Joe Collman

Joe Collman serves as a Technical AI Governance Researcher at the Machine Intelligence Research Institute (MIRI), where he is dedicated to advancing the safety and alignment of increasingly powerful AI systems. His work primarily focuses on developing clearer, more rigorous conditions and specifications for AI safety, such as defining understanding-based evaluation metrics and exploring fit-for-purpose safety architectures. Drawing inspiration from frameworks like David's Open Agency Architecture, Joe aims to create concrete, operational safety targets that can guide both research and policy. His technical expertise in software development—spanning languages like C++, Python, Haskell, and C#—enables him to contribute deeply to the design of robust safety protocols and evaluation tools.

In his role, Joe critically examines current proposals for AI governance, such as Responsible Scaling Policies, highlighting their limitations and advocating for more precise, technically grounded approaches. He actively engages with policy bodies, providing detailed input and comments to shape regulations that align with the realities of advanced AI capabilities. His work emphasizes the importance of aligning incentives across stakeholders to ensure that safety considerations remain central as AI systems become more capable. With a background in mathematics, machine learning, and game design, Joe leverages a multidisciplinary perspective to address complex challenges in technical alignment, governance, and coordination. Ultimately, his high-level aim is to increase the probability that humanity can navigate the transition to a world with superintelligent AI safely, ensuring that technical safety measures are integrated into broader governance frameworks to mitigate existential risks.