Summary The Surveillance AI Pipeline Analyzing Research and Patents arxiv.org
13,440 words - PDF document - View PDF document
One Line
The study uncovers the extensive use of human data extraction in surveillance technologies by elite universities and big tech companies, emphasizing the need for regulation and public involvement.
Slides
Slide Presentation (12 slides)
Key Points
- Computer vision research in the field of artificial intelligence (AI) is contributing to the expansion of mass surveillance.
- An analysis of three decades of computer vision research papers and patents reveals that human data extraction is prevalent in computer vision technology.
- Elite universities and "big tech" corporations are involved in surveillance patents, challenging the perception that only a few entities contribute to surveillance.
- The number of computer vision papers used in surveillance patents has significantly increased over the years.
- Language in computer vision papers and patents often downplays or hides the extent of surveillance, using terms like "objects" to refer to humans.
- Computer vision research is foundational to the paradigm of surveillance and has social and ethical implications.
- The study provides insights into the institutions, nations, and subfields involved in surveillance patents and exposes obfuscating language used in computer vision documents.
- The methodology involved analyzing papers and patents, identifying key dimensions of surveillance AI, and conducting a large-scale computational analysis.
Summaries
30 word summary
The study examines 40,000 research papers and patents, revealing widespread human data extraction in surveillance technologies. It implicates elite universities and big tech companies, calling for regulation and public influence.
61 word summary
"The Surveillance AI Pipeline" analyzes over 40,000 computer vision research papers and patents to expose the prevalence of human data extraction in surveillance technologies. Elite universities and "big tech" companies are implicated in thousands of surveillance patents, challenging the notion that only a few rogue entities are responsible. The study calls for critical examination, regulation, and public influence over surveillance technologies.
159 word summary
"The Surveillance AI Pipeline" is a research paper that examines the relationship between computer vision research and the development of surveillance technologies. The study analyzes over 40,000 computer vision research papers and patents to uncover the prevalence of human data extraction in computer vision. Elite universities and "big tech" corporations are implicated in thousands of surveillance patents, challenging the perception that only a few rogue entities are responsible for surveillance. The research reveals a significant increase in the use of computer vision research in surveillance patents over time. The authors also highlight the use of obfuscating language in computer vision papers and patents to downplay the extent of surveillance. The findings emphasize the need for critical examination of the field and recognition of the social and ethical implications of computer vision technologies. The study aims to empower communities to organize against surveillance, guide policymakers in regulation, shape the research agenda, and enable public knowledge and influence over surveillance technologies.
580 word summary
"The Surveillance AI Pipeline" is a research paper that examines the relationship between computer vision research and the development of surveillance technologies. The authors argue that computer vision, specifically in the field of artificial intelligence (AI), is contributing to the expansion of mass surveillance. They aim to uncover the pathway from computer vision research to surveillance applications and shed light on the normalization of Surveillance AI.
The study analyzes over 40,000 computer vision research papers and downstream patents spanning three decades. The authors find that the majority of these papers and patents focus on extracting data about humans, particularly human bodies and body parts. They present quantitative and qualitative analysis to demonstrate the prevalence of human data extraction in computer vision.
The research also investigates the institutions involved in computer vision research and their connection to surveillance patents. Elite universities and "big tech" corporations, which contribute significantly to computer vision research, are cited in thousands of surveillance patents. This challenges the notion that only a few rogue entities are responsible for surveillance, as many institutions, nations, and subfields involved in computer vision research are implicated.
The study reveals a significant increase in the use of computer vision research in surveillance patents over time, with more than a five-fold increase between the 1990s and 2010s. The linguistic analysis of paper titles also shows a shift towards a greater focus on analyzing humans and semantic categories in computer vision.
The authors highlight the use of obfuscating language in computer vision papers and patents, which downplay or hide the extent of surveillance. Terms like "objects" are used to refer to humans, minimizing the acknowledgment of human data extraction. Images of humans may be included in figures and datasets without explicit mention in the text, further obscuring the connection to surveillance.
The findings challenge the perception of computer vision research as a neutral pursuit and emphasize its foundational role in surveillance. The authors argue for a critical examination of the field and recognition of the social and ethical implications of computer vision technologies.
The research contributes to understanding the widespread extraction of human data in computer vision and the normalization of Surveillance AI. It provides insights into the institutions, nations, and subfields involved in surveillance patents and exposes the obfuscating language used in computer vision papers and patents. The authors hope that their findings will empower communities to organize against surveillance, help policymakers identify regulatory targets, guide researchers in shaping the research agenda, and enable the public to have knowledge and influence over surveillance technologies.
