The Surveillance AI Pipeline (2309.15084v2)

Abstract: A rapidly growing number of voices argue that AI research, and computer vision in particular, is powering mass surveillance. Yet the direct path from computer vision research to surveillance has remained obscured and difficult to assess. Here, we reveal the Surveillance AI pipeline by analyzing three decades of computer vision research papers and downstream patents, more than 40,000 documents. We find the large majority of annotated computer vision papers and patents self-report their technology enables extracting data about humans. Moreover, the majority of these technologies specifically enable extracting data about human bodies and body parts. We present both quantitative and rich qualitative analysis illuminating these practices of human data extraction. Studying the roots of this pipeline, we find that institutions that prolifically produce computer vision research, namely elite universities and "big tech" corporations, are subsequently cited in thousands of surveillance patents. Further, we find consistent evidence against the narrative that only these few rogue entities are contributing to surveillance. Rather, we expose the fieldwide norm that when an institution, nation, or subfield authors computer vision papers with downstream patents, the majority of these papers are used in surveillance patents. In total, we find the number of papers with downstream surveillance patents increased more than five-fold between the 1990s and the 2010s, with computer vision research now having been used in more than 11,000 surveillance patents. Finally, in addition to the high levels of surveillance we find documented in computer vision papers and patents, we unearth pervasive patterns of documents using language that obfuscates the extent of surveillance. Our analysis reveals the pipeline by which computer vision research has powered the ongoing expansion of surveillance.

• The paper quantitatively and qualitatively demonstrates that over 90% of annotated computer vision studies aim to extract human data.
• It reveals a five-fold growth in surveillance-related patents from the 1990s to the 2010s, driven by advances in AI research.
• The study uncovers obfuscating language in research that masks the true extent of human-targeted surveillance applications.

The Surveillance AI Pipeline

The paper "The Surveillance AI Pipeline" by Pratyusha Ria Kalluri et al. conducts an extensive analysis of the relationship between computer vision research and surveillance applications by examining three decades of computer vision research papers and downstream patents. The authors quantitatively and qualitatively assess over 40,000 documents to elucidate the direct pathways through which advancements in computer vision research have facilitated and empowered surveillance technologies.

Their analysis establishes that a predominant share of computer vision research is geared towards extracting human data. Specifically, they quantify that 90% of annotated computer vision papers and patents self-report enabling human data extraction, and 68% explicitly aim to extract data about human bodies and body parts. The paper meticulously identifies types of human data targeted, delineating four key categories: human body parts, human bodies, human spaces, and other socially salient human data.

Through a combination of in-depth content analysis and large-scale computational analysis, the paper not only uncovers the nature of human data extracted but also tracks the growth of surveillance-related studies over time. The number of computer vision papers incorporated into surveillance patents surged more than five-fold from the 1990s to the 2010s. Indeed, nearly 80% of these papers contributed to surveillance patents in the 2010s compared to 50% in the 1990s. This trend marks a significant shift within the field towards developing technologies that enhance surveillance capabilities.

Institutional and National Contributions

This paper also reveals that the contribution to surveillance is a widespread norm across prominent institutions and nations. Key entities, including elite universities and major tech corporations, are responsible for a significant share of the research papers that result in surveillance patents. Notably, this trend is not limited to a few "rogue" institutions but is pervasive across many institutions, nations, and subfields within computer vision. For example, both the US and China, followed by other prominent countries, considerably contribute to the surveillance technology pipeline. The results indicate that 74% of institutions and 83% of nations producing computer vision research see the majority of their patented work being utilized in surveillance technologies.

Obfuscation of Surveillance Intent

A critical aspect illuminated by the research is the frequent use of obfuscating language in papers and patents that mask the true extent and specificities of surveillance applications. The authors find pervasive patterns where the terms like "objects" often subsume humans, thereby deflecting direct acknowledgment of human data surveillance in their terminologies. Moreover, the actual application of many technologies to human surveillance is frequently detailed in supplementary figures and data sets rather than the main text, further veiling the surveillance implications from immediate scrutiny.

Implications and Future Developments

The implications of these research findings are profound, shedding light on the central role computer vision research plays in the expansion and normalization of surveillance technologies. Theoretical implications underline a necessary reevaluation of the purported neutrality in AI research, especially within computer vision. Practically, these findings suggest a nuanced reconsideration of current regulations and ethical guidelines governing AI research and its applications. Moreover, the paper serves as a clarion call for greater transparency and accountability within AI research communities.

As the field progresses, these insights highlight the critical necessity of developing frameworks and technologies that can ensure ethical usage and curb the deleterious impacts of Surveillance AI. Future developments could focus on defining clear boundaries and guidelines for data collection, storage, and application, with overarching governance that addresses the subtleties and implications of human data extraction.

In conclusion, this paper provides a comprehensive and detailed analysis of how computer vision research substantially powers mass surveillance technologies. The revelation of the Surveillance AI pipeline calls for a conscientious reflection within the AI research community to revisit and reform research practices to foster technology development that respects and enhances human dignity and privacy.