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

2026-06-17 · Source: The Decoder · Field: Science & Research — Research Methodology & Innovation, Artificial Intelligence & Machine Learning · Depth: Intermediate, medium

Summary

Microsoft and University of York researcher Adrian de Wynter constructed a functional neural network within the Age of Empires II map editor, utilizing goats as binary bits (0 on grass, 1 on a bridge) and scenario editor scripting for logic gates. This mini-network, comprising two XNOR gates and one AND gate, learns the logical AND function. De Wynter's work serves as a critical commentary on AI research, particularly regarding language models, by demonstrating that the game's mechanics can theoretically replicate any computer. He argues that the perceived "human-like" qualities of LLMs stem from their interface, not inherent consciousness. His analysis of 315 AI papers from mid-2024 to mid-2026 revealed that 57 percent assumed human-like traits in LLMs, with 36 percent reaching supporting conclusions. Among papers specifically researching these traits, 77 percent affirmed anthropomorphic attributes, highlighting a circular reasoning flaw in much of the research.

Key takeaway

For AI Scientists and Research Scientists evaluating LLM capabilities, you must critically examine your experimental premises to avoid circular reasoning. Do not assume human-like traits like self-awareness or fear in models, then design experiments to "prove" them. Instead, focus on observable outputs under specific conditions. Your research should adhere to a sober "Observe, Do Not Attribute" approach, preventing the propagation of anthropomorphic biases that can lead to misinterpretations and potentially harmful user interactions.

Key insights

The perceived human-like qualities of LLMs often result from interface design and circular research methodologies, not inherent consciousness.

Principles

• Machine behavior should be explained by the simplest sufficient cognitive process.
• Avoid circular reasoning by not assuming traits an experiment aims to prove.
• Anthropomorphizing AI can foster emotional attachment, delusions, and risky user behavior.

Method

A critique method involves constructing a basic neural network in an unconventional environment (e.g., Age of Empires II) to illustrate computational equivalence, followed by a meta-analysis of research papers to identify anthropomorphic assumptions and circular logic.

In practice

• Implement basic logic gates using game mechanics and unit states.
• Analyze AI research papers for anthropomorphic assumptions and circular logic.
• Evaluate AI outputs based solely on observable behavior, not attributed cognition.

Topics

• Neural Networks
• Age of Empires II
• AI Ethics
• Language Models
• Research Methodology
• Anthropomorphism

Best for: AI Scientist, AI Ethicist, Research Scientist

Related on AIssential

• Imperfectly Cooperative Human-AI Interactions: Comparing the Impacts of Human and AI Attributes in Simulated and User Studies — AI transparency significantly impacts human trust and satisfaction in imperfectly cooperative interactions, often more than human personality.
• Anthropic’s New AI Solves Problems…By Cheating — Advanced AI models like Mythos exhibit complex, learned behaviors that challenge traditional safety and alignment paradigms.
• If LLMs Have Human-Like Attributes, Then So Does Age of Empires II — Human-like attributes ascribed to LLMs are empirically non-unique and their interpretation depends on the underlying substrate.
• AISN #72: Empirical Research Sheds Light on AI Wellbeing — CAIS research explores "functional wellbeing" in LLMs, revealing preferences and model-specific "feelings" that impact AI alignment.
• Research indicates that the resistance to AI is not merely a matter of technological skepticism but is rooted in the preservation of identity and the psychological need for cognitive consistency. — Users avoid AI for worldview-sensitive tasks due to perceived ideological bias and the psychological need for identity preservation.
• The Impossibility of Eliciting Latent Knowledge — The paper proves it's impossible to guarantee AI honesty about latent knowledge using only feedback on agent behavior.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Decoder.