Full Breakdown
Goat-Based Network in Age of Empires II Shows AI Anthropomorphism
6/18/2026, 11:13:01 PM
Goat-Based Neural Network in Age of Empires II
Adrian de Wynter, a Microsoft AI researcher, used *Age of Empires II*’s scenario editor to build a NAND gate and a 1-bit perceptron. Grass encoded 0, bridges encoded 1, and goat units acted as moving bits; when gate fired, input goats were removed and new goat appeared on output rail, reproducing a neural network. De Wynter says this shows LLM-type computation can be realized in a game.
Context
The experiment tackles the LLM consciousness debate. Ted Chiang likened claims of conscious LLMs to believing Microsoft Word contains dormant interlocutors. Researchers and marketers ascribe anxiety, morality, or personality to chat models, reinforcing a perception that they think like humans. De Wynter’s goat model serves as a reductio ad absurdum, urging a clear separation between algorithmic function and anthropomorphic language.
Figures
Key actors are Adrian de Wynter, a Microsoft AI researcher, and Microsoft, which provides the platform. AI field includes OpenAI, Anthropic and other firms whose products (Claude, ChatGPT, Copilot) are described with human-like traits. Executives Sam Altman, Ilya Sutskever and Dario Amodei have publicly discussed AI consciousness.
Data
De Wynter reviewed 315 computer-science papers from the last two years; 57 % began with the assumption that LLMs have human-like attributes, shaping design and interpretation.
Statements
OpenAI CEO Sam Altman has suggested advancing LLMs could be a step toward an “AI god.” Ilya Sutskever has spoken of LLMs as possessing god-like consciousness. Anthropic CEO Dario Amodei told *The New York Times* he cannot be certain AI is conscious. De Wynter urges disclosures and alignment to clarify model nature, arguing transparency can curb attachment.
Opposition
Same executives who promote AI roadmaps also frame LLMs as quasi-sentient, using language that emphasizes warmth and personality. This framing encourages users to form emotional bonds with chat agents, a phenomenon de Wynter likens to attachment to fictional characters rather than appliances.
Quotes
- “I have this tendency to dial up things to 11 when I really think I need to make a point.” — Adrian de Wynter, Microsoft AI researcher
- “The point of the paper is to formally show that we anthropomorphise too readily, and that sometimes the claims we make with regards to LLM capabilities are too strong.” — Adrian de Wynter
- “I don’t get attached to my toaster, but I definitely get attached to characters on a movie screen.” — Adrian de Wynter
- “Only one rail is active at a time, with a goat acting as the signal carrier.” — Adrian de Wynter
Next Steps
De Wynter recommends experiments that treat LLMs as computational systems rather than human proxies and urges developers to embed disclosures about model capabilities. Adoption of these practices could narrow the gap between functional performance and perceived consciousness, reducing hype-driven expectations while preserving scientific inquiry.
