Kenneth Stanley

Career

Stanley was a professor of computer science at the University of Central Florida, where Stanley held the Charles Millican Professorship. Stanley co-founded Geometric Intelligence with Gary Marcus, Zoubin Ghahramani and Doug Bemis, serving as co-chief scientist; after Uber acquired the company and rebranded it as Uber AI Labs, Stanley led its core research.¹²

Stanley later led the Open-Endedness team at OpenAI and served as CEO of Maven. At Lila Sciences, Stanley is Senior Vice President of Open-Endedness, leading work on methods for continual, open-ended discovery and creativity in AI systems.²¹

Research

Stanley's research centers on neuroevolution and open-ended search. With colleagues, Stanley created NeuroEvolution of Augmenting Topologies (NEAT) and its HyperNEAT and ES-HyperNEAT extensions, compositional pattern producing networks, novelty search and minimal criterion coevolution, along with interactive systems such as Picbreeder, NERO and Galactic Arms Race.²¹

In 2019, with Jeff Clune, Joel Lehman and others, Stanley co-authored Go-Explore, an approach to hard-exploration problems in reinforcement learning, and POET, which generates increasingly complex learning environments together with their solutions. The 2022 Evolution through Large Models paper applied code-generating language models as mutation operators in genetic programming.⁴⁵³

Teaching and writing

With Joel Lehman, Stanley wrote Why Greatness Cannot Be Planned: The Myth of the Objective, which examines open-ended exploration, rather than fixed objectives, as a route to discovery. Springer published the book in 2015, and its argument starts from a finding in artificial intelligence research.¹²⁶

Work and contributions

• Evolution through Large Models (ELM) Published research

  Co-author with Joel Lehman and others of "Evolution through Large Models" (2022), which uses code-generating language models as mutation operators in genetic programming; Stanley's homepage lists ELM among the methods Stanley co-invented.³²

• NeuroEvolution of Augmenting Topologies (NEAT) Published research

  Listed on Stanley's homepage as an inventor, with colleagues, of NEAT and its extensions HyperNEAT and ES-HyperNEAT, along with compositional pattern producing networks and novelty search.²

Timeline

1. 2015Published Why Greatness Cannot Be Planned, co-written with Joel Lehman, with Springer.⁶
2. Jan 7, 2019Co-authored POET, a method for generating learning environments and their solutions together.⁵
3. Jan 30, 2019Co-authored Go-Explore, a new approach to hard-exploration problems in reinforcement learning.⁴
4. Jun 17, 2022Co-authored "Evolution through Large Models" on language models as mutation operators.³

Other documented roles

Roles the sources document that are not confirmed as current.

• Open-Endedness Team Leader, OpenAI²¹
• Head of Core AI Research, Uber AI²
• Charles Millican Professor in Computer Science, University of Central Florida²

Where they work

• Lila Sciences 1 person in the wiki

Related people

• Jeff CluneReinforcement learning · Agents
• David HaReinforcement learning · Agents
• Noam BrownReinforcement learning · Agents
• Denis YaratsReinforcement learning · Agents
• Shimon WhitesonReinforcement learning · Agents
• Katja HofmannReinforcement learning · Agents
• Siddharth NarayananReinforcement learning · Agents
• John LangfordReinforcement learning · Machine-learning theory
• Jakob FoersterReinforcement learning · Agents
• Josh AlbrechtAgents · Reinforcement learning