A Perceptron in Age of Empires II
A developer demonstrates building functional NAND gates, a perceptron, and even a training circuit within Age of Empires II, proving the game's functional and Turing completeness. This technically impressive and highly unconventional feat uses in-game assets like goats and terrain to represent bits and logic. The project then leverages this demonstration to make a philosophical argument about how the substrate of an LLM influences the perception of its anthropomorphic attributes.
The Lowdown
This article details the impressive feat of constructing a perceptron and its training circuit entirely within the confines of Age of Empires II. Leveraging the game's scenario editor, the author demonstrates how basic game mechanics can be manipulated to create complex computational logic, ultimately connecting this intricate build to a broader philosophical discussion on the nature of AI.
- The project begins by proving Age of Empires II is functionally and Turing-complete, laying the theoretical groundwork for in-game computation.
- NAND gates are constructed using 'goats' as signal carriers, and terrain features like 'grass' and 'bridges' to represent binary 0s and 1s, with 'ice' rails managing gate readiness and race conditions.
- A simple bipolar 1-bit perceptron is then built, using XNOR gates for inner product calculations and an AND gate for thresholding, with goats also managing concurrency.
- An ansatz-based training circuit is also implemented, employing various terrain types (bamboo for XNOR, forests for AND, baobabs for OR, deep water for NOT) to compute error and update weights.
- Beyond the technical demonstration, the work argues that changing the 'substrate' or representation of an LLM alters how we perceive its attributes, suggesting that research into LLM anthropomorphism must avoid presuming these attributes exist from the outset.
This unconventional yet brilliant project not only showcases the hidden computational power within a classic RTS game but also offers a tangible metaphor for a nuanced philosophical debate about AI and human perception.
The Gossip
Philosophical Perceptions
The paper's central philosophical argument, concerning how an LLM's substrate affects the perception of its anthropomorphic attributes and the implicit assumptions in asking about such properties, sparked considerable debate. Some commenters agreed with the premise that asking a question like 'Do LLMs have empathy?' inherently assumes the possibility of empathy in LLMs, while others strongly disagreed, arguing that merely asking a question does not equate to making an assumption about the answer. The overall philosophical depth and rigor of the paper were also questioned, with some suggesting a lack of engagement with existing philosophical literature on emergence and materialism.
Engineering in Empires
Commenters expressed significant admiration for the sheer technical ingenuity required to implement logic gates and a perceptron within Age of Empires II. Many discussed the game's robust map editor, its trigger/effect system, and more advanced scripting capabilities (like the XS language in the Definitive Edition and Python tools) that enable such complex constructions. There was a strong sense of nostalgia and appreciation for the creativity that the game's engine has fostered over the years.
Link Logistics & Legacy Learning
Several users noted that the initial link to the ArXiv paper was broken, which was quickly remedied by another commenter providing the correct URL. The discussion also broadened to include related projects and games that have historically incorporated neural networks, such as 'Creatures' (1996), highlighting a long-standing fascination with integrating AI concepts into game design.