Microsoft researcher uses "Age of Empires" goats to make a point about AI consciousness
A Microsoft researcher has built a functioning neural network inside Age of Empires II using digital goats to prove that AI can run on anything. The absurdist experiment highlights how easily human-like chat interfaces trick users into believing AI is truly sentient.
-
A Microsoft researcher used digital goats to show that AI logic can run in surprisingly unconventional systems.
-
The experiment suggests chat interfaces may make AI seem more human than it really is.
-
Human-like conversation alone is not evidence that an AI is conscious or sentient.
Microsoft researcher Adrian de Wynter built neural network components into Age of Empires II, in which goats served as signal carriers for the AI to function.
In the game, the characters move when you give them commands. The researcher realized he could use this to create logic gates (the tiny switches inside a computer chip that process information).
And for the case of AI, instead of electrons flowing through a copper wire, it’s digital goats walking down a path.
De Wynter argues that neural-network logic can be recreated in unusual subjects, even if the goats sound a bit gimmicky.
Plus, the researcher argues that a friendly chat interface may be doing more work than the AI itself to appear intelligent.
The main point here is that AI doesn’t need big, expensive chips to be operational. The same principles can be created with Lego bricks or large groups of people.
“Many anthropomorphic measurements in AI are measurements of presentation, rather than of an actual system’s behavior,” argued de Wynter, revealing that chatbots are largely of a sugar-coated nature, and great at telling you what you want to hear.
Why this matters
The researchers claim that humans naturally anthropomorphize things, which basically means that the nature of chat interfaces encourages users to treat AI like people.
By changing the presentation while keeping the computation, the author is trying to isolate how much anthropomorphism comes from the interface rather than the system itself.
And as human-like language creates an illusion of understanding – my Gemini assistant has started calling me “mate” out of nowhere – the same output from goats would feel less convincing.
So, as a kind of sarcastic case in point, the goats show us that we have to stop holding up a mirror to AI, as it bleats out its content all the same.
