Center for Human-Compatible AI

The Center for Human-Compatible Artificial Intelligence (CHAI) was launched on August 29, 2016, at the University of California, Berkeley, under the leadership of Stuart Russell, who holds the Smith-Zadeh Chair in Engineering. CHAI was established with a founding grant of $5.5 million from the Open Philanthropy Project, with the goal of ensuring that increasingly capable AI systems remain beneficial to humanity. CHAI is a multi-institution research group headquartered at UC Berkeley with faculty principal investigators at Cornell University, the University of Michigan, and Princeton University. The center's co-principal investigators include computer scientists Pieter Abbeel, Anca Dragan, Bart Selman, Joseph Halpern, Michael Wellman, and Satinder Singh Baveja, as well as cognitive scientists Tom Griffiths and Tania Lombrozo. Mark Nitzberg serves as Executive Director. The center's research is organized around several core themes: foundations of rational agency and causality, value alignment and inverse reinforcement learning, human-robot cooperation, multi-agent perspectives and applications, models of bounded or imperfect rationality, and models of human cognition. A signature contribution is Cooperative Inverse Reinforcement Learning (CIRL), which formalizes the value alignment problem as a two-player game where the AI must learn human preferences through interaction rather than having objectives specified in advance. CHAI has produced hundreds of research publications since its founding, with papers accepted at premier conferences including NeurIPS, and has trained approximately 30 PhD students in AI safety. Alumni have been placed at leading institutions including Stanford, Princeton, MIT, and DeepMind. The center hosts approximately seven interns annually and convenes an annual workshop bringing together around 250 researchers, practitioners, and policymakers. Beyond technical research, CHAI engages in significant policy advocacy. Stuart Russell authored the popular science book Human Compatible: Artificial Intelligence and the Problem of Control (2019), delivered a widely viewed TED talk on three principles for creating safer AI, and has spoken at the World Economic Forum in Davos and the Nobel Week Dialogues. Russell co-created the Slaughterbots video campaigning against autonomous weapons, which has been viewed over 70 million times and won a Gold Medal at the Cannes Corporate Media and TV Awards. He holds independent observer status in the Global Partnership on AI and has testified before the U.S. Senate on AI regulation. CHAI's sponsors include Open Philanthropy, the Future of Life Institute, the Leverhulme Trust, and CITRIS. Partner organizations include the Leverhulme Centre for the Future of Intelligence, the Center for Long-Term Cybersecurity at Berkeley, the Berkeley Existential Risk Initiative, and ICT4Peace. The center has received over $18 million in total funding since its founding.