Limin Ye^1,2 (), Liqiang Huang³, Ke Zhou⁴, Ming Meng^1,2; ¹Philosophy and Social Science Laboratory of Reading and Development in Children and Adolescents (South China Normal University), Ministry of Education, China, ²School of Psychology, South China Normal University, ³Department of Psychology, The Chinese University of Hongkong, ⁴School of Psychology, Beijing Normal University

Humans often rely on intuition to facilitate solving complex problems swiftly. However, what cognitive mechanisms underlie intuition? Here we explore this question by modeling large-scale behavior data, i.e., over 360 million valid behavioral responses that were collected through an online game simulating the Traveling Salesman Problem (TSP) on a smartphone APP platform. TSP is a classic combinatorial optimization problem, where a salesman aims to find the shortest path traversing specified cities (nodes) and returning to the starting point. Through the TSP we investigate human decision-making capabilities in scenarios where optimal solutions are hard to compute as the number of nodes increases. The findings were three-fold. First, as the number of nodes increases, the computational difficulty of solving the TSP rises exponentially, but human efficiency in solving the problem does not decrease proportionally, suggesting that humans indeed base their choices on intuition rather than exhaustive calculation. Second, humans tend to connect adjacent nodes sequentially and avoid route crossings. Performance in this task correlates with the visual perception of the overall structure, with better scores achieved when the convex hull angle of the formed route is larger. We therefore hypothesize that in solving complex problems, individuals often plan based on visual construction of simple mental representations. Finally, we have trained a deep learning neural network to learn human algorithms, achieving a 95% accuracy in predicting human solutions. Using it as a benchmark, with the nearest-neighbor heuristic as the baseline, we simulated visual physiological structures and established an exploratory Bayesian feature pyramid model based on global centralized adaptive feature modulation. This model closely approximates the predictive capabilities of neural networks and human performance, exhibiting both robustness and interpretability. This work provides a novel framework for understanding the cognitive mechanisms of humans in solving the TSP based on visual intuition.

Acknowledgements: The research was supported by grants from the STI2030-Major Projects (2021ZD0204200), the Sino-German Center for Research Promotion (M-0705), and the National Natural Science Foundation of China (31871136).