Collective incentives reduce over-exploitation of social information in unconstrained human groups

Collective behavior arises from countless individual decisions, yet the processes governing dynamically interacting individuals in human collectives under realistic conditions remain poorly understood. A crucial trade-off in collective systems involves using personal versus social information. Over-reliance on personal information hinders information spread, while over-reliance on social information limits exploration and leads to environmental over-exploitation. This trade-off is vital across various social contexts, from social foraging to technological innovation and agricultural practices. Individuals must integrate personal and social information strategically across different timescales, but the underlying mechanisms remain largely unknown. Most prior studies used static or simulated social information sources or restricted participant choices. To understand real-world human collectives, paradigms allowing complex dynamics in naturalistic environments and novel computational models accounting for individual perspectives and spatial constraints are needed. Spatial constraints fundamentally shape the costs and benefits of social information use.

Existing literature highlights the importance of balancing personal and social information in collective decision-making. Studies on social foraging and the evolution of social learning emphasize the potential disadvantages of relying too heavily on either personal or social information. Game-theoretic models, particularly producer-scrounger games, predict increased scrounging (relying on others' discoveries) when resources are concentrated. However, most experimental studies have used simplified settings with static or simulated social information, limiting our understanding of real-world collective dynamics. The need for paradigms that allow the study of complex social dynamics in naturalistic environments, combined with computational models that capture individual perspectives and spatial constraints, has been increasingly recognized.

This study employed an immersive-reality experiment where groups of four participants searched for rewards ('coins') in a 3D virtual environment. The environment featured two resource distributions: concentrated (few rich patches) and distributed (many poor patches). Incentives were manipulated: individual (reward based on personal success) and group (reward based on group success). High-resolution visual-spatial data (25 Hz) were collected, including participant coordinates, orientation, velocity, coin count, and visual information about other participants. A 2x2 experimental design was used, with each group completing four rounds (two in each resource distribution). To infer underlying decision mechanisms from unconstrained interactions, an unsupervised Social Hidden Markov Decision Model was developed. This model used state-dependent variables (turning angles, distance to successful group members, relative bearings) to probabilistically assign participants to latent states ('Individual Exploration' and 'Social Relocation'). The model also incorporated time-dependent state predictors (exploitation visibility, distance to visible patches, number of visible exploiting players, time since last success) to quantify the influence of various factors on the probability of switching between states. Additionally, time-lagged Gaussian process regressions were used to map group-level spatio-temporal dynamics and their relationship to foraging success. A separate control condition with solitary foragers was also included.

Behavioral analyses revealed individual-level benefits of high social information use and spatial proximity in concentrated environments, but this came at the cost of group performance. Group-level incentives buffered the negative collective consequences of excessive scrounging. The Social Hidden Markov Decision Model showed that participants were more sensitive to social information in concentrated environments, frequently switching to a 'Social Relocation' state where they approached successful group members. Group-level incentives reduced overall responsiveness to social information and promoted higher selectivity over time. Multilevel Poisson regressions revealed individual benefits from high scrounging rates and close proximity in concentrated environments, contrasting with the negative impact of these factors on collective performance. Time-lagged regressions revealed a collective exploration-exploitation trade-off, with shorter timescales favoring proximity in concentrated environments and longer timescales favoring distance for better exploration. Solitary foragers generally outperformed those in groups.

The findings demonstrate how individual-level social information use can have opposing effects on individual and collective outcomes. In concentrated environments, scrounging is individually beneficial but collectively detrimental. Group incentives effectively mitigated this conflict by reducing reliance on social information, improving collective performance. The adaptive tuning of decision strategies over time, particularly the increased selectivity in group-incentivized participants, highlights the importance of strategic social learning. The results support theoretical predictions from producer-scrounger games, but also highlight the crucial role of spatial constraints and perception in modulating social decision-making. The apparent contradiction between the individual benefits and collective costs of high social information use is explained by the fact that participants could discover new patches while moving towards others, reducing the costs of frequent social information use even in distributed environments. The study extends prior research emphasizing selective and strategic social learning rather than simple copying or innovation.

This study provides a mechanistic link between individual social information use and collective dynamics in naturalistic interactions. Group incentives enhance collective performance by reducing individually beneficial yet collectively costly exploitation of social information. The research showcases a methodology that moves beyond highly constrained experiments, towards a science of unconstrained behavior in naturalistic and socially interactive environments. Future research could investigate collective foraging scenarios without zero-sum outcomes, systematically vary group sizes, include patches of varying quality, and examine how individuals update strategies based on observed payoffs.

The study primarily focused on participants from WEIRD (Western, Educated, Industrialized, Rich, and Democratic) societies, limiting the generalizability of the findings to other cultural contexts. The virtual environment, while naturalistic, is still a simplification of real-world foraging scenarios. While the model effectively captured latent decision-making processes, additional factors influencing foraging behavior might not have been fully accounted for. The limited range of group sizes (four participants) also limits the generalizability of findings to other group sizes.