The study was conducted by Marta C. Couto and Fernando P. Santos from the Socially Intelligent Artificial Systems group at the Informatics Institute of the University of Amsterdam, together with Christian Hilbe from the Interdisciplinary Transformation University in Linz. Using mathematical and computational models from evolutionary game theory, the researchers examined how individuals learn to act in strategic situations where outcomes depend partly on the behaviour of others, such as negotiation, task allocation, and decisions about cooperation. In these models, each individual has a certain sensitivity to outcomes: some are strongly inclined towards strategies that appear to work best, while others learn more erratically and sometimes persist with suboptimal choices.

Most existing theories and models assume that this sensitivity is the same for everyone and remains constant over time. The new research challenges both assumptions. The team investigated what happens when individuals differ in how strongly they respond to rewards, and whether that very trait can itself evolve. The finding is surprising: high sensitivity to rewards is not always the most successful approach.

To explain this, the researchers analysed classic game-theoretic scenarios representing social dilemmas. In a donation game, one person can help another by paying a cost, for example by investing time or resources for someone else’s benefit. Helping is good for the recipient but always costly for the giver in the short term. In these competitive situations, highly sensitive learners quickly discover that withholding help pays off: they donate less and earn more. Because this provides a clear advantage over less sensitive individuals, an evolutionary arms race towards short-term gain emerges, even if the group as a whole ultimately fares worse.

In a different type of scenario, the snowdrift game, the dynamic reverses. The researchers illustrate this with the example of a shared office kitchen: everyone values cleanliness, but everyone hopes that someone else will clean the sink. If no one acts, everyone loses. In this type of situation, lower sensitivity to immediate outcomes can become a strategic asset. Those who are less focused on the direct result clean less often, thereby encouraging their more sensitive colleagues to take on the task. Similar phenomena are described in psychology as strategic incompetence, and in biology as the Red King effect: by reacting more slowly or caring less about small gains, the burden can be shifted to others.

Over the long term, however, these short-term patterns play out very differently. In donation games, the most sensitive learners in competitive environments are rewarded increasingly often, pushing the population as a whole towards a strong focus on immediate gain. In most snowdrift scenarios, the evolutionary trajectory differs, with the distribution of sensitivities across the population shifting depending on the social context. The study thus demonstrates that the evolution of decision-making styles depends strongly on the type of social situation in which individuals find themselves.