Political Economy of Reinforcement Learning (PERLS) Workshop

Neural Information Processing Systems (NeurIPS)

December, 2021

Workshop Draft Schedule (Tuesday, December 14th)

All details subject to change.

Workshop draft schedule - Tuesday, December 14th
Time (ET) Event
8:00am Welcome
8:20am [Plenary] Culturing PERLS (Mark Nitzberg)
Chairperson: Aaron Snoswell
8:55am 5 minute break

Theme 1:

Vagueness & Specification

Theme Summary: How should RL systems be designed in the face of difficult specification problems and normative indeterminacies? How do these problems change in particular domains or as RL systems become more capable?
Chairperson: Thomas Gilbert
  • Stuart Russell
  • Mireille Hildebrandt
[Panel discussion]
  • Stuart Russell
  • Mireille Hildebrandt
  • Salomé Viljoen
  • Natasha Jaques
10:40am 10 minute break

Theme 2:

Legitimacy, Accountability,
& Feedback

Theme Summary: To what extent does RL automate many feedback mechanisms for oversight and control that social institutions (e.g. governments) presently enforce or take for granted? How can stakeholders acquire a voice in system evaluation?
Chairperson: Jakob Foerster
  • "Legitimacy" in the Computational Elicitation of Preferences in Mechanism Design (Jake Goldenfein)
  • The Role of Explanation in RL Legitimacy, Accountability, and Feedback (Finale Doshi-Velez)
  • Evaluating Reinforcement Learners (Michael Littman)
[Panel discussion]
  • Jake Goldenfein
  • Finale Doshi-Velez
  • Michael Littman
  • Evi Micha
  • Ivana Dusparic
12:15 Noon 45 minute lunch break
1:00pm Poster session on Gather Town - for accepted papers, please see here.
2:55pm 5 minute break

Theme 3:

Tools for democratization

Theme Summary: What new policy tools (technical or otherwise) could be used to achieve democratic control and oversight of RL systems? What organizational theories are needed to inform or shape how democratization should proceed?
Chairperson: Michael Dennis
  • Antimonopoly as a Tool for Democratization? (Ayse Yasar)
  • Reinforcement of What? Shaping the Digitization of Judgement by Reinforcement Learning (Frank Pasquale)
  • Metrics are Tricky (Rachel Thomas)
[Panel discussion]
  • Ayse Yasar
  • Frank Pasquale
  • Rachel Thomas
  • Mason Kortz
4:30pm Closing remarks (Thomas Krendl Gilbert)