The rapid rise of computational power allows ever more capable AI agents to be trained in simulation. A simulator, of course, does not fully reflect reality nor human preferences. How can AI agents learn useful, human-aligned behaviors in simulation that transfer to new settings and people?
I consider this question from the lens of generalization, human-AI coordination, and open-ended learning, as part of the Autonomous Assistants team at Google DeepMind.
Dec 2023: Co-organized the 2nd Workshop on Agent Learning in Open Endedness (ALOE) at NeurIPS 2023 🌱, seeking to bridge ideas of open-ended evolution in ALife with self-supervised machine learning. The event was a lot of fun and drew out a special community of researchers.
Dec 2023: Joined DeepMind as a Research Scientist.
Nov 2023: Released minimax, a library for rapid experimentation with autocurricula methods for RL in JAX, including new parallelized and multi-GPU/TPU versions of PLR and ACCEL.
Sep 2023: Became a Doctor (of computers). You can find my dissertation on arXiv.