Schedule

Welcome from the organizers

Katherine Lee (The GenLaw Center), Hoda Heidari (K&L Gates Initiative at Carnegie Mellon University), Alexandra Givens (Center for Democracy and Technology), Paul Ohm (Georgetown Law Center)

Bio: Zachary Lipton is the Chief Technology Officer and Chief Scientist of Abridge and the Raj Reddy Associate Professor of Machine Learning at Carnegie Mellon University. At CMU, he directs the Approximately Correct Machine Intelligence (ACMI) lab, whose research focuses include the theoretical and engineering foundations of robust and adaptive machine learning algorithms, applications to both prediction and decision-making problems in clinical medicine, natural language processing, and the impact of machine learning systems on society. A key theme in his current work is to take advantage of causal structure underlying the observed data while producing algorithms that are compatible with the modern deep learning power tools that dominate practical applications. He is the founder of the Approximately Correct blog (approximatelycorrect.com) and a co-author of Dive Into Deep Learning, an interactive open-source book drafted entirely through Jupyter notebooks that has reached millions of readers. He can be found on Twitter (@zacharylipton), GitHub (@zackchase), or his lab's website (acmilab.org).

Catered by TATTE

Bio: Emily Lanza is Senior Counsel for Policy & International Affairs at the U.S. Copyright Office, where her portfolio includes work on the Office’s AI Initiative and international copyright developments in the Asia Pacific region. She earned a JD from Georgetown University Law Center, a MPhil in Archaeological Heritage from the University of Cambridge, and a BA in Anthropology from the University of Pennsylvania. Prior to her legal career, she held several curatorial and public service positions at libraries and museums in the United States and the United Kingdom.

Abstract: Today’s generative AI models hold a considerable amount of power and have been built off of original content which comes from artists and creators. To build an ethical framework around generative AI, it is imperative to prioritize transparency, accountability, and inclusivity throughout the development, deployment, and utilization phases to ensure that the technology serves the collective good while respecting fundamental human rights and values. This presentation will cover the practices that Shutterstock has put in place, provide examples of other types of shared success models, and contemplate quality of content as the next priority in artist compensation.

Bio: Sejal Amin is the Chief Technology Officer at Shutterstock. She is a senior technology executive and product development expert with over 25 years of experience leading large teams through cultural, operational and technology transformation for SaaS initiatives. In her role as CTO, Sejal is responsible for delivering a technology vision and strategy that transforms Shutterstock's technology platform to deliver a new and unparalleled experience to customers and contributors.

Recently, she was Chief Product and Technology Officer for Khoros, a Vista Portfolio company where she was integrating a distributed Product and Technology organization while defining a Product and Operational strategy to execute on the company's vision and growth goals. Just prior to that, she was CTO for the Thomson Reuters Tax and Accounting Tax Professionals Business. She has a wide range of technology leadership experience across several business units at Thomson Reuters managing global product development teams and portfolios of growing size and complexity for 15+ years. Over the years, she led several enterprise-wide transformation initiatives focused on project to product transitions, organizational transformations, technology portfolio redesigns, and building high performance product development cultures to keep pace with a rapidly changing technology environment.

Sejal was recognized on the BT150 Transformational Executive List in 2021 and has a number of professional associations including Board Director for the Value Stream Consortium.

Preliminary abstract: Unlearning is the problem of removing knowledge from an AI system after it has been trained, without retraining the whole AI from scratch. Unlearning is a major technical challenge that remains an open subject of research: we will survey the state-of-the-art in unlearning research, including the major approaches to unlearning, the ways that researchers measure the quality of unlearning methods, and the long-term challenges facing scientists and practitioners who need to erase specific knowledge from an AI.

Bio: David Bau is an assistant professor at Northeastern University Khoury school of Computer Sciences. He is a pioneer on deep network interpretability and model editing methods for large-scale AI such as large language models and image synthesis diffusion models. He is leading an effort to create a National Deep Inference Fabric.

Bio: Professor Solow-Niederman’s scholarship sits at the intersection of law and technology. Her research focuses on how to regulate emerging technologies, such as artificial intelligence, in a way that reckons with social, economic, and political power. With an emphasis on algorithmic accountability, data governance, and information privacy, Professor Solow-Niederman explores how digital technologies can both challenge longstanding regulatory approaches and expose underlying legal values.

Professor Solow-Niederman’s work has been published or is forthcoming in the Harvard Journal on Law & Technology, the Northwestern University Law Review, the Southern California Law Review, and the Berkeley Technology Law Journal, among other law reviews and peer-reviewed journals. Her piece on data breaches was selected as a winner of the 2017 Yale Law Journal Student Essay Competition. Professor Solow-Niederman is a member of the Electronic Privacy Information Center (EPIC) Advisory Board. She is also a faculty affiliate at Harvard University’s Berkman Klein Center for Internet & Society and a visiting fellow at the Yale Law School Information Society Project, where she has worked with the Media Freedom and Information Access Clinic on a series of FOIA requests concerning state government use of AI.

Abstract: Differential privacy is the gold standard of computing statistics on private tabular data, creating plausible deniablity for each record contributor. But what is differential privacy not suitable for, and what could go wrong when it is applied in generative AI? In this talk we want to shed light on what differential privacy is, and what it is not by clarifying common misconceptions. We will then discuss how differential privacy comes at a cost to utility and fairness, and how applying it to different data modalities can impact the provided guarantees.

Bio: Niloofar Mireshghallah is a post-doctoral scholar at the Paul G. Allen Center for Computer Science & Engineering at University of Washington. She received her Ph.D. from the CSE department of UC San Diego in 2023. Her research interests are Trustworthy Machine Learning and Natural Language Processing. She is a recipient of the National Center for Women & IT (NCWIT) Collegiate award in 2020 for her work on privacy-preserving inference, a finalist of the Qualcomm Innovation Fellowship in 2021 and a recipient of the 2022 Rising star in Adversarial ML award.

Panelists: Sejal Amin, David Bau, Alicia Solow-Niederman, Niloofar Mireshghallah, Kat Walsh

[questions]