Microsoft researcher Kenneth Stanley built a functional neural network using goats, bridges, and ice ramps inside Age of Empires II's map editor. The unconventional project serves as a sharp critique of how AI researchers conduct their work.
Stanley's analysis of 315 papers revealed a methodological problem. More than half of the surveyed papers assume language models possess human-like traits before experiments begin. This assumption shapes their conclusions, creating circular reasoning that inflates the perceived capabilities of AI systems.
By constructing a working neural network from medieval game mechanics, Stanley demonstrates that any sufficiently complex system can appear intelligent when observers project human-like qualities onto it. The goat-based network functions mathematically, but calling it "intelligent" requires the same interpretive leap researchers make with language models.
The critique targets a widespread practice in AI science. Researchers often frame their work using language that presupposes human cognition in machines. They measure "understanding," "reasoning," or "common sense" without first establishing whether these terms apply to silicon-based systems. This linguistic choice influences how findings get reported and understood by the broader scientific community and the public.
Stanley's work exposes confirmation bias baked into AI research design. When you start with the assumption that a language model thinks like a human, you interpret its outputs through that lens. A pattern match becomes "reasoning." Statistical pattern matching becomes "understanding." The goat network proves the same phenomenon occurs regardless of what generates the patterns.
This isn't a claim that language models lack all meaningful capabilities. Rather, Stanley argues for epistemic honesty. Researchers should explicitly separate what large language models demonstrably do from what we assume they do based on surface-level similarities to human behavior. The distinction matters because it shapes which problems get funded, which approaches get pursued, and which claims reach the public.
The goat project gained attention because it uses absurdist humor to make a serious