Powerful advances in machine learning over the past decade have brought with them explosive growth in the use of computational decision-making tools. These tools have a direct impact on hundreds of millions of lives. Yet they are often opaque, poorly understood, and impossible to hold to account. As algorithmic decision-making systems become increasingly ubiquitous, and our reliance upon them increasingly widespread, it is crucial to develop an understanding of how fairness, justice, and value considerations apply to these tools.
At JFI, we believe this must be a collaborative enterprise, engaging the technical knowledge and lived experience of a diverse range of academic experts, government policymakers, community advocates, and the general public. Our work focuses on leveraging our technical and subject-matter expertise to enable this broad spectrum of voices to meaningfully participate in informed debate on these unprecedented ethical, social, and political questions.
JFI is advancing this development in three areas.
If the public, lawmakers, administrators, and judges are ill-equipped to understand automated decision tools, then widespread use of these tools presents fundamental challenges to principles of democratic oversight and government accountability. Therefore, as these systems are used more and more, it is of critical importance to develop explanations of these systems that are both accessible to general audiences and helpful for evaluating their use: public education is indispensable. JFI is producing both text-based and interactive explanations that will start to fill this need.
Alongside this work in public education, we are promoting a more robust and comprehensive treatment of digital ethics and governance within higher education. Supporting promising new researchers in the field, we build on their expertise to develop course materials that can be adapted for use in a broad range of fields, from computer science to social work. At a structural level, JFI is building partnerships and infrastructure to facilitate cross-disciplinary engagement around these questions, so that thinkers from a broad base of disciplines – computer science, data science, and statistics, of course, but also philosophy, history, Africana studies, and law – can more easily combine their knowledge and experience.
Developing good public policy around automated decision systems is a daunting prospect: it requires a grounding not just in machine learning and statistics, but also in municipal history, administrative law, and decision theory. In addition, it requires input and thought from a huge range of people: people whose lives will be affected by these systems, people who work at the agencies implementing them, public servants, community leaders, and ordinary citizens. JFI works closely with governments and partner organizations to provide the background knowledge and frameworks they need. On the technical side, we are developing prototype tools that these bodies can use to more effectively evaluate and more equitably implement such systems: novel feature detection algorithms, generative models, and decision support tools that can take user-specified values, goals, and constraints into account.
work / single.html