Microsoft researcher builds goat-powered neural network in Age of Empires 2 to show why we should ‘stop assuming that LLMs behave like humans just because they were trained with natural language’
A Microsoft Principal Scientist has published a research paper that builds a fully functional neural network within Age of Empires II: Definitive Edition to make a pointed critique about AI hype and anthropomorphization.
Adrian de Wynter’s paper, titled “If LLMs Have Human-Like Attributes, Then So Does Age of Empires II,” uses the game’s custom map editor to construct working NAND gates and other computational logic gates. This unusual approach serves a serious purpose: to challenge the assumption that large language models possess human-like intelligence simply because they generate natural language responses.
De Wynter argues against the widespread tendency to perceive human-like qualities in artificial systems—a phenomenon driven partly by our psychology and partly by AI company messaging. The paper suggests that both LLMs and non-intelligent systems can produce outputs that appear human-like without possessing genuine cognition.
The research has gained viral attention, reflecting growing concerns among researchers about LLM misconceptions. De Wynter’s work demonstrates that sophisticated outputs don’t require consciousness or human-level intelligence—they merely require sufficient computational complexity. By showing that even a video game can serve as a platform for building neural networks, the paper makes its philosophical point clear: computational behavior shouldn’t be conflated with human cognition.