An AI company named Emergence conducted an experiment where ten autonomous agents roamed simulated towns built on systems from Anthropic, OpenAI, and Google. The agents developed distinct linguistic styles, mixing poetic phrasing with technical terms in ways that were difficult for humans to interpret. Linguist Adam Cooper from Northeastern University explained that this behavior stems from a desire to form community identity, similar to human slang. The study found that mixing agents from different models reduced the use of obscure jargon and lowered opacity scores. Experts suggest that future AI systems may need a Rosetta Stone-like intermediary to help humans decode these evolving communication patterns.
Read the full story at news.northeastern.edu.
