Microsoft researcher builds a working neural network out of goats in Age of Empires II to critique AI science

What changed

A Microsoft researcher built a functioning neural network inside Age of Empires II using only goats, bridges, and ice ramps created in the game’s map editor. This is not a programming stunt but a critique aimed at AI research practices. By replacing digital neurons with virtual goats wandering a structured game map, the project proves the underlying mathematics of neural networks do not depend on the human-like appearance or conversational interface often assumed in language model studies.

Why builders should care

The researcher analyzed 315 AI papers and found that more than half begin by treating language models as if they have human intentions or understanding before any testing. This biases conclusions about AI capabilities. For builders and operators evaluating or developing AI systems, it signals a need for clearer criteria that separate engineered math from anthropomorphic interpretations. Misreading models as “thinking” entities can lead to overestimating reliability and risks.

The practical takeaway

AI tools and interfaces can behave identically while appearing fundamentally different. Whether a language model is accessed as chat text or through wandering goats in a game, the underlying processing remains statistical math. This work pressures AI developers and product designers to rethink how user experience affects perception and trust in systems. It also encourages AI policymakers and investors to demand rigor that avoids projecting human traits onto models prematurely.

What to watch next

Expect more critiques challenging AI hype narratives based on anthropomorphizing models. Engineers and researchers will need to develop objective benchmarks that do not assume human-like agency in language models. For product teams, this means examining AI features through a practical performance and risk lens rather than emotional or conversational appeal. The real test will be how these lessons influence AI development standards and user expectations.

AI Quick Briefs Editorial Desk