A Microsoft AI researcher used Age of Empires II to challenge claims about AI consciousness. His experiment questions how easily people attribute human-like traits to large language models.
Media and technology professionals are increasingly confronted with claims about artificial intelligence possessing human-like qualities. A recent experiment by a Microsoft AI researcher highlights how easily people can misinterpret the capabilities of large language models (LLMs), raising important questions for those who build, monetize, and distribute AI-driven content.
Adrian de Wynter, a researcher at Microsoft, constructed a basic neural network inside the classic real-time strategy game Age of Empires II. Using the game's scenario editor, he created a functioning NAND gate and a 1-bit perceptron, with digital goats serving as the bits. De Wynter explained that this setup mimics the core technology behind LLMs like Claude, ChatGPT, and CoPilot, though in a simplified form. Grass represented 0, bridges represented 1, and goats acted as signal carriers. When the gate operated, goats were removed and new ones placed to represent output, demonstrating the mechanics of a neural network within the game.
De Wynter's work was motivated by what he described as a widespread tendency to anthropomorphize AI systems. He reviewed 315 computer science papers from the past two years and found that 57 percent began with the assumption that LLMs have human-like traits. He argued that this bias shapes research design, interpretation, and conclusions, often leading to overstatements about AI's capabilities.
He pointed out that if people can build a neural network in Age of Empires II that functions like those in popular chatbots, but no one claims the game's goats are conscious, then similar claims about LLMs should be scrutinized. The difference, he said, is the interface: interacting with an LLM through a chat window encourages users to see human-like qualities, while the same technology embedded in a video game does not evoke the same perception.
De Wynter has played Age of Empires II since its 1999 release and chose it for its familiarity and its ability to illustrate the gap between representation and interpretation. He noted that the problem of anthropomorphizing LLMs often starts before research even begins, as researchers and users alike project human attributes onto these systems simply because they process natural language.
He also addressed the commercial incentives behind these perceptions. De Wynter observed that people are more likely to buy products-whether toasters, phones, or LLMs-when they can empathize with them. He suggested that marketing often amplifies claims about AI's capabilities, as companies seek to differentiate their products in a crowded market.
Executives and scientists have contributed to the debate by making public statements about AI consciousness. For example, OpenAI CEO Sam Altman has implied that developing LLMs could lead to the creation of an "AI god," while Anthropic CEO Dario Amodei has said he cannot be certain whether AI is conscious. These statements, according to De Wynter, further blur the line between technical reality and public perception.
De Wynter does not dismiss the possibility that LLMs could possess some form of consciousness, but he emphasized that the criteria for such judgments are often subjective and based on human-like benchmarks. He encouraged researchers and users to focus on the actual properties of LLMs, rather than projecting expectations shaped by human experience.
He recommended that companies provide clear disclosures and alignment techniques to help users understand the true nature of LLMs. De Wynter noted that users often become attached to AI systems that appear to have warmth or personality, which can influence both user behavior and purchasing decisions.
For those interested in how brands leverage cultural connections to drive engagement, a recent story on Dentsu's relaunch of 360i as a social-first solution for US brands explores similar themes of perception and audience attachment. Read more in this coverage of Dentsu's strategy.
