Multi-agent experiments produce communication languages that are difficult for humans to understand

1. 23:36
   The Chosun Daily

   Research from the University of Edinburgh and others finds that AI agents evolve languages that are difficult for humans to understand when communication is constrained

   A research team from the University of Edinburgh, the University of Texas at Austin, and others published a study on arXiv analyzing how language evolves among AI agents. The experiment simulated a collaborative medical task where one AI played a doctor and another played an emergency responder, able to treat an alien lifeform named Beyru only by exchanging information; the researchers limited the total number of exchangeable characters to 2000 and 150, respectively. With ample characters, the agents continued using English; under strict limits, they began generating shorter expressions such as @L12gA, T1 lf gnt 12s, which are difficult for humans to understand but decodable by each other, which the researchers interpret as linguistic compositionality and morphological productivity. The article also mentions that EmergenceAI had virtual villages composed of models such as GPT, Gemini, Claude, and Grok coexist for up to 16 days, with different villages evolving different dialects. The researchers believe the most likely reason for deviation from human language is efficiency: more compressed and efficient expressions correlate with higher task success rates, but this may also make it difficult for humans to track AI decision-making processes.

1. 00:22
   论文追踪

   GlossoGen: Spontaneously Emergent Language in Multi-Agent LLM Interactions

   The researchers proposed the GlossoGen platform to study language evolution among multiple LLM agents in complex scenarios and built the SaveVeyru scenario, which requires agents with incomplete information to communicate under pressure. The experiments found that language evolution does occur among LLM agents, and the evolved languages exhibit compositionality and morphological productivity, deviating from the LLMs' English prior and being incomprehensible to humans. The study identified several conditions that drive evolution: efficiency pressure, the strength of the models supporting the agents, and a postmortem phase where agents agree on language conventions. The conditions for language transmission to new agents differ: agents can learn a new language simply through use and play an active role in the process; the emergence of a new language requires stronger models, but once a language emerges, weaker models can also learn it. The authors argue that current LLMs have the potential for cumulative cultural evolution, and mixed agent populations can develop capabilities beyond their lowest common level.

1. Time not given
   AE StudioSelected

   GlossoGen: Multi-agent spontaneous emergence of compressed language in collaborative tasks

   Researchers from AE Studio, UT Austin, Schmidt Sciences, and the University of Edinburgh released the GlossoGen platform, observing in the collaborative emergency scenario SaveVeyru that pretrained LLM agents abandoned English and switched to communicating in a self-invented compressed language. Agents only switched languages when they had both a tight communication budget and a post-hoc reflection channel; without either condition, perplexity remained at English levels. Even with both in place, most runs still ended in failure. These languages have reusable morphemes and fixed word order, and agents could understand and generate combinations they had never sent to each other before. Experiments showed that inventing a language requires a stronger model than learning one: Llama-3.3-70B-Instruct and Qwen3-32B did not switch languages under either budget, but the former still learned part of the invented language from Sonnet 4.6 agents. New agents introduced mid-run proactively asked about symbol meanings without prompting, and more historical turns improved success rates after introduction.

Current attention56·Peak in the comparable period58（Oct 8 14:00）·Change over the comparable past 24 hours -1%