Lila Sciences builds open-ended search team for autonomous science

Lila Sciences is treating open-ended search as core infrastructure for its push toward scientific superintelligence. Ken Stanley, senior vice president of open-endedness, is leading a team focused on methods that generate diverse, unexpected possibilities rather than simply optimizing toward a fixed goal. The approach draws inspiration from biological evolution and from scientific serendipity, where useful discoveries can emerge from paths that initially look unpromising.

Starting in 2025, Stanley assembled researchers with backgrounds spanning quality-diversity algorithms, natural language processing, human-algorithm interfaces, robotics, materials science, and artificial life. Their work includes developing new algorithms, adapting existing ones in unconventional ways, and combining large language models with mechanisms that push systems away from repetitive suggestions. Joel Simon is also designing LLM harnesses that can track prior ideas and request more divergent ones.

Lila plans to connect this research to AI Science Factories, automated labs that verify model outputs through experiments. Current areas include mRNA therapeutics and quantum dots, with scientists and computer researchers working together to adapt abstract methods to practical constraints. In the fall of 2026, the team plans to release papers and a white paper covering four areas of open-ended research, including idea generation, judging interestingness, model representations, and the broader feedback loop of hypothesis and experiment.