ArXiv:2605.31514 Shows Age of Empires II Neural Net Matches LLM Attribute Claims

A neural network trained on Age of Empires II exhibits the same anthropomorphic attributes researchers ascribe to LLMs, per arXiv:2605.31514. The paper trains a simple network on the game, proves Age of Empires II is functionally and Turing-complete, and states that any entity in a sufficiently powerful substrate presents equivalent attributes. It rejects generalised assumptions of morality or understanding independent of substrate and proposes a null assumption of non-uniqueness for experiments. Related analyses in agentic workflow literature, including arXiv:2303.08774 on LLM agents, similarly document goal-directed outputs without substrate-specific criteria. The paper's approach surfaces measurement gaps also noted in arXiv:2212.03551 on emergent abilities, where consistent behaviour alone fails to distinguish understanding from pattern completion across game and language domains.