Inês Terrucha
Research Center Trustworthy Data Science and Security, University of Duisburg-Essen
Tue
,
Nov 10, 2026
10:00 am
EST
Nov 10, 2026
10:00 am
In-person
One Portsoken
714
Portsoken Street
London, E1 8PH, UK
London, E1 8PH, UK
The Roux Institute
Room
714
100 Fore Street
Portland, ME 04101
Portland, ME 04101
Network Science Institute
2nd floor
2nd floor
Network Science Institute
101 Belvidere Street
4th floor
Boston, MA 02115
4th floor
Boston, MA 02115
Network Science Institute
2nd floor
2nd floor
Talk recording
As AI systems become more integrated into decision-making, it is increasingly important to understand how delegation to AI affects behavior in social dilemmas and what drives people to delegate their choices to AI. Building on previous findings that delegation can increase prosocial behavior in a collective-risk dilemma setting, we investigate whether these effects generalize across different dilemmas and types of AI agents. A second key goal is to examine what factors lead people to choose delegation after experiencing both delegation and self-play. To this end, we conducted a preregistered experimental program across three social dilemmas: the Prisoner’s Dilemma, the Snowdrift, and the Stag-Hunt Game. Each participant made decisions under three modes: self-play, mandatory delegation, and optional delegation. The design crossed two factors: (1) Delegation order (delegation-first vs. self-play-first, with all participants ending in optional delegation), and (2) AI agent type, where four modes of delegation were tested: rule-based, goal-oriented, supervised-learning, and LLM-based agents. This yielded eight treatment conditions per game. Each condition included N = 100 participants, yielding a total sample size of N = 2400 across all three social dilemmas. We found that delegation generally decreases cooperation. Most importantly, participants’ delegation choices in the optional phase were strongly associated with whether they earned more money when delegating or when deciding for themselves. In other words, participants were more likely to delegate again when delegation yielded higher payoffs and less likely to delegate when self-play had been more profitable, even though they could freely program the agents to their preferences. Together, these findings provide the most systematic evidence to date on how delegation influences prosocial behavior across various social dilemmas. They also highlight that delegation is not a fixed preference but a strategic payoff-driven choice, shaped directly by individuals’ past experiences with delegation versus self-play.
About the speaker
I’m a postdoctoral researcher in the Human Understanding of Machines and Algorithms lab, led by Prof. Nils Köbis, where I study how interactions with AI affect individual behavior and, collectively, shape a hybrid society. With a background in physics, I completed a joint PhD at Ghent University and Vrije Universiteit Brussel under Prof. Tom Lenaerts, combining evolutionary game theory with behavioral experiments to examine how people solve collective-risk dilemmas when acting with or through AI. This work led to three publications, including one in PNAS. My current research spans from incentivized experiments on delegation to AI in social dilemmas to experience-sampling studies of everyday AI use and its effects on emotional and social well-being. What I value most about this mixed-methods approach is being able to carry insights from the lab into the real world and back again.
Share this page:



