Overview
- Microsoft researcher Adrian de Wynter published an essay, video demonstrations and code last weekend (June 20–21) showing a working NOT‑AND gate and a 1‑bit perceptron built inside Age of Empires II using terrain as bits and goats as signal carriers.
- De Wynter used the game’s scenario editor to assemble basic logic and training circuits to demonstrate how computation can run on an unexpected substrate and to argue that the same surface behaviors that prompt us to call LLMs ‘human‑like’ can arise from framing and implementation.
- As part of the paper he reviewed more than 300 recent AI papers and reported that a majority began with human‑like assumptions about LLMs and that studies explicitly testing for human attributes often concluded they existed.
- He made all materials public on GitHub to let others reproduce and inspect the work and to press for a scientific change: tests should start from a null assumption and be designed to prove human‑like claims rather than assume them.
- The demonstration renews debate over anthropomorphism and research bias, highlights the concept of substrate‑independent computation such as Turing‑completeness, and could prompt calls for stricter experimental design though no institutional policy changes have been reported yet.