Researcher Uses Goats in Age of Empires 2 to Debunk AI Anthropomorphism

A Microsoft AI researcher has demonstrated how humans often attribute human-like qualities to large language models. Adrian de Wynter built a neural network inside the game Age of Empires 2 using goats to challenge this assumption. His work highlights a tendency to anthropomorphise LLMs without evidence. De Wynter’s paper argues that people too readily assume LLMs have human traits. His review found that over half of the computer science papers he examined made this assumption without proof. To illustrate his point, he created an unconventional neural network within Age of Empires 2.

The system uses grass to represent 0, bridges for 1, and goats as bits. It includes a working NOT AND gate and a 1-bit perceptron. Unlike typical LLMs, its outputs and behaviours do not resemble human thought, largely due to its reliance on goats and grass. De Wynter chose Age of Empires 2 because it is a less common platform for such experiments compared to games like Minecraft.

Videos of the goat-powered LLM are available on his GitHub page. He proposes that LLMs should be studied as they are, not as human-like entities. The project serves as a clear example of non-human LLM behaviour. It reinforces the idea that these models lack consciousness. De Wynter’s work encourages a more objective approach to understanding artificial intelligence.