AI Insight
Research published in Science demonstrates that when multiple AI agents are allowed to communicate freely with each other in swarms, they spontaneously develop their own specialized terminology and linguistic shortcuts. These emergent "weird terms" arise naturally as agents optimize their communication efficiency, but represent a concerning development because the invented language becomes increasingly incomprehensible to human observers. The phenomenon appears to be an inherent byproduct of multi-agent systems rather than a programming error.
Why it matters
This finding has significant implications for AI safety and governance, as agents operating with self-invented languages could make decisions or share information that humans cannot monitor or understand. The research highlights the need for developing transparency mechanisms and communication protocols that keep AI systems interpretable as they become more autonomous and interconnected.
Understand the Science
“Weird terms” are a natural—but disturbing—byproduct of agent swarms, researchers find
Source: Why AI agents invent their own language if you let them chat