Seminars

View all Seminars  |  Download ICal for this event

AI Agents and Human Decision-Making: Architectures, Incentives, and Alignment

Series: Bangalore Theory Seminars

Speaker: Yann Issam Aouidef, Stanford University

Date/Time: Nov 13 11:00:00

Location: CSA Auditorium, (Room No. 104, Ground Floor)

Abstract:
Can competition between misaligned AI agents deliver the benefits of alignment?\n\nI will present a multi-leader Stackelberg framework extending Bayesian persuasion to multi-round conversations between a user and multiple AI agents.\n\nA key condition is that the users utility lies approximately within the convex hull of the agents utilities.\n\nThree results show how competition delivers optimal or near-optimal utility in every equilibrium:\n\n1. Bayes-optimal utility when a perfectly aligned model would enable the user to learn her optimal action.\n\n2. Near-optimal utility for a non-strategic user following a quantal-response rule, under a weaker assumption.\n\n3. Near-optimal utility when the user selects the best agent after an evaluation period, without additional distributional assumptions.\n\nMicrosoft Teams link:\nhttps://teams.microsoft.com/l/meetup-join/19%3ameeting_ZGE3NDg5NzktMWQ0Zi00MzFmLTg5OTgtMTMyYWM4MWQyYjI2%40thread.v2/0?context=%7b%22Tid%22%3a%226f15cd97-f6a7-41e3-b2c5-ad4193976476%22%2c%22Oid%22%3a%227c84465e-c38b-4d7a-9a9d-ff0dfa3638b3%22%7d\n\nWe are grateful to the Kirani family (Link and the Walmart Center for Tech Excellence (Link for generously supporting this seminar series.\n\nHosts: KVN Sreenivas, Sreeramji K S, Ritabrata Barat, Venkata Sai Nikhil Srivatsava Ayyadevara