TLDR: A forthcoming talk at the Berkman Klein Center will feature MIT’s Jacob Andreas discussing how language models can better represent their knowledge and uncertainty, with a focus on coherence and calibration. The event, moderated by Berkman Klein’s Chief AI Scientist Josh Joseph, will delve into the societal and governance implications of AI’s responsible use.
The Berkman Klein Center for Internet & Society at Harvard University is set to host a significant discussion titled ‘Belief, Uncertainty, and Truth in Language Models’. This event aims to tackle fundamental questions surrounding the capabilities of advanced AI, specifically what it means for a language model to ‘know’ something and how it should convey its level of uncertainty to human users.
Leading the discussion will be Jacob Andreas, an Associate Professor of Electrical Engineering and Computer Science at MIT. Andreas, whose research at MIT’s EECS and CSAIL, and previous PhD work at the Berkeley NLP Group and Berkeley AI Research Lab, focuses on the computational foundations of language learning, will present novel approaches to developing language models. These models are designed not only to interpret the world but also to introspect and model their own understanding. His methodology emphasizes optimizing for ‘coherence and calibration’ beyond mere accuracy, aiming for models that are both more factually consistent and more reliable in expressing their confidence levels.
Moderating this crucial conversation will be Josh Joseph, the Berkman Klein Center’s Chief AI Scientist. Joseph, holding the first role of its kind at BKC, is at the forefront of initiatives exploring the agency of AI systems, developing new benchmarks beyond traditional ‘intelligence’ metrics, and building internal AI research capabilities. He also serves as a Visiting Scientist at MIT and a Lecturer on Law at Harvard Law School.
The discussion is expected to bridge the gap between cutting-edge technical advancements and broader societal concerns. It will explore critical governance questions, including the ethical standards that should guide how large language models present information and express reliability or doubt. The conversation will conclude by examining the policy implications arising from these complex issues, navigating the intersection of computer science and its impact on society.
Also Read:
- Montclair State’s CHSS Initiates Year-Long AI Distinguished Speaker Series
- Organizations Redefine Governance to Mitigate AI Risks and Drive Prosocial Innovation
The event is scheduled for Wednesday, October 1, 2025, from 12:30 PM to 1:30 PM ET. It will take place at 1557 Massachusetts Ave., Multipurpose Room, 5th Floor, Cambridge, MA 02138 US. The announcement was last updated on September 22, 2025.
