Navigating the labyrinth of involution: LLM-assisted smart agent-based simulation of social resource competition

Involution is an inefficient equilibrium of positional competition in which intensified competitive input fails to generate proportional positional advancement under scarce social resources and weak exit options. This study conceptualizes involution as non-advancement-generating input under positional scarcity and develops a hybrid model combining rule-based agent-based modeling (ABM) with large language model (LLM) assisted smart agent-based modeling (SABM). Agents in SABM make bounded movement decisions and provide natural-language rationales. ABM enables scalable pattern assessment and sensitivity analysis. The results identify five patterns: homogeneous crowding and the embedded-catfish agent intensify involution, whereas agent-attribute heterogeneity, advancement-path diversification, and competitive-exit activation mitigate it. The catfish effect is conditional rather than linear: a single sufficiently embedded catfish agent can amplify involution, while a sufficiently larger number of catfish agents may shift the system toward broader agent-attribute heterogeneity and mitigate involution. These findings clarify how positional scarcity, homogeneity, embeddedness, catfish-agent quantity, path structure, and exit options jointly shape involution dynamics.