A research paper submitted to arXiv on May 29, 2026, and revised on June 1, 2026, challenges the validity of attributing human-like traits to large language models (LLMs). Author Adrian de Wynter argues that ascribing generalized anthropomorphic attributes—such as morality or deep natural language understanding—to LLMs may be empirically incorrect. To demonstrate this, the researcher trained a simple neural network on the strategy video game Age of Empires II. The study posits that if LLMs are deemed to possess human-like qualities based on their behavior, then any entity operating within a sufficiently powerful substrate, such as LEGO sets or the Greater Boston Area, could also be interpreted as having similar attributes.

The paper contends that these purported anthropomorphic characteristics are not unique to LLMs and that interpreting them depends heavily on the underlying system, or substrate, in which they function. The author shows that the game Age of Empires II is both functionally- and Turing-complete, suggesting that behavior observed in complex systems does not necessarily imply human-like cognition.

Consequently, the research argues that assuming these attributes exist independently of the system is often circular or uninformative. To address this, the author proposes a 'null' assumption for future research, which presumes LLM non-uniqueness rather than anthropomorphism when designing experiments. This approach aims to establish more explicit, empirically grounded criteria for evaluating the behavior of AI systems, moving away from subjective human interpretations.

A research paper submitted to arXiv on May 29, 2026, and revised on June 1, 2026, challenges the validity of attributing human-like traits to large language models (LLMs). Author Adrian de Wynter argues that ascribing generalized anthropomorphic attributes—such as morality or deep natural language understanding—to LLMs may be empirically incorrect. To demonstrate this, the researcher trained a simple neural network on the strategy video game Age of Empires II. The study posits that if LLMs are deemed to possess human-like qualities based on their behavior, then any entity operating within a sufficiently powerful substrate, such as LEGO sets or the Greater Boston Area, could also be interpreted as having similar attributes.

The paper contends that these purported anthropomorphic characteristics are not unique to LLMs and that interpreting them depends heavily on the underlying system, or substrate, in which they function. The author shows that the game Age of Empires II is both functionally- and Turing-complete, suggesting that behavior observed in complex systems does not necessarily imply human-like cognition.

Consequently, the research argues that assuming these attributes exist independently of the system is often circular or uninformative. To address this, the author proposes a 'null' assumption for future research, which presumes LLM non-uniqueness rather than anthropomorphism when designing experiments. This approach aims to establish more explicit, empirically grounded criteria for evaluating the behavior of AI systems, moving away from subjective human interpretations.