Awards

“State-of-the-art AI based on large language models (LLMs) and vision-language models (VLMs) exhibits unprecedented abilities but is poorly understood,” says Briger Family Distinguished Associate Professor of Computer Science, Associate Professor of Linguistics, and Associate Chair of Computer Science Ellie Pavlick. “As a result, even models with ‘guardrails’ exhibit unpredictable behavior, making them risky to deploy in applications where human understanding and control is paramount.”

In response, Ellie and her collaborators are conducting new research to enable us to understand and control so-called “black box” AI by creating tools that inspect, diagnose, and manipulate high-level algorithms. It’s multidisciplinary work, leveraging insights from cognitive science, neuroscience, and philosophy, and it’s just won her a Defense Advanced Research Projects Agency (DARPA) Young Faculty Award (YFA). 

The objective of the YFA program is to identify and engage rising research stars in junior faculty positions at American academic institutions. It provides funding, mentoring, and industry and DoD contacts to awardees early in their careers, focusing on untenured faculty and emphasizing those without prior DARPA funding. The long-term goal of the YFA program is to develop the next generation of academic scientists, engineers and mathematicians.

“What’s different about our approach,” says Ellie, “is that we’re going to harness advances in reverse engineering these systems. Most existing methods seek to change model behavior while continuing to treat the model itself as a black box. In contrast, our approach could enable auditing and control of how models behave by directly inspecting and editing the numerical parameters that define their structure and logic.”

Also a Research Scientist at Google Deepmind, Ellie leads the Language Understanding and Representation (LUNAR) Lab, which seeks to understand how language “works” and build computational models that can understand language the way that humans do. Their projects focus on language broadly construed and often include the study of capacities more general than language, including conceptual representations, reasoning, learning, and generalization. 

Recently, Ellie was interviewed on television newsmagazine 60 Minutes, received Brown’s Early Career Research Award, took part in a new AI discussion series hosted by Brown’s Office of the Provost, and was awarded a grant for her work with language acquisition and information retrieval that set a Brown CS record. She was a keynote speaker at the 2019 International Conference on Computational Semantics (IWCS) and gave invited talks at the 2019 New England Machine Learning and Harvard Linguistics Universals Colloquium Series. 

Ellie joins Brown CS faculty member Stefanie Tellex, who received a DARPA Young Faculty Award in 2015.

For more information, click the link that follows to contact Brown CS Communications Manager Jesse C. Polhemus.