Human social interactions are rarely purely cooperative or competitive; instead, they exist on a continuous spectrum. Existing research often simplifies this complexity by focusing on binary choices in economic games, neglecting the dynamic shifts in behavior along this continuum. This study addresses this gap by introducing a novel framework that examines the continuous trade-off between cooperation and competition. The central research question is to understand the computational and neural mechanisms underlying this dynamic interplay. The study's significance lies in its potential to advance our understanding of how the brain manages the complex decisions involved in navigating social situations, where cooperation and competition are not mutually exclusive but rather interwoven and context-dependent strategies. This nuanced approach could have significant implications for understanding social behaviors, resolving conflicts, and improving group performance in various social settings.

Prior research on cooperation and competition predominantly uses economic games with binary choices. These studies suggest that cooperativeness is influenced by three factors: (i) environmental factors (resource availability and distribution), (ii) individual predispositions and social biases, and (iii) dyadic interactions (reciprocity, trust, and reputational information). Economic models often assume players will reach an optimal equilibrium, but this is not always observed in reality. Individual differences and biases significantly impact behavior. Mentalizing processes, particularly inferring others' intentions and adapting behavior accordingly, are also crucial. While Bayesian models have successfully captured behavior in some economic games, a comprehensive model incorporating environmental influences, individual biases, and the continuous adjustment of cooperation levels has been lacking. Previous neuroimaging studies implicated brain regions like the temporo-parietal junction (TPJ), medial prefrontal cortex (mPFC), and anterior cingulate gyrus (ACC) in cooperative and competitive behaviors, often by processing social prediction errors. However, a clear understanding of how these regions integrate social context and Bayesian signals to modulate cooperation remains elusive.

To address these issues, the researchers designed a new game, the "Space Dilemma." This game utilizes a continuous spatial location as a parametric measure of the cooperation-competition continuum. Two players simultaneously choose locations in a linear space; the player closer to a randomly appearing target wins a reward, with the reward distribution manipulated to create three social contexts: (i) cooperative (equal reward sharing), (ii) competitive (winner takes all, loser loses), and (iii) intermediate (winner takes all, loser neither gains nor loses). The game was played over multiple trials across the three contexts. Twenty-seven pairs of participants (one in an fMRI scanner, one in an adjacent room) played the game. fMRI data were acquired during gameplay, while behavioral data were collected from both players. The researchers hypothesized that behavior would be influenced by personal predispositions, social context, and the co-player's behavior. A Bayesian model was developed to predict participant behavior, incorporating parameters for social bias, context influence, and the weight of the co-player's actions. The model's parameters were then used as regressors in fMRI analyses, investigating the neural correlates of social prediction errors (measured by Kullback-Leibler divergence, KLD) and their relationship to subsequent behavior. Distinct fMRI GLMs were used to analyze neural activity, focusing on the TPJ, mPFC, ACC, and paracingulate cortex. ROI analysis and time series analysis further explored how neural activity predicted behavioral changes.

The study found that behavior in the Space Dilemma was significantly influenced by social context, with increased competitiveness observed in the competitive and intermediate contexts compared to the cooperative context. Participants exhibited reciprocal behavior, adapting their actions based on their co-player's previous moves, indicating a tit-for-tat strategy. Quantitative analysis showed that a Bayesian model incorporating social bias, context-dependent risk, and co-player's influence best predicted participant behavior, outperforming simpler models. fMRI analysis revealed that the right TPJ encoded both unsigned (magnitude) and signed (direction) prediction errors related to the co-player's competitiveness, with distinct subregions processing each type of error. The anterior TPJ (arTPJ) encoded a directionally specific prediction error, predictive of future changes in cooperativeness, particularly in the intermediate context. Furthermore, the posterior dorsomedial prefrontal cortex (pDMPFC), ACCg, and paracingulate cortex (PaCg) showed context-dependent activity related to updating cooperation levels for self and other. Activity in ACCg and PaCg correlated with the model's social bias and tit-for-tat parameters, suggesting a role in modulating behavior based on social context and co-player actions. The analyses suggest a self-other gradient in ACC, reflecting the processing of self and other's cooperation levels.

The findings demonstrate that the brain employs a complex interplay of computational and neural mechanisms to navigate the continuous trade-off between cooperation and competition. The Bayesian model successfully captures this dynamic behavior, highlighting the importance of integrating social context, individual biases, and online learning from interactions. The dissociation of signed and unsigned prediction errors in the TPJ provides further insight into the flexibility of social information processing. The role of the mPFC, ACCg, and PaCg in integrating various factors (social context, social bias, co-player actions) to modulate behavior is consistent with their known involvement in social cognition and mentalizing. This work advances our understanding beyond binary models, showing how subtle behavioral adjustments are critical in dynamic social interactions.

This study provides a comprehensive understanding of the computational and neural underpinnings of the continuous cooperation-competition trade-off. The Space Dilemma game and Bayesian model effectively captured the dynamic interplay of context, bias, and co-player influence on behavior. Neuroimaging results showed distinct roles for TPJ subregions in processing social prediction errors and their impact on subsequent behavior. The mPFC, ACCg, and PaCg also play crucial roles in integrating information to adjust cooperation levels. Future studies could explore the impact of uncertainty, task difficulty, and group dynamics on this trade-off, deepening our understanding of complex social behavior.

The study used same-sex pairs of participants, which might limit the generalizability of findings to mixed-sex interactions. The sample size, while sufficient for detecting significant effects, could be expanded to increase statistical power. The Space Dilemma, while novel, is a simplified model of real-world social interactions. The Bayesian model, though successful, might not capture all the nuances of human decision-making in such contexts. The fMRI data analysis relied on specific model parameters, raising potential issues related to model assumptions and the interpretation of correlations.