This paper demonstrates the uses of facial recognition as an auxiliary dataset to reidentify anonymous individuals in public spaces and using dating apps. It also demonstrates the ability to use facial recognition to collect "Personally Predictive Information" such as social security numbers from anonymous individuals using by combining face recognition with data mining algorithms and statistical re-identification techniques. This raises significant concerns about the role of data mining as online and offline worlds blend increasingly seamlessly with augmented reality technologies.

Synthesizes perspectives from multiple fields to propose an economic theory of privacy, showing the relative value of control and disclosure of private information. Individual disclosers of information are often subject to an asymmetrical power relationship to the economically powerful collectors of information, establishing a fraught ethical terrain.

Examines how the U.S. Genetic Non-Discrimination Act (2009) could not have predicted and regulated certain big-data applications of genetic testing in the workplace. The author shares several helpful examples of how seemingly neutral genetic testing practices lead to racialized disparities in the workplace.

This important investigative reporting shows how predictive software used in many U.S. courts’ sentencing and bail proceedings is systematically biased against non-white defendants. Although the predictive algorithm does not directly account for demographic factors when producing a score for likely recidivism, it does use social and economic data that is functionally a proxy for historically-biased social and political relations. Additionally, because the algorithm is treated as a proprietary secret it is not available for public examination or adversarial procedures in court.

This paper challenges the notion that data mining and machine learning is an inherently unbiased method for decisions making. The authors examine bias and fairness in big data through the lense of US laws that bar discrimination in employment. Because algorithms would appear to outsource decisions to machines—and thus appear to escape the legal requirement that victims of discrimination must be able to demonstrate an intent to discriminate—the best hope for legal relief may come from doctrines of disparate impact. The authors examine the challenges of using antidiscrimination law to address big data and show the challenges of utilizing existing methods to remedy discrimination.

A widely influential collection of essays in critical information theory that centers classificatory systems in the study of technoscience. Illustrates the ethical dynamics of information systems by demonstrating that while classification systems can be deeply oppressive by "torquing" the possibilities of their lives, people are often able to find degrees of agency within the same system that oppresses.

In this widely cited piece, the authors offer a broad interrogation of big data as a technology, analytical method and a mythology. They provide six "provocations" to inform the growing field of critical data studies: big data changes the definition of knowledge; claims to objectivity and accuracy are misleading; bigger data are not always better data; taken out of context, big data loses its meaning; just because it is accessible does not make it ethical; limited access to big data creates new digital divides.

Examines the paradox that giving individuals more control over their personal data tends to lead them to release more sensitive data by increasing their confidence that the collector of the data has good intentions. This shows how some approaches to increasing privacy controls may backfire.

Proposes the adoption of "Consumer Subject Review Boards" as an IRB-analog for commercial applications of big data.

Commercial applications of big data effectively skirt privacy regulations meant to protect Personally Identifiable Information (PII). This article proposes an approach to mitigate "predictive harms" through procedural data due process.

Data & Civil Rights is an annual conference that brings together leaders in civil rights, government, industry, and academic research, to address an urgent question: How do we preserve and strengthen civil rights in the face of data-driven technological change?

The code of conduct produced by the Data Science Association.

A broad look at ethical issues in big data, with a eye toward business practices.

Algorithms are increasingly important power-brokers in society, but critics and journalists rarely have the tools necessary to unpack exactly how algorithms leverage power over society and individuals. This report details methodological tools for understanding and reverse engineering the input/output relationships in algorithms. The author details the mechanics behind the atomic decisions that algorithms make, including prioritization, classification, association, and filtering.

Investigative reporting that details how companies compile predictive profiles of customers by correlating disparate data points. These profiles allow the companies to micro-target customers using highly personal information that the customers may not know they have shared.

This technical paper illustrates mathematical techniques for achieving "fairness in classifications" made in machine learning. The authors identify how algorithms can be trained to follow a fairness constraint so similar persons are treated similarly without causing discrimination or losing the utility of the classification system. The technical insights are applied to the problem of statistical parity in affirmative action admission policies at colleges.

This essay examines how algorithmic decision-making relies on classifications systems that are not inherently fair or neutral representations of human behavior or groupings. The authors contend that the best way to understand and regulate machine learning as a sociotechnical system is by giving values an analytic priority over technical concerns—machine-made decisions are ultimately representations about what human programmers value.

A report from the Obama Administration examining how privacy can be protected while maximizing the social, economic and scientific utility of big data.

An exploration of how scholarly papers will change with the advent of collaborative writing tools and data storage options.

Set of guidelines for handling and sharing scientific data.

Seeks to debunk the idea that machine learning is fair by default by virtue of . Emphasizes how historically unfair social categories are multiply encoded in any adequately rich data space. This occurs because many data points that appear neutral at first look (such as credit scores) are functionally proxies for histories of social, political and economic exclusion (such as redlining encouraged by federal housing policies). Additionally, the disparate sample sizes inherent in modeling minority and majority populations disadvantages the minority population when machine learning is used to make population-wide decisions.

An examination of how the new epistemological paradigms of data analytics lead to new ethical challenges.

PMID: 19197046
PMCID: PMC2745217.

