Uploaded February 2026 | Updated September 2026, 3 hours ago
Achieving a powerful synergy in human-AI collaboration requires extreme versatility in agent capabilities, both at the level of generalization across modalities, languages, tasks, and domains in its own reasoning and problem solving, as well as in its role-taking and interactive strategy enactment in collaboration. This talk first explores model generalization from the inside out, building on insights into the complementary affordances of disparate data sources as they are transformed through a learning process, and then turning outwards to application across five tasks, including event ordering, question answering, form understanding, multi-lingual relation extraction, and reasoning path prediction. The concept of learning from contrasting cases is demonstrated to facilitate easy-to-hard generalization by fostering more nuanced instruction following skills. The concept of scaffolding is also used to illustrate how representations of reasoning may extend model capabilities beyond a natural frontier, either through augmentations injected into the learning process or as part of inference-time interaction with a human or agent partner or with the environment. Extending 3 decades of research into AI-supported collaborative work, the talk ends with a discussion about work-in-progress extending model generalization research into an agentic setting where intelligent collaborative agents participate in collaboration with another agent, a human collaborator, or a group of humans.
Dr. Carolyn Rosé is the Kavcic-Moura Professor of Language Technologies and Human-Computer Interaction in the School of Computer Science at Carnegie Mellon, Program Director for the Masters of Computational Data Science Program, and the Language Technologies Institute undergraduate programs. Her research group’s highly interdisciplinary work, published in nearly 350 peer-reviewed publications, is represented in the top venues of 5 fields: namely, Language Technologies, Learning Sciences, Cognitive Science, Educational Technology, and Human-Computer Interaction, with awards in 4 of these fields. She is a Past President and Inaugural Fellow of the International Society of the Learning Sciences, Senior member of IEEE, Founding Chair of the International Alliance to Advance Learning in the Digital Era, and Executive Editor (formerly Co-Editor-in-Chief) of the International Journal of Computer-Supported Collaborative Learning. She also serves as a 2020-2021 AAAS Leshner Leadership Institute Fellow for Public Engagement with Science, with a focus on public engagement with Artificial Intelligence.
Achieving a powerful synergy in human-AI collaboration requires extreme versatility in agent capabilities, both at the level of generalization across modalities, languages, tasks, and domains in its own reasoning and problem solving, as well as in its role-taking and interactive strategy enactment in collaboration. This talk first explores model generalization from the inside out, building on insights into the complementary affordances of disparate data sources as they are transformed through a learning process, and then turning outwards to application across five tasks, including event ordering, question answering, form understanding, multi-lingual relation extraction, and reasoning path prediction. The concept of learning from contrasting cases is demonstrated to facilitate easy-to-hard generalization by fostering more nuanced instruction following skills. The concept of scaffolding is also used to illustrate how representations of reasoning may extend model capabilities beyond a natural frontier, either through augmentations injected into the learning process or as part of inference-time interaction with a human or agent partner or with the environment. Extending 3 decades of research into AI-supported collaborative work, the talk ends with a discussion about work-in-progress extending model generalization research into an agentic setting where intelligent collaborative agents participate in collaboration with another agent, a human collaborator, or a group of humans.
Dr. Carolyn Rosé is the Kavcic-Moura Professor of Language Technologies and Human-Computer Interaction in the School of Computer Science at Carnegie Mellon, Program Director for the Masters of Computational Data Science Program, and the Language Technologies Institute undergraduate programs. Her research group’s highly interdisciplinary work, published in nearly 350 peer-reviewed publications, is represented in the top venues of 5 fields: namely, Language Technologies, Learning Sciences, Cognitive Science, Educational Technology, and Human-Computer Interaction, with awards in 4 of these fields. She is a Past President and Inaugural Fellow of the International Society of the Learning Sciences, Senior member of IEEE, Founding Chair of the International Alliance to Advance Learning in the Digital Era, and Executive Editor (formerly Co-Editor-in-Chief) of the International Journal of Computer-Supported Collaborative Learning. She also serves as a 2020-2021 AAAS Leshner Leadership Institute Fellow for Public Engagement with Science, with a focus on public engagement with Artificial Intelligence.
