BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:AI Center Seminar - AI Fundamentals series - Emanuel Tewolde  -  "
 Benchmarking Cooperation-Sustaining Mechanisms and LLM Agents in Social Di
 lemmas"
DTSTART:20260615T143000
DTEND:20260615T153000
DTSTAMP:20260928T233022Z
UID:9906351e1f5e95cdee02772a3fea76ce99257b89e13179a636100dac
CATEGORIES:Conferences - Seminars
DESCRIPTION:Emanuel Tewolde\nThe talk is jointly organized by the EPFL A
 I Center and the DLAB as part of the AI fundamentals seminar series.\n\
 nHosting professor: Prof. Robert West (DLAB)\n\nTitle\nBenchmarking Cooper
 ation-Sustaining Mechanisms and LLM Agents in Social Dilemmas\n\nAbstract\
 nIt is increasingly important that LLM agents interact effectively and saf
 ely with other goal-pursuing agents\, yet\, recent works report the opposi
 te trend: LLMs with stronger reasoning capabilities behave _less_ cooperat
 ively in mixed-motive games such as the prisoner's dilemma and public good
 s settings. Indeed\, our experiments show that recent models---with or wit
 hout reasoning enabled---consistently defect in single-shot social dilemma
 s.\n   \nTo tackle this safety concern\, we present the first comparativ
 e study of game-theoretic mechanisms designed to enable cooperative outcom
 es between rational agents _in equilibrium_. Across four social dilemmas t
 esting distinct components of robust cooperation\, we evaluate four famili
 es of mechanisms: (1) repeating the game for many rounds\, (2) reputation 
 systems\, (3) third-party mediators to delegate decision making to\, and (
 4) contract agreements for outcome-conditional payments between players. A
 mong our findings\, we establish that contracting and mediation are most e
 ffective in achieving cooperative outcomes between capable LLM models\, an
 d that repetition-induced cooperation deteriorates drastically when co-pla
 yers vary.  Moreover\, we demonstrate that the mechanisms become _more ef
 fective_ under evolutionary pressures to maximize individual payoffs.\n\nB
 io\n\nEmanuel Tewolde is a fourth-year Computer Science PhD student at Car
 negie Mellon University advised by Vincent Conitzer. His research focuses 
 on algorithmic game theory\, reinforcement learning\, and LLM agents\, wit
 h an emphasis on the safety\, coordination\, cooperation\, and alignment o
 f AI systems. Ultimately\, he strives to understand how to enable artifici
 al intelligence and humans to effectively achieve better social outcomes i
 n strategic interactions.\n\nEmanuel’s work is supported by the Cooperat
 ive AI PhD Fellowship\, and has previously received the AAAI 2025 Best Pos
 ter Award. He has previously worked on AI research agents with Meta FAIR\,
  and researched ML methods for renewable energy systems with the Fraunhofe
 r-Gesellschaft. Prior to CMU\, he completed Master’s and Bachelor’s de
 grees in Mathematics at Imperial College London and the Technical Universi
 ty of Darmstadt respectively.
LOCATION:ELE 117 https://plan.epfl.ch/?room==ELE%20117 https://epfl.zoom.u
 s/j/65116073738?pwd=RSMy92kZvAC22JaNqVTJ8tzFyhVJvv.1
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR