Referring, Inferring, and Preferring

Philosophy of mind, cognitive science, and AI lack a unified model to explain how systems form meaningful propositional attitudes—states like belief, desire, and intention. This work introduces and defends the Trialectics framework as a response to this theoretical gap. We propose that any system capable of forming meaningful propositional attitudes—that is, internal states with truth-evaluable, compositionally structured content that can be rationally evaluated and guide goal-directed behavior—must instantiate three interdependent functional capacities: Reference (R), the ability to establish semantic links; Inference (I), the capacity to process and transform content systematically; and Preference (P), the function of internal evaluation and goal-orientation. The central thesis is that these capacities are not modular but mutually conditioning: meaningful content arises only from their dynamic entanglement, where reference shapes and is shaped by inference and preference, and so on. By embedding evaluation (P) within the cognitive loop, Trialectics offers a naturalized solution to the problem of normativity in reasoning and reconciles representational requirements with enactivist critiques by defining reference functionally rather than metaphysically. The framework’s utility is demonstrated by showing how it implicitly structures existing cognitive architectures (e.g., BDI, Active Inference) and provides a powerful diagnostic and design tool for AI, particularly concerning alignment, generalization, and interpretability. Ultimately, Trialectics posits a fundamental architectural condition for any system—biological or artificial—to achieve the flexible, offline, and rationally evaluable propositional attitudes characteristic of sophisticated intelligence.