By: Erin Vollick
9 Oct, 2023
Jeff Clune, Vector Institute: AI that learns to improve itself
October 9, 2023
Jeff Clune, Canada CIFAR AI Chair, Vector Institute
Can AI replicate the inventiveness and creativity of natural evolutionary processes? As he works towards AI that might endlessly learn and improvise on its own, Canada CIFAR AI Chair Jeff Clune brings that eventuality ever closer.
“CIFAR has been visionary in how it provides scientific funding — it finds talented scientists, gives them resources and then steps back to let them do the work.”
Jeff Clune, Canada CIFAR AI Chair, Vector Institute
Clune’s efforts to expand the frontiers of AI have been hugely influential. His work has been cited nearly 30,000 times, and a 2014 NeurIPS proceedings paper on deep neural networks by Clune and collaborators has garnered more than 9,000 citations.
His collaboration on POET (paired open-ended trailblazer) — an AI agent that solves ever-increasing environmental challenges that it builds for itself — inspired a wave of research into AI self-generating environments. Clune and trainee Jenny Zhang’s recent work on OMNI (open-endedness via models of human notions of interestingness) is likewise generating great interest in the AI community. With OMNI, the team demonstrates how large language models can help AI agents select which skills are interesting to learn and, by extension, can lead to self-improving AI.
For Clune, whose move to Canada in 2021 was supported by the Pan-Canadian AI Strategy, “CIFAR has been visionary in how it provides scientific funding — it finds talented scientists, gives them resources and then steps back to let them do the work.” He also notes CIFAR’s support in helping him recruit and be competitive for “some of the best students in the world.”