Peter Henderson

I'm an assistant professor at Princeton University, leading the Polaris Lab. I work on reinforcement learning, alignment, and law.

Areas of interest:

I care about building deep reinforcement learning agents that matter: those that help us solve major challenges for the public good, reliably, safely, and semi-autonomously.
- - Novel RL methods in latent representation space for discovery and long-horizon strategic decision-making.
  - Long-horizon value function learning, planning, and alignment.
  - Alignment of representations and values across domains (e.g., text to control). Does alignment in one domain transfer to others?
  - Fighting Goodhart's Law through generalizable optimization targets and evaluations.
  - Automated frameworks for building better rules for AI (and humans!).

I also care a lot about shaping the law to make the United States ready for a post-AGI world.
- - Gradual loss of human control over the rule of law as AGI becomes more prevalent in legal processes.
  - Empowering people and labor do reduce wealth and power inequality that AGI exacerbates.
  - Understanding what a new world of intellectual property law looks like after AGI.

I'm looking for:

Students and postdocs to join Polaris Lab (PhD/Postdoc expression of interest; Undergrad/MSE expression of interest). (I'll do my best to get back to you but can't make any promises due to the large volume of emails I receive.)
If you just want to collaborate, or if you want to be a visitor in the lab, please fill out this form here!
Great seed investments in new AI-first companies (DM me on X or Bluesky with the pitch!)

Some things I did before:

Launched products at Amazon and AWS.
Did research at Meta Fundamental AI Research (FAIR).
Got my JD & PhD in Computer Science from Stanford, with the great Dan Jurafsky.
Did RL Research at McGill & Mila with David Meger & Joelle Pineau.
Worked at the California Supreme Court with Justice Mariano-Florentino Cuéllar; led the Domestic Violence Pro Bono Project at Stanford Law School.

Google Scholar · GitHub · X · Bluesky · Publications