The study employed a methodology that involved analyzing papers and patents from the Conference on Computer Vision and Pattern Recognition (CVPR) from 1990-2020. Various databases were used to gather the data, and content analysis was conducted by a team of experts. The researchers identified key dimensions of surveillance AI and found that the majority of papers and patents focused on extracting data about humans, with an emphasis on human bodies and body parts.
Large-scale computational analysis of over 40,000 papers and patents was conducted to study the breadth and variation of surveillance across years, institutions, nations, and subfields. Surveillance indicator words were identified, and the corpus was scanned for patents containing these words. The majority of papers with downstream patents were found to be used in surveillance patents.
The study also provided background information on surveillance and computer vision, highlighting the impact of surveillance on marginalized communities and ethical concerns surrounding dataset collection and curation practices in computer vision.
696 word summary
"The Surveillance AI Pipeline" is a research paper that explores the connection between computer vision research and the development of surveillance technologies. The authors argue that computer vision, particularly in the field of artificial intelligence (AI), is contributing to the expansion of mass surveillance. They aim to uncover the pathway from computer vision research to surveillance applications and shed light on the normalization of Surveillance AI.
The study analyzes three decades of computer vision research papers and downstream patents, totaling over 40,000 documents. The authors find that the majority of annotated computer vision papers and patents report their technology enables the extraction of data about humans, specifically human bodies and body parts. They present both quantitative and qualitative analysis to demonstrate the prevalence of human data extraction in computer vision.
The research also examines the institutions that produce computer vision research and their involvement in surveillance patents. Elite universities and "big tech" corporations, which are prolific in computer vision research, are cited in thousands of surveillance patents. This challenges the narrative that only a few rogue entities contribute to surveillance, as the majority of institutions, nations, and subfields that author computer vision papers with downstream patents are implicated in surveillance.
The study reveals a significant increase in the use of computer vision research in surveillance patents over the years. Between the 1990s and 2010s, there has been a more than five-fold increase. The linguistic analysis of paper titles also shows a shift towards an increased focus on analyzing humans and semantic categories in the field of computer vision.
The authors highlight the obfuscation of language in computer vision papers and patents, which often downplay or hide the extent of surveillance. Terms like "objects" are used to refer to humans, minimizing the acknowledgment of human data extraction. Figures and datasets may contain images of humans without explicit mention or discussion in the text, further obscuring the connection to surveillance.
The findings challenge the perception of computer vision research as a neutral pursuit and reveal its foundational role in the paradigm of surveillance. The authors argue that progress in computer vision is closely tied to the expansion of Surveillance AI. They emphasize the need for a critical examination of the field and a recognition of the social and ethical implications of computer vision technologies.
The research contributes to the understanding of the pervasive extraction of human data in computer vision and the normalization of Surveillance AI. It provides insights into the institutions, nations, and subfields involved in surveillance patents and exposes the obfuscating language used in computer vision papers and patents. The authors hope that their findings will serve as a tool for communities to organize against surveillance, for policymakers to identify regulatory targets, for researchers to shape the research agenda, and for the public to exercise knowledge and power over surveillance technologies.
The methodology involved analyzing papers and patents from the Conference on Computer Vision and Pattern Recognition (CVPR) from 1990-2020. The researchers used various databases to gather the data and employed a team of experts to conduct content analysis. They identified key dimensions of surveillance AI, including data type, transferal, and use. They found that the majority of papers and patents extracted data relating to humans, with a focus on human bodies and body parts.
The study also conducted a large-scale computational analysis of over 40,000 papers and patents to study the breadth and variation of surveillance across years, institutions, nations, and subfields. The researchers identified surveillance indicator words and scanned the corpus for patents containing these words. They found that the majority of papers with downstream patents were used in surveillance patents.
The analysis of the evolution across years showed that the number of computer vision papers with downstream patents stabilized in the early 2000s and remained above 200 every year until 2018 when it suddenly dropped by nearly half. The linguistic evolution demonstrated changes in the focus of papers and patents, with certain words becoming more polarized in their association with surveillance.
The study also provided additional background on surveillance and computer vision, highlighting the impact of surveillance on marginalized communities and the ethical concerns surrounding dataset collection and curation practices in computer vision.
969 word summary
The Surveillance AI Pipeline is a research paper that analyzes the connection between computer vision research and the development of surveillance technologies. The authors argue that computer vision, particularly in the field of artificial intelligence (AI), is contributing to the expansion of mass surveillance. They aim to uncover the pathway from computer vision research to surveillance applications and shed light on the normalization of Surveillance AI.
