The Control and Intelligent Systems Lab studies artificial intelligence through the lens of control theory: treating generative models as dynamical systems that can be understood, principled, and controlled with the same rigor and reliability we have in other engineering disciplines.
At the Control and Intelligent Systems Lab, our research lies at the intersection of control theory, optimization, and artificial intelligence with a focus on the principled design of reliable and controllable AI technologies. Our work adapts mathematical principles from control theory, traditionally used to ensure safety and predictability in engineered systems. This enables us to understand, control, and ultimately improve the behavior of AI systems (spanning language applications, embodied systems, and swarms of agents), so we can live in a world where AI systems are controllable: they reliably do what we want them to do, and they never do what we don't want them to do.
We are very proud to be an academic lab at UC Berkeley, a public university. Beyond research and teaching excellence, our ultimate goal is to serve society by educating the public, and to train elite engineers so that they may go on to contribute to the public good. In that spirit, we publish openly on arXiv, always share our code, and upload video summaries of our papers to increase the accessibility of our work. We also engage in collaborations and advising with public and private sector partners, take on service to the academic community, and regularly participate in outreach.
