We drive policy outcomes and spark debate through reports, scholarly articles, regulatory comments, direct advocacy efforts together with coalition allies, articles and op-eds, and participation in events including public panels, conferences, and workshops. Here's a selection of our recent work.
In a paper presented at the 2020 Conference on Fairness, Accountability, and Transparency in Machine Learning, we describe how and when private companies collect or infer sensitive attribute data, such as a person’s race or ethnicity, for antidiscrimination purposes.
Our empirical research showed that Facebook’s “Special Audiences” ad targeting tool can reflect demographic biases. More broadly, we provide experimental proof that removing demographic features from a real-world algorithmic system’s inputs can fail to prevent biased outputs.
* Northeastern University
We co-led the Pretrial Risk Managment Project of the John D. and Catherine T. MacArthur Foundation. As part of this project, we published a critical issue brief on pretrial risk assessment, which focused on two questions. First: Why do many in the civil rights community oppose the use of pretrial risk assessment instruments? Second: What concrete reform strategies are available that would avoid risk assessment instruments, or would sharply limit their role?
Our empirical research showed that Facebook’s ad delivery algorithms effectively differentiate the price of reaching a user based on their inferred political alignment with the advertised content, inhibiting political campaigns’ ability to reach voters with diverse political views.
* Northeastern University, ** University of Southern California
Our empirical research showed that Facebook itself can skew the delivery of job and housing ads along race and gender lines, even when advertisers target broad audiences.
* Northeastern University, ** University of Southern California
Without active measures to mitigate them, bias will arise in predictive hiring tools by default. This report describes popular tools that many employers currently use, explores how these tools affect equity throughout the entire hiring process, and offers reflections and recommendations on where we go from here.
In the first rigorous, independent evaluation of Facebook’s new ad transparency plans, we urge the company to improve its ad transparency tools to enable meaningful public scrutiny.
Automated decisions are increasingly part of everyday life, but how can the public scrutinize, understand, and govern them? This Upturn and Omidyar Network report maps out the landscape, providing practical examples and a framework to think about what has worked.
Bail reform is rapidly underway. But at the same moment that jurisdictions work to reduce the true risks of pretrial release through reform policies, jurisdictions across the country are also adopting statistical tools that will blindly predict such risks remain as high as ever. This forthcoming article charts how jurisdictions can avoid making costly errors in their adoption of pretrial risk assessment tools.
This framing paper, prepared for the NetGain Partnership, explores how automated decisions are shaping the lives of vulnerable people and groups, and offers suggestions and direction for interested funders and the broader social sector.
The encryption debate is generally framed as a struggle between civil liberties and national security. We partnered with Consumer Reports to shed light on why encryption is critical for consumers’ safety and well-being.
Together with the Leadership Conference, Upturn releases the latest version of our scorecard that evaluates the police body-worn camera policies in 75 major U.S. cities. It continues to show a nationwide failure to protect the civil rights and privacy of surveilled communities.
Today, most major police departments that use body-worn cameras allow officers unrestricted footage review. This report explains why police departments must carefully limit officers’ review of body-worn camera footage, and calls for “clean reporting” to be adopted by all police departments.
This article draws lessons primarily from the domain of criminal justice, to illustrate three structural challenges that can arise whenever law or public policy contemplates adopting predictive analytics as a tool. It then offers some ideas for solutions.
Drawing on computer science expertise, we propose a new governance strategy, using cryptography to prove that a decision is rule-bound and correct, even when the decision comes from a “black box” that is secret or is too complex for direct human inspection.
In a report for the Open Society Foundations, we review the different types of brokerage and profiling products sold by data brokers, survey the relevant legal landscape, and recommend an impact-driven, bottom-up approach to further investigation of data-driven profiling by data brokers.
This project maps the ways that data at scale may pose risks to philanthropic priorities and beneficiaries, identifies key questions that funders and grantees should consider before undertaking data-intensive work, and offers recommendations for funders to address emergent data ethics issues.
We find that at least 20 of the nation’s 50 largest police forces have used a predictive policing system, with at least an additional 11 actively exploring options to do so. Vendors shield the technology in secrecy, and informed public debate is rare. Early research findings suggest that these systems may not actually make people safer.
Together with the Leadership Conference, Upturn releases a scorecard that evaluates the police body-worn camera policies in 50 major U.S. cities. It shows a nationwide failure to protect the civil rights and privacy of surveilled communities.
A technical assessment of the present and potential future monitoring capabilities available to internet service providers.
We explain how online lead generation works, describe the risks and legal complexities specific to lead generation for online payday loans, document the widespread use of search ads by payday lead generators, and recommend interventions.
A “missing manual” for policy professionals seeking to better understand technology’s impact on financial underwriting and marketing.
How and where, exactly, does big data become a civil rights issue? This report begins to answer that question, highlighting key instances where big data and civil rights intersect.
Users in China can’t freely explore the Internet because of the regime’s “Great Firewall.” But special software tools—when they work—can help users around those barriers. We proposed a new approach to developing circumvention tools, a strategy called “collateral freedom.”