Discussions of justice and big data often focus on the risks of inclusion in big data systems—police profiling, redlining, credit scores, etc. This paper examines the justice consequences of exclusion from those same systems. Some socioeconomically disadvantaged populations exist on the periphery of big data ecosystems and thus their preferences and needs are not adequately accounted for in machine learning. The author advocates for a legal/policy principle he names "data antisubordination," which seeks to ameliorate the risk that big data will create group disadvantages.

Information poverty occurs when localized conditions result in people not being able to attain desired information reliably, such as people who are homeless or incarcerated. The authors of this paper examine how social stigma around extreme body modification complicate information poverty. People who may not otherwise be "information poor" are not able to find necessary information about extreme body modification practices and safety because their community is structured around secrecy. This study shows how in an era of highly networked information tensions of group membership, secrecy and safety play out in terms of information access and poverty.

The database structures and user interfaces of social media platforms are centered around the sharing of information, including sharing information about other people. However, many of our assumptions about how privacy functions focus on the individual as the locus of control over information. The authors of this paper examine how teenagers use social media in a networked fashion and advocate for a symmetrical model of networked privacy that treats privacy as a distributed phenomenon.

This public letter represents the Council for Big Data, Ethics and Society’s perspective on proposed changes to the Common Rule. In particular, it focusses on the mistaken assumption that public datasets are inherently low-risk to the subjects of data science research.

00000.

00003.

00001
Because new auxillary datasets that can be used to potentially be used to re-identify data are constantly released, the risk of re-identification from ad-hoc techniques rises over time. This paper shows how the risks of re-identification are not only unknown, but unknowable. Due to this, the authors advocate for a precautionary approach to privacy protection, placing the burden of proof on the releasing party to account for potential risks. Such a precautionary approach uses policy to incentivize caution with data releases.

An essential contribution into data privacy theory, law and practice. This book is centered on the author’s theory of "contextual privacy," arguing that privacy is less about control over disclosure of information than about control over the flow of information.

Makes the case for a skeptical approach to big data analytics that cuts a middle path between boosterism that ignores scientific limits and cynicism that rejects demonstrable utility of big data. O’Neill identifies the pitfalls of trusting big data too much as: people get addicted to metrics as an easy route to actionable insight; too much focus on numbers, and not enough focus on the behaviors for which the numbers are a proxy; data analytics tends to allow incorrect or irrelevant framing of problems; and using datasets as proxies for real-world social/behavioral conditions enables perverse incentives for gaming the proxies. The pitfalls of not trusting data enough include: big data enables measurement of the value of business decisions with more specificity; big data resources puts "quants" inside of business strategy; big data enables discussion of negative business news without treating it as unwarranted skepticism; missing the feedback loop between models and society.

An examination of data sharing practices in astronomy, arguably the most data-intensive field of science. Using an extensive survey at a major research institution, the authors show that although astronomers are by and large philosophically supportive of data sharing, practical and infrastructural barriers limit the extent of sharing large datasets.

Proposes alternative principles and procedures for research ethics review where the Common Rule and the peculiar practices of IRB’s are not applicable.

This report is a collection of concrete examples of how big data intersects with important civil rights issues, such as credit and insurance access, predatory financial products, and employment background checks. It helpfully illustrates how the civil rights consequences of big data technologies are dependent on the political and legal context. For example, predictive policing could have more negative civil rights outcomes because the algorithms are considered proprietary under IP law and thus are not auditable by experts.

Useful pedagogical tools for teaching themes of technology ethics, including Internet and data ethics.

Despite our tendency to rely on formal laws and regulations to protect privacy rights, much of what actually protects privacy can be understood as latent structural constraints. These constraints operate somewhat like legal constraints by imposing excessive costs on the widespread breaking of implicit societal norms.

Examines the impact of big data methods on the insurance industry. While algorithmic/machine learning predictive methods may provide more accurate risk distributions in some contexts, it also risks using unexpected and/or irrelevant correlations to impose disparate costs on vulnerable populations and may excessively invade privacy.

On March 17, 2014, the Data & Society Research Institute, the White House Office of Science and Technology Policy (OSTP), and New York University’s Information Law Institute co-hosted "The Social, Cultural, & Ethical Dimensions of "Big Data."" The purpose of the event was to convene key stakeholders and thought leaders from across academia, government, industry, and civil society to examine the social, cultural, and ethical implications of "big data," with an eye to both the challenges and opportunities presented by the phenomenon.

This report offers a framework for understanding ethical issues in social media research and provides a helpful decision tree for determining what obligations researchers have toward the subjects of data-intensive social science.

This classic essay in the philosophy of technology argues that scholars should not assume technological artifacts are neutral in a sociopolitical sense. Rather, technology always embodies its sociopolitical context and scholarship about technology needs to account for that context.

http://wwtstories.org/BonesOfTheMilkyway/bonesofthemilkyway.html
A collaborative astronomy project that uses big data techniques to distribute scientific inquiry.

"Privacy" and "utility" are typically portrayed as having an inverse relationship in big data applications: decisions to increase privacy will typically decrease the utility of the dataset, and vice versa. This paper shows that much of this interpretation relies on the operationalization of "privacy" and "utility" within computer science experiments, which varies widely across the field.