UC San Diego has launched the Center for Principled Agentic Intelligence, a new interdisciplinary research center focused on AI alignment, human oversight, accountability, and independent evaluation of increasingly autonomous artificial intelligence systems.
The center launches with $1 million in support over three years for its initial research initiatives, along with $250,000 in shared computing resources provided by the Halıcıoğlu School of Data Science and Computing.
Known as CPAi, the center is a partnership between UC San Diego’s Halıcıoğlu School of Data Science and Computing and the School of Arts and Humanities.
Researchers from data science, computer science, high-performance computing, philosophy, arts, and humanities will collaborate on questions surrounding increasingly capable AI agents.
Unlike conventional AI systems primarily designed to respond to individual prompts, agentic AI can plan and execute multiple actions in pursuit of broader objectives.
CPAi will organize its research around three principles: alignment, trust, and responsibility.
Alignment will examine whether systems behave according to the intentions of developers and users. Trust will focus on obtaining verifiable evidence that systems perform reliably. Responsibility will examine accountability for the people and organizations creating and deploying AI technologies.
The center will initially operate two major initiatives.
AI Alignment and Society, co-led by Mikhail Belkin and Eddy Keming Chen, will examine how AI training and evaluation affect system behavior and how alignment can be tested as systems become more capable.
Researchers will also examine how meaningful human oversight can be maintained and what independent evaluation methods may be necessary.
A second initiative, Agentic AI System Evaluation, will develop new methods for independently assessing the reliability and performance of AI agents.
The team plans to create an open-source testbed capable of simulating real-world tasks and analyzing why an agent fails, whether it stays within defined boundaries, whether its behavior can be audited, and how reliably it completes complex multi-step assignments.
The center ultimately wants to make rigorous AI evaluation tools available beyond the companies developing the systems themselves.
KEY QUOTES:
“As artificial intelligence becomes increasingly capable and autonomous, universities have an important role to play in ensuring that its development serves people and society.”
Pradeep K. Khosla, Chancellor Of UC San Diego
“The Center for Principled Agentic Intelligence builds on that legacy by bringing together researchers who can examine not only what increasingly capable AI systems can do, but how we understand them, evaluate them, and ensure they remain responsive to human goals.”
Rajesh K. Gupta, Dean Of The Halıcıoğlu School Of Data Science And Computing
“The humanities are essential to answering these questions and to ensuring that AI evolves in ways that reflect our values and serve the common good.”
Cristina Della Coletta, Dean Of The School Of Arts And Humanities
