Keynote Speaker
August 3 (Monday)
Dr. Larry Heck is a Professor with a joint appointment in ECE and Interactive Computing at the Georgia Institute of Technology. He holds the Rhesa S. Farmer Advanced Computing Concepts Chair and is a Georgia Research Alliance Eminent Scholar. From 2017-2021, he was the CEO of Viv Labs and SVP at Samsung where he led Bixby for North America. From 2014-2017, he was a Principal Scientist at Google, leading an advanced Dialogue effort in Google Research. From 2009-2014, he was the Chief Scientist of the Microsoft Speech products team and later a Distinguished Engineer in Microsoft Research. In 2009, he was a co-founder of Microsoft’s Cortana personal assistant. From 2005-2009, he was Vice President at Yahoo! responsible for Search and Advertising quality. From 1998-2005, he was with Nuance Communications and served as Vice President of R&D. He began his career as a researcher at the Stanford Research Institute (1992-1998) initially in acoustics and later in speech research with the Speech Technology and Research (STAR) Laboratory. Funded by the US government’s NSA and DARPA, his SRI Speaker Recognition team was the first to successfully create large-scale deep neural network (DNN) deep learning technology in the field of speech processing (1998 National Institute of Standards and Technology Speaker Recognition evaluation) and the first to deploy a major industrial application of deep learning.
Dr. Heck received the PhD in Electrical Engineering from the Georgia Institute of Technology in 1991. He is a Fellow of the IEEE, published numerous scientific papers and holds 50+ United States patents. He was inducted into the Academy of Distinguished Engineering Alumni at Georgia Tech, received the Distinguished Engineer Award from the Texas Tech University Whitacre College of Engineering, and was inducted as a Fellow in the National Academy of Inventors in 2025.
The Conversation Inside the Model: Probing and Controlling Interactional Structure in Multimodal Foundation Models
Conversational AI has become strikingly fluent, yet most systems are still optimized to produce a plausible next utterance rather than to manage how an interaction unfolds over time. Human conversation, by contrast, is a jointly coordinated, stateful process: participants continuously track where the exchange stands and shape it through finely timed interactional behavior. Decades of work in conversation analysis describe this machinery in detail, but it has rarely been connected to the internal mechanics of the foundation models now driving spoken and multimodal agents. In this talk I take the view that conversation can be modeled as a controlled, stateful dynamical system, and I ask a question that is increasingly testable: do pretrained multimodal foundation models already represent interactional structure in their internal states, and if so, can we recover that structure and steer it? I will outline a research program that treats interactional competence as something to be identified, observed, and causally controlled inside these models, present early evidence from work in progress, and discuss what controllable conversational state could mean for the next generation of digital humans. This is an active effort, and I will share our most recent findings.