If LLMs Have Human-Like Attributes, Then So Does Age of Empires II (arxiv.org)

A recent exploration from Microsoft and the University of York highlights the flawed tendency in AI research to attribute human-like qualities to large language models (LLMs). The study employs a neural network trained on the classic video game Age of Empires II to argue that the anthropomorphic attributes ascribed to LLMs, such as morality and understanding, may not be uniquely indicative of these models. This suggests that if other sufficiently powerful systems, like a video game or even mundane environments, can exhibit similar behaviors, then attributing human-like qualities to LLMs could stem from misinterpretations rather than inherent traits. The significance of this work lies in its challenge to contemporary assumptions within AI research. By presenting the idea of "non-uniqueness" in LLMs, it calls for a rigorous reevaluation of how researchers frame their experiments. The paper advocates for a 'null assumption' approach in experiments, urging scientists to avoid presupposing anthropomorphic characteristics to ensure that their findings are valid and meaningful. As AI continues to evolve and integrate into society, clarifying these assumptions is crucial for ethical and effective AI deployment and interactions, ultimately shaping future research methodologies.