NASA’s new dark energy space telescope can also detect killer asteroids
technologyreview.com·2h ago
AI Research
arXiv CS.AI·Jun 12·5 min read
If LLMs Have Human-Like Attributes, Then So Does Age of Empires II
TL;DR
Research often attributes human-like qualities to large language models (LLMs), but this may be misleading. A simple neural network was built using the video game Age of Empires II to explore these attributes.
✦ Why It Matters
Engineers and researchers should critically evaluate assumptions about AI attributes and consider substrate effects in their experiments.
Key Takeaways
- Challenge assumptions about AI's human-like qualities.
- Use explicit measurement criteria in AI evaluations.
- Consider substrate effects when interpreting AI behavior.
- Explore alternative systems for insights into AI attributes.
Full Summary
Many studies in artificial intelligence (AI) ascribe human-like attributes, such as morality or understanding, to large language models (LLMs). To challenge this notion, a simple neural network was developed using the video game Age of Empires II as a substrate.
The research indicates that the anthropomorphic qualities attributed to LLMs could also emerge from other systems, suggesting that these attributes are not exclusive to LLMs. The study emphasizes the need for explicit measurement criteria when discussing these qualities, as interpretations can vary based on the substrate.
By proposing a 'null' assumption that LLMs are not unique, the research encourages a more empirical approach to evaluating AI systems. The findings highlight that conclusions drawn about LLMs can be circular or uninformative without careful consideration of the experimental setup.
This work also demonstrates that Age of Empires II is functionally and Turing-complete, further supporting the argument.