UC Berkeley

This dissertation, entitled Programming Insight: Human and Machine Intelligence in the Petabyte Age, explores the ideal of “intelligent” organizations through a close reading of a data fusion and analysis environment for government and enterprise by Palantir Technologies. Using approaches from science and technology studies, human-computer interaction, and new media theory, the project links understandings of computation and interactivity with emerging infrastructures for knowledge practices. My work examines the relationships between epistemology, ethics, and aesthetics for a digital platform that works in support of organizations that, in the words of former NSA director Michael Hayden, “kill people based on metadata.”

My dissertation examines the digital and human processes that purport to transform digital traces into knowledge and insight in domains from finance to counter-terror. Commonly cast as a generic surveillance technology, I argue that Palantir’s products index a range of technical and cognitive performances that allow analysts and their organizations to see digital traces as constitutive of agential activities like data exfiltration, fraud, and terrorism. Through an analysis of publicly available materials including demonstration videos, conference proceedings, technical papers, and patent filings, I present the rhythms of a digital environment that elicits a hybrid investigative intelligence in contexts where data are presumptively and overwhelmingly computerized. Drawing on contemporary and historical work in science and technology studies and human-computer interaction, the project traces these instances of knowledge production as composite epistemic performances among humans and machines. The environment and interface for these performances are constructed as a procedurally designed site for improvised interactions – as a space of play. What results are investigative narratives that name, explain, and target instances of elusive, anomalous, or threatening behavior mediated by digital systems. These structuring categories, however, are neither fundamental nor disinterested, but emerge from highly cathected figures of organizational risk including whistleblowers, malicious insiders, cybercriminals, and terrorists. I contend that such data-based, organizational “intelligence” is a constructed effect of distributed cognitive assemblages, and that a variety of agents and epistemic mediators – from machine learning algorithms, to graph visualizations, to experienced human analysts – shape the resulting possibility space, the concepts invoked, and the knowledge produced.

The project contributes to critical work on new media, risk, and expertise, as I analyze how recent Big Data technologies present these risks as networked phenomena rendered from massive sets of relational digital data. In particular, I argue that contemporary critiques of risk management must engage with a growing domain of problems and techniques centered on the detection of anomalies. The challenges of anomaly detection encompass not only the identification and management of risk to prevent negative outcomes, but also related opportunities for the exploitation of anomalies, most prominently the pricing of assets in financial markets. The mechanics of knowledge discovery and data mining (KDD) systems and how they represent and recirculate anomaly and risk – among organizations that alternately seek to exploit the risk profiles of others and minimize their own – brings studies of digital media interfaces and infrastructure into conversation with science and technology studies and political thought. The construction of such systems may signal a shift away from the nation-state as the center of technopolitical critique, as a national government or agency becomes one customer and/or vendor in a larger field of non-democratic firms and organizations exchanging access to datasets, analysis tools, and other technologies of modern governance.