Dreyfus, 1972).

However, the above predictions were far from the reality of the slow development of AI software and hardware. In the 1950s and 1960s, millions of dollars (mainly from various branches of the Department of Defense) were spent on “centers of excellence” for AI research at universities such as MIT, Stanford, and Carnegie Mellon. By the mid-1970s, funding for AI research began to dry up, a period known as 60 million. C. Elish and D. Boyd called the computer science sector the “AI winter.” A scathing British government report (Lighthill, 1973) essentially declared AI projects a failure. In addition, as priorities changed at U.S. defense agencies that funded AI research (such as the Defense Advanced Research Projects Agency), the resources available for unrestricted basic research shrank, narrowing the scope of experiments researchers could conduct (Edwards, 1996; Mirowski, 2003).

While most AI research in the first few decades was theoretical or limited to experiments in academic laboratories, work in the field of “expert systems” in the late 1970s marked the first time that AI research could be explicitly and successfully applied to commercial industry (Russell & Norvig, 1995, pp. 21-22). These “expert systems,” also called “knowledge systems” or “knowledge-based systems,” were conceived as a supplement to, and sometimes a replacement for, complex decision support in professional settings, such as medical diagnosis. Information was collected from human experts (usually just one or two) and encoded into the rules and procedures that make up the computer system (Forsyth, 1993). In this way, expert systems were designed to simulate human expert decision making in complex environments. They flourished throughout the 1980s, but by the mid-1990s their popularity had waned. Expert systems began to be seen as “fragile,” working only in limited environments and with less-than-perfect results.

As hopes for widespread adoption of expert systems faded, a new set of techniques for achieving computer intelligence came to prominence. 4 This work involved a different set of assumptions about intelligence than the previous mainstream approach, which centered on the role of logic and rationality in abstract models of the world. In addition to developments in “behavior-based” robotics (Brooks, 1991), techniques from the field of machine learning also attracted research and development attention, including natural language processing, computer vision, and neural networks. Neural networks have their roots in cybernetic concepts of command and control, and in fact, neural networks were considered one of the leading approaches to solving artificial intelligence problems in the 1950s, so named because the concepts behind the way they work were loosely inspired by the function of neurons in the brain. However, the technology was soon ridiculed by the leading researchers of the time and was considered an unviable approach to artificial intelligence (Olazaran, 1996). In the 1980s, new research reignited interest, demonstrating that it could be used effectively for certain types of problems, such as object and speech recognition (Olazaran, 1996). Although more discrete than “expert systems,” research in the field of machine learning and neural networks showed promise for transitioning to successful products such as optical character recognition. Machine learning and deep learning, described by many as a “rebranding” of neural networks. Combined with large data sets and concentrated talent, machine learning is achieving things that seemed impossible a few years ago. These advances, tightly coupled with commercial companies, are only possible in the context of huge data sets, higher computing power, and widespread commercial commitment to big data.