Online Speech, Computational Fact Checking & Ethics in Computer Science
[Return to Nexus Newsletter]
By Kaitlin Barnes
For those who work with data, big or small, the question often begins as, how much of my data is valuable?
The consensus is simple: we’re drowning in data. Over the past few decades, and predominately spurred by the modern Internet, information has become extremely democratized and accessible. As we move towards a world where news both breaks and is disseminated across social media sites, such as Twitter, there’s an even greater need for detecting data credibility in real time. With the massive amounts of data available, algorithms are increasingly relied upon to make sense of that information. This growing reliance is generating challenging questions about the “fairness” of these algorithmic solutions.
Earlier this year, the Hariri Institute for Computing co-sponsored the Journalism & the Search for Truth in an Age of Social Media conference, which was hosted by BU’s College of Communication and chaired by James Katz, Director of the Division of Emerging Media Studies and Juliet Floyd, Professor of Philosophy. The conference analyzed “social media’s interaction with journalism and democracy…from philosophical, ethical, practical and political perspectives.” As part of the Hariri Institute’s sponsorship, Institute Director Azer Bestavros contributed to program development and invited Giovanni Luca Ciampaglia, a Research Scientist at Indiana University, to partake in the conference’s Trolling, Computer Moderation, and Algorithms panel session.
On the heels of that conference, the Institute hosted an encore Data Science Initiative (DSI) Colloquium and Roundtable to spur additional dialog on the opportunities to address real world problems through computational tools. The discussion dove into the convergence of online political speech, computational fact checking, and the evolving role of ethics in computer science. The event featured presentations by Giovanni Luca Ciampaglia and Brian Ulicny, Senior Director of Data Science for Thomson Reuters, who focused on fact checking tools and real-time assessment of information credibility.
The presenters were then joined by BU faculty members, John Byers (Computer Science), Spencer Piston (Political Science), Nancy Ammerman (Sociology), Mark Crovella (Computer Science), Ran Canetti (Computer Science, RISCS Center), Dylan Walker (Information Systems), Katz, Floyd, and Bestavros. Participants came into the discussion with research interests that span from how online reputational ratings influence commerce, to how racial and class-based prejudice influence policy opinion, to how people use technology on individual and societal levels. They raised questions about how the field of computer science can play a larger role in framing problems, what does it mean for an algorithm to be fair, and how to address bias with computational tools.
As the discussion turned to the problematic trend of increasing opacity of computational methods, the group grappled with the complexities associated with making algorithms completely transparent. Crovella highlighted how these problems are compounded by unintended consequences that arise from scaling up solutions as the Internet takes on the role of “an accelerator of social sorting.” To tackle these questions, Floyd and Ammerman suggested a need for computer-related solutions to incorporate theories and perspectives from social sciences in ways that allow for accessible explanations of algorithmic outputs. Walker proposed that a diversification of methods could potentially mitigate the risk of bias arising from reliance on a sole algorithm. Canetti and Byers added a layer of intricacy to the fray by tying in the threat of adversarial behavior and the need for computational tools to also address societal issues of security.
This discussion is part of ongoing work at the Institute involving the Center for Reliable Information Systems & Cyber Security (RISCS), BU/MIT Technology & Cyberlaw Clinic, and faculty within various academic departments such as Philosophy and Political Science. By leveraging BU’s disciplinary breadth, the Institute hopes to cultivate cross-disciplinary collaborations that expand the frontier of research on these complex topics.
Looking forward, the Institute is planning a spring 2018 symposium, slated to address algorithmic accountability. Conference organizers Juliet Floyd, Ran Canetti, and Azer Bestavros are working to bring together experts from a wide range of disciplines to further explore what constitutes algorithmic fairness and how to be ethically responsible in the development and usage of computational tools.
One thing this ongoing discussion makes clear: there will always be a need for human interaction and interpretation of data, no matter how sophisticated our computational tools become.