Berkeley AI Safety encompasses research initiatives at the University of California, Berkeley dedicated to ensuring advanced artificial intelligence systems remain beneficial to humanity. The primary entity is the Center for Human-Compatible Artificial Intelligence (CHAI), founded in 2016 by a group of academics led by Berkeley computer science professor Stuart J. Russell, co-author of the widely used textbook Artificial Intelligence: A Modern Approach. CHAI focuses on developing methods to align AI behavior with human values, addressing risks posed by increasingly capable Artificial intelligence systems.
Founding and Funding
CHAI was established in 2016 with a mission to refocus AI research on safety and human compatibility. The founding faculty included prominent researchers from multiple institutions: Russell, Pieter Abbeel, and Anca Dragan from Berkeley; Bart Selman and Joseph Halpern from Cornell University; Michael Wellman and Satinder Singh Baveja from the University of Michigan; and Tom Griffiths and Tania Lombrozo from Princeton University. In its inaugural year, the Open Philanthropy Project recommended that Good Ventures provide CHAI with $5,555,550 in support over five years. Subsequent grants from OpenPhil and Good Ventures have totaled over $12,000,000, including funding for collaborations with the World Economic Forum and the Global AI Council.
Research Focus
CHAI's research centers on value alignment, particularly using inverse reinforcement learning, where AI systems infer human values by observing behavior rather than following explicit instructions. This approach aims to create AI that can learn and adapt to human preferences in complex, real-world scenarios. The center has also investigated human-machine interaction, including the design of systems where an AI can be safely interrupted or switched off, even if the AI has the capability to override such controls. These topics connect to broader challenges in Machine learning and Deep learning, where ensuring safety is increasingly critical as models grow in capability.
Broader Context
Berkeley's AI safety work is part of a wider academic and industry movement. Other institutions, such as the Future of Humanity Institute and the Machine Intelligence Research Institute, address similar concerns. CHAI's research complements efforts at organizations like OpenAI and Anthropic, which also prioritize alignment. The center's focus on human-compatible AI has influenced discussions on existential risk from artificial general intelligence, as highlighted in Russell's book Human Compatible.
Impact and Collaborations
Through partnerships with the World Economic Forum and the Global AI Council, CHAI has extended its influence beyond academia, engaging policymakers and industry leaders. The center's funding from OpenPhil and Good Ventures underscores its role in the philanthropic push for AI safety. As of the mid-2020s, CHAI continues to publish research and train students, contributing to a growing body of knowledge that aims to ensure Generative AI and other advanced systems remain under human control.
See Also
- Existential risk from artificial general intelligence
- Future of Humanity Institute
- Future of Life Institute
- Human Compatible
- Machine Intelligence Research Institute