BERI-ALL Collaboration
About
The BERI-ALL Collaboration is a partnership between the Berkeley Existential Risk Initiative (BERI) and the Autonomous Learning Laboratory (ALL) at the University of Massachusetts Amherst's Manning College of Information and Computer Sciences. BERI is a US-based 501(c)(3) public charity that collaborates with university research groups working to reduce existential risk by providing them with free services and operational support. The Autonomous Learning Laboratory, formerly known as the Adaptive NetWorks (ANW) Laboratory, was founded by Andrew Barto and is now co-directed by Philip S. Thomas and Bruno Castro da Silva. The lab conducts foundational artificial intelligence research with emphases on AI safety and reinforcement learning, and particularly the intersection of these two areas. Its long-term goals include developing more capable artificial agents, ensuring that AI systems are safe and well-behaved, improving understanding of biological learning, and forging stronger links between studies of learning across computer science, engineering, neuroscience, and psychology. The trial collaboration between BERI and ALL began in August 2020, with BERI providing operational support funded through its general use funds. In January 2022, BERI converted the trial into a main collaboration, supported by a $250,000 grant from the EA Funds Long-Term Future Fund. This funding was used to hire a full-time ML software engineer to develop the Seldonian Toolkit, a software library that makes it easier for researchers and practitioners to apply and create Seldonian machine learning algorithms with high-confidence safety and fairness guarantees. Seldonian algorithms were introduced by Philip Thomas and collaborators in a 2019 paper published in Science. The framework centers real safety concerns in machine learning by allowing users to specify safety and fairness constraints that algorithms must satisfy with high probability. The Seldonian Toolkit, developed as a collaboration between ALL and BERI, includes components for engine, experiments, tutorials, and a GUI. In 2022, the BERI-ALL Collaboration received $150,000 from Jaan Tallinn through the Survival and Flourishing Fund's S-Process grant round. BERI and ALL also jointly organized a Seldonian Toolkit Competition in March-April 2023, designed for undergraduate and graduate students in the US and Canada to learn about and develop safe and fair machine learning algorithms. ALL currently includes approximately 8 doctoral students and one staff member (Grants and Contracts Coordinator). Notable alumni of the lab include Richard Sutton and Sridhar Mahadevan.
Theory of Change
The BERI-ALL Collaboration aims to reduce existential risk by accelerating research at the intersection of AI safety and reinforcement learning. By providing operational support, staffing, and funding to the Autonomous Learning Laboratory, BERI removes administrative bottlenecks and enables researchers to focus on developing machine learning algorithms with provable safety and fairness guarantees. The development of the Seldonian Toolkit creates practical tools that allow both researchers and industry practitioners to responsibly apply machine learning to high-risk applications, potentially establishing safety-by-design as a norm in the field. The causal chain runs from operational support to faster research iteration to widely-available safe ML tools to reduced risk from deployed AI systems.
Details
- Start Date
- -
- End Date
- -
- Expected Duration
- -
- Funding Raised to Date
- -
- Last Updated
- Apr 3, 2026, 1:18 AM UTC
- Created
- Apr 3, 2026, 1:18 AM UTC