Imagine an AI agent working side-by-side as your programming assistant. Or picture one that helps you whip up a delicious dinner.
Pop culture has given us a variety of futuristic assistive devices that could step into these techno-utopian roles. Think Baymax from Big Hero 6. Or C-3PO from Star Wars. Or Rosey the Robot from The Jetsons.
At the University of California San Diego, AI agents are making the leap from fiction to fact in a new computer science lab dubbed PEARLS (Pragmatically Exploring Agents with Reinforced Languages). The lab is led by Prithviraj (Raj) Ammanabrolu, an assistant professor who joined the Jacobs School of Engineering’s Department of Computer Science and Engineering (CSE) in 2024.
Ammanabrolu and his student researchers are creating trustworthy and responsible AI agents that can communicate and collaborate with humans in natural language and learn how to accomplish day-to-day tasks efficiently.
“To achieve these goals, we have a relatively interdisciplinary lab that works on everything from the core machine and reinforcement learning algorithms to cognitive science and human-centered design of AI systems,” said Ammanabrolu.
Ultimately, the researchers aim to imbue AI agents with the ability to align themselves to human preferences after getting feedback. This feedback comes in two forms: humans that tell AI agents how to improve and environments that allow them to observe the effects of their actions. As part of their strategy, the team is building and using neurosymbolic world models that guide the agent’s actions.
Ammanabrolu is also affiliated with the startup tech company, MosaicML, which was acquired by Databricks in 2023. As a part-time researcher for them, he is leading an effort to scale Reinforcement from Human Feedback (RLHF), a set of techniques that shapes how Large Language Models (LLMs) learn from large populations of people, especially in industrial enterprise scenarios.
Like many who move to San Diego, Ammanabrolu couldn’t resist the triple allure of beach, sun, and great food. He was also drawn by the university’s collaborative environment – a necessity for his interdisciplinary research.
“My hometown is a beach city in India, and the noise of the ocean waves helps me think,” said Ammanabrolu. “It also helped that of all the universities I interviewed, the CSE faculty were the most chill, and the PhD students seemed relatively much happier.”
Ammanabrolu has been in San Deigo for roughly a year. From sparring at a kickboxing gym to walking in scenic Balboa Park, he seems to have found his footing. And it often comes with sand between his toes.
“Seriously, some days I just get a chair and go write my grant proposals during the day at Windansea Beach with a nice order of boba,” he said.
Where to grab pearl tea on the way to PEARLS lab? Yun Tea House is Ammanabrolu’s current favorite. Maybe soon, he will entrust an AI agent with his go-to order: a Monkey King with 50% sugar, 0% ice, and, of course, boba.
--By Kimberley Clementi