Seminar: From Rulesets to Reasoning: Building More Thoughtful Devices in the Age of Generative AI
Evan King
Applied Research Scientist
Useful Sensors
Tuesday, March 25
10:30 - 11:30AM
1100 Torgersen Hall
Abstract
Despite promises of more flexible user interactions, natural language interfaces for smart devices often fail to meet user goals when those goals deviate from rigid rulesets. A typical smart assistant cannot interpret a request to “make it cozy in here”, nor can a smart thermostat explain “why the heat turned off again”. These limitations stem from an inability to reason about devices’ internal states and capabilities as they relate to user goals. However, the emergence of large language models is rapidly reshaping this landscape. Driven by recent advances in generative AI, this talk advocates for a paradigm shift from rigid rulesets to flexible on-device reasoning to drive development of smarter, more “thoughtful” devices. The talk will explore the new research opportunities and challenges that emerge when everyday devices are empowered to generate their actions and explanations dynamically rather than relying on preconfigured rules.
Biography
Evan King is an Applied Research Scientist at Useful Sensors, a startup from the creators of TinyML and TensorFlow Lite at Google. His work focuses on the development of real-time, on-device speech recognition and goal-oriented reasoning for user interfaces. He holds a PhD in ECE from the University of Texas at Austin, where his dissertation research introduced some of the first smart devices and spaces powered by on-device generative AI. He has previously held research positions at the Air Force Research Labs and Toyota.