I will present our work on understanding and modelling strategic
decision-making in the face of existential risks that has been performed
with students and international collaborators in my group in Brussels.
The workhorse of this research is the collective risk dilemma (CRD), a
formalization of the problem of climate negotiations, wherein the goal
is to reach a common good, but selfish considerations may result in
failure with potential cataclysmic outcomes. I will first introduce
this dilemma and will present a series of behavioral experiments that
reveal how people act in such scenarios and how changing uncertainties
affect the outcomes. Second, I will show how both multi-agent
evolutionary and reinforcement learning approaches can match the
experimental observations, providing thus hypothesis generators for
future experiments. Third, I will introduce the idea of delegating
the decision-making in the CRD to autonomous agents and discuss how the
experimental results are changed when humans are not in the loop. This
research fits into the ongoing work to understand the origins of
cooperation and coordination and how AI may help (or not) to reach
better outcomes. Prior to this story and in light of my sabbatical
visit to Umeå, I will quickly introduce my group and some of the other
work we do on building intelligent methods for health applications,
decision-making based on expert advice and AI governance.