Jeff Clune is a Professor of Computer Science at the University of British Columbia (UBC) and a Research Scientist at OpenAI. His research spans open-endedness, evolutionary algorithms, neural architecture search, and curiosity-driven learning in AI systems.
Research program
Clune's work sits at the intersection of evolutionary computation, neural architecture search, and deep learning. It includes work on NEAT (NeuroEvolution of Augmenting Topologies), evolving neural network architectures through evolution, which predates modern neural architecture search. His research on quality-diversity algorithms studies methods that maintain a diversity of solutions rather than converging on a single optimum, an approach connected to AI robustness. In curiosity-driven learning and open-endedness, he studies AI systems that self-generate learning challenges, enabling continuous capability growth without fixed objectives. His safety-adjacent work includes research on deceptive alignment (with Nguyen et al.) examining how neural networks can learn to represent one behavior during evaluation while retaining another.
Darwin Gödel Machine
Clune is a co-author of the Darwin Gödel Machine paper (arXiv 2505.22954, a Sakana AI and UBC collaboration, 2025). The paper presents the first empirical demonstration of an LLM autonomously rewriting its own training code: the system tests modifications on its own compute, verifies improvements, and keeps changes that improve performance. The paper frames it as an empirical step toward recursive self-improvement, distinct from prior theoretical treatments.
Relative to Bostrom's original Gödel Machine concept (2003), the contribution is that the Darwin Gödel Machine uses empirical performance checks on actual compute rather than formal proofs. This makes it practically implementable but also means its self-modifications cannot be formally verified.
Career
Clune received his PhD from Michigan State University in 2009, advised by Christoph Adami. He held postdoc and faculty positions at Cornell University and the University of Wyoming. He served as Research Director at Uber AI Labs (2017–2020) and as a Research Scientist at OpenAI (2020–2021). He has been a Professor at the University of British Columbia since 2021, holding a concurrent Research Scientist position at OpenAI since 2021.
Relationships
- co-authored: The Darwin Gödel Machine (Sakana AI, May 2025) — Darwin Gödel Machine (with Sakana AI)
- related: Sakana AI — institutional collaborator on Darwin Gödel Machine
- related: Recursive Self-Improvement (RSI) — the Darwin Gödel Machine is an empirical demonstration relevant to this concept
- related: AI Scheming — open-endedness and deceptive alignment research is relevant to scheming concerns