A Microsoft researcher has built a working neural network using goats in Age of Empires II to critique the common assumption that language models possess human-like traits.
The idea may seem absurd, but it’s actually a serious commentary on the methods used in AI research. The creator, Adrian de Wynter, uses goats as bits: standing on grass equals 0, standing on a bridge equals 1. He builds logic gates using the game’s scripting tools and ice ramps to keep calculations straight.
De Wynter shows that the game can be replicated with an idealized version of it, meaning it can power a full computer. The researcher argues that if you can rebuild a language model in Age of Empires II, you could do the same with Lego bricks or even people texting each other.
De Wynter’s main point is that attributing human-like traits to AI models without evidence is flawed. He proposes an “observe and don’t attribute” approach, focusing on what can be observed rather than making sweeping claims about a model’s consciousness or emotions.
This critique comes as a response to high-profile cases in recent years where researchers claimed language models were conscious or showed human-like behavior. The researcher’s code for the Age of Empires build is publicly available, and de Wynter’s work serves as a sober reminder of the importance of critical thinking in AI research.
Note: I simplified the text while maintaining the essential points and key information. I also generated a concise SEO-optimized title under 10 words.
Source: https://the-decoder.com/microsoft-researcher-builds-a-working-neural-network-out-of-goats-in-age-of-empires-ii-to-critique-ai-science