Cross-Species Computational Mechanisms of Abstract Rule Learning and Cognitive Flexibility in Dynamic Environment

Abstract

Abstract rule learning and cognitive flexibility are central to adaptive behavior because they allow organisms to identify regularities, generalize knowledge to unfamiliar situations, and modify behavioral strategies when environmental contingencies change. Although these capacities have traditionally been investigated through distinct experimental paradigms, growing evidence suggests that they may depend on interacting computational processes. However, existing approaches have often examined rule learning, generalization, and cognitive flexibility as partially separate computational problems, limiting a unified understanding of how these processes interact across species. This article develops an integrative computational framework that synthesizes theoretical and empirical research on abstract rule learning, generalization, and cognitive flexibility across humans, rodents, and nonhuman primates. It examines reinforcement-learning models, hierarchical representations of task structure, sequential hypothesis testing, and recurrent neural network approaches by evaluating the computational operations they explain, the assumptions they require, and their limitations for cross-species interpretation. Recent comparative studies provide evidence that humans and other animals can learn structured regularities, while also revealing differences in learning efficiency, cross-domain transfer, and adaptation to changing rules (Antoun et al., 2025; Bähner et al., 2025). These findings suggest that behavioral similarity should not automatically be interpreted as computational equivalence. Building on this literature, the article proposes an integrative conceptual framework comprising four computational operations: rule acquisition, structural generalization, change detection, and adaptive revision. By separating these operations analytically, the framework provides a basis for distinguishing shared computational principles from species-specific implementations. The framework is intended as a set of analytical distinctions and testable hypotheses rather than a validated computational model. The framework provides a basis for designing more comparable cross-species experiments, developing discriminative computational models, and evaluating whether observed similarities in behavior reflect shared mechanisms or different underlying processes.