The study analyzes three decades of computer vision research papers and downstream patents, totaling over 40,000 documents. The authors find that the large majority of annotated computer vision papers and patents self-report their technology enables the extraction of data about humans, specifically human bodies and body parts. They present both quantitative and qualitative analysis to demonstrate the prevalence of human data extraction in computer vision.
The research also examines the institutions that produce computer vision research and their involvement in surveillance patents. Elite universities and "big tech" corporations, which are prolific in computer vision research, are cited in thousands of surveillance patents. This challenges the narrative that only a few rogue entities contribute to surveillance, as the majority of institutions, nations, and subfields that author computer vision papers with downstream patents are implicated in surveillance.
The study reveals a significant increase in the number of computer vision papers used in surveillance patents over the years. Between the 1990s and 2010s, there has been a more than five-fold increase in the use of computer vision research in surveillance patents. The linguistic analysis of paper titles also shows a shift towards an increased focus on analyzing humans and semantic categories in the field of computer vision.
The authors highlight the obfuscation of language in computer vision papers and patents, which often downplay or hide the extent of surveillance. Terms like "objects" are used to refer to humans, minimizing the acknowledgment of human data extraction. Figures and datasets may contain images of humans without explicit mention or discussion in the text, further obscuring the connection to surveillance.
The findings of this study challenge the perception of computer vision research as a neutral pursuit and reveal its foundational role in the paradigm of surveillance. The authors argue that progress in computer vision is closely tied to the expansion of Surveillance AI. They emphasize the need for a critical examination of the field and a recognition of the social and ethical implications of computer vision technologies.
The research contributes to the understanding of the pervasive extraction of human data in computer vision and the normalization of Surveillance AI. It provides insights into the institutions, nations, and subfields involved in surveillance patents and exposes the obfuscating language used in computer vision papers and patents. The authors hope that their findings will serve as a tool for communities to organize against surveillance, for policymakers to identify regulatory targets, for researchers to shape the research agenda, and for the public to exercise knowledge and power over surveillance technologies.
The methodology of the study involved analyzing papers and patents from the Conference on Computer Vision and Pattern Recognition (CVPR) from 1990-2020. The researchers used the Microsoft Academic Graph, paper-patent citation linkages, and the Google Patents database to gather the data. The content analysis was conducted using a team of six experts who used an inductive-deductive methodology to analyze the documents. They identified key dimensions of surveillance AI, including data type, data transferal, and use of data. They found that the majority of papers and patents extracted data relating to humans, with a focus on human bodies and body parts. They also analyzed the transferal of human data, finding that it is often transferred to other institutions and not under the control of the datafied person. The researchers also examined the institutional use of data and identified three categories: modeling or categorizing humans, soft influence, and hard control. They found that many subfields of computer vision contribute to Surveillance AI, even ones not explicitly connected to human data.
The study also conducted a large-scale computational analysis of over 40,000 papers and patents to study the breadth and variation of surveillance across years, institutions, nations, and subfields. The researchers identified surveillance indicator words and scanned the corpus for patents containing these words. They validated each word through manual inspection and created a list of approved surveillance indicator words. They then scanned each paper's downstream patents to identify patents containing these words. They found that the majority of papers with downstream patents were used in surveillance patents.
The analysis of the evolution across years showed that the number of computer vision papers with downstream patents stabilized in the early 2000s and remained above 200 every year until 2018 when it suddenly dropped by nearly half. To study the linguistic evolution, the researchers computed the log-odds ratio of words appearing in 1990s paper titles versus 2010s paper titles. They found that there were changes in the focus of papers and patents, with certain words becoming more polarized in their association with surveillance.
The study also provided additional background on surveillance and computer vision. Surveillance is a technology of social control that is intrinsically tied to the production of power relations. It perpetuates inequalities and disproportionately impacts marginalized communities. Computer vision has rapidly risen in the past decade due to the availability of image and video data. However, dataset collection and curation practices often lack considerations of informed consent, privacy, and the mitigation of negative social stereotypes. The field has been criticized for its focus on efficiency, universality, and impartiality, which can lead to the erosion of privacy and the reproduction of social stereotypes.
Overall, the study provides a comprehensive analysis of the Surveillance AI pipeline, examining the data extraction, transferal, and use of human data in computer vision research and applications. It highlights the pervasive nature of surveillance in the field and raises important ethical considerations.