A researcher built a neural network of goats, bridges, and paths in Age of Empires 2. The idea conveys a message: Just because an AI seems intelligent, it doesn’t mean it thinks or feels like a human.
Those who talk or write with ChatGPT, Claude, Gemini, and others quickly feel they are facing real intelligence. The AIs write texts, answer questions, and sometimes even seem empathetic.
For Microsoft researcher Adrian de Wynter, that’s exactly the problem. Just because a machine seems convincing does not mean, in his opinion, that it thinks or feels. (GitHub)
To demonstrate this, he initiated an unusual experiment: he built a neural network in Age of Empires 2 – with goats.
What exactly did the researcher build? De Wynter used goats, paths, and bridges in the strategy game to recreate a very simple version of a neural network.
The animals take on the role of individual neurons and move through the network according to fixed rules. The system can only solve very simple tasks, but it fulfills the same basic idea as modern AI: information is processed, leading to a result.
Simply put, the goats act as switches: they trigger certain actions or block them. Depending on which path they take or which bridge is passable, information is passed on – very similarly to artificial neurons that receive, process, and pass on signals to other neurons.
However, the researcher did not want to prove that goats are intelligent. He wanted to show how easily humans tend to attribute more to a system than what is actually there. So the research thesis goes.
Why should goats say something about ChatGPT?
The researcher’s thesis is provocative: If one assigns consciousness to an AI solely because it understands language and gives convincing answers, then theoretically one should also assign consciousness to their goat network.
Of course, no one seriously believes that the animals in Age of Empires think. For this reason, de Wynter considers the conclusion that modern language models must inevitably possess some form of consciousness to be wrong.
In his view, the ability to process information or sound human does not yet prove that a machine has feelings, desires, or a sense of self. (GitHub)
On De Wynter’s GitHub page, there are videos showing the goat LLM in action – to outsiders, the whole thing seems incomprehensible, which proves the point for de Wynter: The same mechanisms that drive ChatGPT or Claude operate here. Just with goats, and of course simplified.
The collected results and the analysis of the data can be read here for anyone interested.
What do you think of such experiments? Let us know in the comments if you believe AI can really develop consciousness someday?
A few goats in Age of Empires will not revolutionize AI research. However, they show remarkably well how quickly humans tend to see more in machines than may actually be there. Because while AI seems increasingly human, many people use the technology not just for work or learning, but also for personal conversations and emotional closeness.
Your opinion is important to us!
Do you like the article? Then let us know!