Poor Environments Make People More Prosocial to Humans but Not to AI: A Social Boundary of Foraging Theory

Abstract

The richness of an environment can influence when individuals choose to act to help others. Recent research grounded in a foraging-theoretic framework has shown that, compared with rich environments, people are more willing to act to obtain rewards in poor foraging environments, and that this environmental influence is stronger when decisions benefit others (Vogel et al., 2026). However, it remains unclear whether this environmental influence on prosocial behaviour extends to artificial intelligence (AI), an increasingly prevalent social agent. The present project therefore aims to examine how resource environments shape prosocial decisions toward humans and AI and to further investigate the potential psychological, computational, and neural mechanisms underlying these effects. The project comprises four studies, with Study 3 consisting of two experiments (Studies 3a and 3b), and is designed to yield a total effective sample of 740 participants. Combining behavioural experiments, computational modelling, causal manipulations, and electroencephalography (EEG), the project will systematically examine how resource environments and agent type jointly shape prosocial decision-making. Study 1 (target effective N = 200) will extend the foraging decision-making paradigm developed by Vogel et al. (2026). By manipulating rich and poor foraging environments and including both human and AI recipients, the study will examine how environmental richness influences prosocial behaviour toward different types of agents. The primary prediction is an interaction between environment and agent type: compared with a rich environment, a poor environment is expected to increase the tendency to help human recipients, whereas helping toward AI recipients is expected to show a different environmental effect. Study 1 will also measure perceived opportunity costs and other psychological variables relevant to human–AI prosocial decision-making, and will use a supplementary task to explore whether the environmental effect extends to other forms of social interaction. Study 2 (target effective N = 200) will use a repeated trust game to further examine how resource environments and agent type influence investment decisions during dynamic social interactions. Participants will repeatedly interact with human and AI trustees in either a rich or poor environment while receiving information about the trustees’ behavioural intentions. Using reinforcement-learning models, we will compare several preregistered candidate models to examine whether the observed behavioural differences are better accounted for by changes in the relative weighting of social and instrumental modes of decision-making, changes in learning parameters, or differences in initial value expectations. Study 3 will use two experiments to causally test the potential mechanisms underlying these effects (target effective N = 300). Study 3a (N = 150) will directly manipulate perceived opportunity costs and compare helping decisions toward human and AI recipients to test the causal role of opportunity costs in human–AI differences in prosocial behaviour. Study 3b (N = 150) will manipulate cognitive load using 2-back and 0-back tasks and employ a customer-service feedback task involving human and AI agents to examine whether changes in available cognitive resources influence decisions to exert additional prosocial effort for humans versus AI. Study 3b will also measure perceived task load, scenario realism, and perceived human-likeness of the AI agents. Study 4 (target effective N = 40) will use EEG to further investigate the neural temporal dynamics through which environmental richness and agent type influence prosocial decision-making. Participants will make prosocial decisions toward human and AI recipients in rich and poor environments and receive corresponding outcome feedback. By comparing neural activity during the decision and feedback stages across environments and agent types, this study will explore the potential neural mechanisms underlying differential effects of resource environments on prosocial decisions toward humans and AI. Overall, the present project aims to test whether the influence of foraging environments on prosocial behaviour depends on the type of agent being helped, and whether the established effect whereby poor environments promote prosocial behaviour generalises from human–human interactions to human–AI interactions. We expect that the effect of environmental scarcity on prosocial behaviour will not generalise uniformly across all agents, but may depend on the extent to which the recipient is represented as an entity with a mind and social significance. This project will further delineate the boundary conditions of foraging theory in the context of human–AI interaction and reveal how environmental scarcity shapes behavioural investment toward different social agents.