Study Uses Age of Empires Editor to Show Dangers of Treating AI as Cognizant

Recent documentaries have tried to demystify how computers actually work. Dylan Reibling’s The End of the Internet argues that phrases like “cloud computing” obscure the physical infrastructure—cables, valves and the like—while Amanda Kramer’s So Unreal traces how fantasies from The Matrix and Terminator shape public ideas about cyberspace and the technology built around them.

Programmer Adrian de Wynter took that critique into a playful but pointed experiment. In a paper titled “If LLMs Have Human-Like Attributes, Then So Does Age of Empires II,” he used the game’s scenario editor to replicate basic functions of language learning models with goats, grass and bridges.

De Wynter says he was shocked that 57% of the more than 300 papers he reviewed began by assuming AI displayed something approximating consciousness. He avoids the term “artificial intelligence,” preferring “language learning model,” and demonstrates how removing a personable chat panel leaves behind the alienating code we’re used to.

llm, age of, scenario editor, language model, goats, cloud computing, the matrix, consciousness, chat panel, amanda kramer