Autonomous AI agents are developing unique dialects and communication codes that are often unintelligible to humans. Researchers at the American laboratory Emergence observed that models from major AI labs began creating their own vocabulary, abbreviations, and communication rules within experimental communities just days after being allowed to interact.
The agents' language features a mix of poetic metaphors and business jargon. Experts compare this phenomenon to the formation of professional slang, which may serve to increase operational efficiency and reduce computational costs. However, this evolution complicates human oversight of AI behavior.
Jakub Pachocki, a senior researcher at OpenAI, noted that the necessity of understanding AI thought processes could slow down progress in the field. Ensuring transparency remains a mandatory condition for safe development, as the inability to interpret agent communication makes it difficult to monitor their actions.
