Abstract:

This study investigates a puzzling phenomenon: Why do individuals with exceptional mathematical abilities often fail when applying these skills to socially complex numerical environments like stock markets? We explore the cognitive and neural mechanisms underlying social numerical inductive reasoning (SNIR) - the process where people must identify numerical rules while simultaneously adapting to others' decisions in multi-agent settings. Traditional approaches have studied numerical reasoning and social cognition separately, however, their critical interaction in economic decision-making remains unclear. Our research specifically investigate how brain regions responsible for numerical rule acquisition compete with regions for intentional inference, providing a new explanation for bounded rationality in complex social-numerical environments.

Our work integrates three established theoretical frameworks: dual-system theory (which distinguishes between fast, intuitive thinking and slow, deliberative reasoning), theory of mind (our ability to understand others' mental states), and Arthur's bounded rationality in complex systems (which explains how rational decision-making becomes limited in complex environments). By integrating these perspectives, we extend traditional dual-system theories to account for the interaction between mathematical and social cognition. Previous brain imaging studies have separately identified the neural basis of numerical inductive reasoning (primarily in DLPFC/FPC) and social cognition (primarily in TPJ/mPFC), but have not examined their competitive or cooperative interactions in socioeconomic contexts. Our approach bridges this gap by investigating how these brain systems dynamically reorganize under varying conditions.

We employ a multi-modal approach combining behavioral experiments, ERP, and fMRI techniques to examine the brain activity underlying SNIR: First, we have developed a specialized experimental paradigm that combine numerical sequences (such as “35, 37, 41, 47”) with multiplayer social contexts (e.g., real-time El Farol Bar Problem simulations, where participants must decide whether to attend a potentially crowded location). The task uniquely isolate SNIR-specific processes by manipulating cognitive load, social load, and incentive structures. Second, we will implement time-resolved event-related potential (ERP) analyses to distinguish between quick, intuitive reasoning (System 1, associated with alpha-band oscillations) and more deliberate, analytical reasoning (System 2, linked to theta-band oscillations) in both numerical and social cognition. Third, we will conduct the functional magnetic renounce imaging (fMRI) analyses, including multi-voxel pattern analysis (MVPA), representational similarity analysis (RSA), and dynamic causal modeling to map neural networks during SNIR tasks. This approach revealed distinct neural patterns for numerical rule acquisition (DLPFC/FPC) versus ToM (TPJ/mPFC) and captured their dynamic interaction.

Our research aims to validate a dual-pathway model of SNIR with several expected outcomes: We anticipate identifying distinct neural signatures for the cognitive pathway (DLPFC-FPC axis for numerical rule acquisition) and social pathway (TPJ-mPFC axis for ToM). These pathways should exhibit differential activation depending on task demands. We anticipate that when facing complex numerical rules, System 2 (associated with the FPC) will dominates in non-social tasks. However, in socially complex numerical tasks, the ToM System 1 (associated with the TPJ) will prioritizes intention inferences, potentially suppressing numerical rule acquisition processes. We predict that contextual factors will modulate system dynamics, with evolutionarily familiar contexts enhancing ToM System 1 activation and loss-avoidance contexts strengthening mPFC-DLPFC connectivity for thinking about others' thinking (recursive mentalizing) at the expense of pure numerical acquisition.

This research reveals why individuals like Newton—brilliant at discovering patterns in the physical world—could fail dramatically in stock trading: SNIR demands not just mathematical reasoning but also recursive mentalizing, creating a dual-pathway model where social intuition often overrides numerical deliberation. Our findings redefine bounded rationality in multi-agent systems as emerging from competition between cognitive and social neural networks rather than from pure computational limitations. The significance of this work extends across multiple domains. For complexity economics, it provides micro-level neural evidence supporting the theory that bounded rationality emerges from social-cognitive constraints, explaining market inefficiencies despite individual intelligence. For education, our results suggest that enhancing SNIR might require training that specifically targets the integration of analytical thinking with social cognitive processes, particularly in conflict-rich environments. For artificial intelligence, our findings suggest that effective AI systems for economic applications should integrate both analytical thinking and ToM processes for economic simulations. By clarifying the neural basis of socioeconomic decision-making, this work offers a valuable insights for enhancing human-AI collaboration in complex decision environments.

Key words: complexity economy, inductive reasoning, bounded rationality, dual system