At the last AI Lab’s weekly Research Meeting, VUB PhD student Elias Fernandez shared his team’s work with the audience. The study focuses on human behaviour in the context of collective endeavours characterized by uncertain, long-term, and non-linear returns. For instance, anthropogenic climate change, public health measures or even group hunting.
The researchers operationalized scenarios in a collective-risk dilemma, where players can invest into a public good over a number of rounds, and will only observe their payoff when the game ends. The jeopardy of crossing a dangerous threshold is able to transform a traditional public goods game, where players incur in the well-known tragedy of commons, into a coordination game, where success depends on surpassing a coordination barrier.
Behavioral experiments indicate that, when the danger of collective loss is high, slightly more than half of the experimental groups are able to coordinate and avoid the dangerous threshold. However, uncertainties over environmental variables, such as the placement of the threshold, revert the game back into a prisoner’s dilemma, decreasing group success.
When there is no uncertainty, participants invest earlier and in a more polarised manner
The effect of uncertainty about the number of rounds the game will take is hence enlightened. For instance, how much time the players have to avoid the consequences of surpassing a dangerous verge. Surprisingly, the results indicate that, for low levels of this timing uncertainty, not only collective success does not decrease significantly, but they observe a behavioral shift. Contrarily to what happens when there is no uncertainty, participants invest earlier and in a more polarised manner.
Moreover, a behavioural analysis of the experimental data reveals that, under timing uncertainty, participants of successful groups tend to reciprocate in a similar fashion to the group analogous of the Tit-for-Tat strategy, where players only increase their investments if the group does the same. Such a result indicate that certain behavioral ecosystems are more successful than others in achieving the equilibrium that is socially optimum.
Interested in more research? Stay tuned for our reviews of the research meetings. You can read at the AI Lab News about the next session which will welcome Anna Jon-And, guest researcher from Stockholm University: “Minimal prerequisites for processing language structure: A model based on chunking and sequence memory”.