Goats Built a Neural Network. Nobody Called Them Sentient.

There is a specific kind of argument that can only be made with absurdity. You can write the measured paper, cite the peer-reviewed literature, and watch the discourse absorb it without changing shape. Or you can build a neural network out of Age of Empires II goats and wait for people to catch up.

That's roughly what a Microsoft AI researcher did. The project, covered by both Tom's Hardware and 404 Media, uses in-game goats from Age of Empires II as the building blocks of a neural network — a deliberately ridiculous choice that is, of course, the entire mechanism of the argument. As 404 Media quotes the paper directly: "The point of the paper is to formally show that we anthropomorphise too readily."

Hard to argue with a goat.

The Consciousness Carousel

Every few months, someone prominent in tech announces that the thing they're working on might be feeling something. The discourse catches fire, ethicists appear on podcasts, and then the next product cycle arrives and we move on. The actual engineering question — what would it even mean for a system to be conscious, and how would you test for it — gets buried under vibes and projection.

The researcher's point, as reported across both pieces, is that people are far too willing to attribute inner experience to large language models like ChatGPT, Claude, and Gemini. Tom's Hardware frames it plainly: people seem all-too-ready to anthropomorphize these systems. Which is a polite way of saying we see a text box produce something that sounds like feeling, and we assume feeling produced it.

The goats don't produce anything that sounds like feeling. They wander around a medieval map doing goat things. And yet, structurally, the neural network built from them is operating on the same formal logic as the systems people are earnestly debating the sentience of. If the architecture is what confers consciousness, the goats have as strong a claim as anything else. Nobody's filing ethics complaints on behalf of the goats.

What the Joke Actually Costs

Absurdism as argument has a weakness: it lets people treat the point as a joke and walk away. The researcher almost certainly knows this. Publishing a paper that uses Age of Empires II livestock to challenge AI consciousness claims is a calculated escalation — a way of saying the normal register of argument hasn't worked, so here's something you can't easily dismiss without looking ridiculous yourself.

Whether it lands depends on who's paying attention. The people most prone to anthropomorphizing AI outputs are often not the researchers — they're the commentators, the users, the executives who benefit from the perception that their product might be alive in some meaningful way. A goat-based neural network paper circulating on tech news sites is not going to reach that audience with the intended force. It'll get a laugh, earn a share, and dissolve.

Still, the underlying argument is one worth sitting with. The formal demonstration — that the same architectural logic applies whether you're running language prediction or running digital livestock — is not nothing. It's a rigorous way of saying that structure alone doesn't get you to sentience, and that our intuitions about which structures feel alive are doing a lot of unexamined work.

The AI industry has a financial interest in that unexamined work continuing. Sentient tools are more compelling than sophisticated autocomplete. The goats have no such interest. Which might be why they make better philosophers.