Beyond Language: Why Artificial Intelligence Should Separate Communication from Reasoning

Modern large language models combine natural-language communication and problem-solving within the same architecture. This conceptual position paper argues that future artificial intelligence systems should separate human-language processing from internal reasoning. It distinguishes routine cognition from constructive reasoning and proposes an architecture comprising a language interface, a semantic intermediate representation, and a nonlinguistic reasoning engine. Potential benefits include reduced linguistic overhead, improved multilingual efficiency, and greater computational capacity for exploration, verification, and problem-solving. The paper also examines semantic loss, conversion cost, interpretability, and the difficulty of defining reasoning operationally, and concludes with a research agenda for experimentally testing the proposal.