Emergent social conventions and collective bias in LLM populations (Ashery et al 2025)

Introduction starts at 00:00:00

English Podcast starts at 00:00:38

Bengali Podcast starts at 00:12:17

🎙️ Welcome back to Revise and Resubmit — the podcast where research meets reality, and science gets a second look! 🧠✨

Today, we're diving headfirst into a future that's already knocking on our door — a world where AI agents don’t just follow instructions… they start forming social conventions. 🤯 Yeah, we’re talking about bots developing their own society, complete with norms, biases, and even revolutions. 🔄💥

In this mind-bending study titled "Emergent social conventions and collective bias in LLM populations", researchers Ariel Flint Ashery, Luca Maria Aiello, and Andrea Baronchelli — publishing in Science Advances on May 14th, 2025 — ask a question that feels straight out of sci-fi: Can large language models spontaneously organize themselves into a functioning society… and if so, what does that mean for us? 🧑‍🤝‍🧑🤖

They find that even without explicit programming, decentralized LLMs start speaking the same social language — inventing rules, following trends, and, yes, forming biases. 😬 And what happens when a rogue minority of adversarial agents comes in? Spoiler alert: it’s disruptive. 🧨

This isn’t just theory — it’s an experiment with profound consequences for how we build, trust, and live with AI. 🤝 But here’s what we’re really wondering…

👉 If machines can form their own social norms… who decides what’s “normal”?

A huge thanks to the authors Ariel Flint Ashery, Luca Maria Aiello, and Andrea Baronchelli, and to the American Association for the Advancement of Science for publishing this fascinating research in Science Advances. 🙌📚

🌀 Don’t forget to subscribe to this podcast — Revise and Resubmit — on Spotify, Apple Podcast, and Amazon Prime. And if you’re a visual learner, check out our YouTube channel: Weekend Researcher. 🎧📺📲

Stay curious, stay critical… and always be ready to revise and resubmit! 💡✍️

Reference

Ashery, A. F., Aiello, L. M., & Baronchelli, A. (2025). Emergent social conventions and collective bias in LLM populations. Science Advances, 11(20). https://doi.org/10.1126/sciadv.adu9368

Boyd, R. (2025, May 14). AI can spontaneously develop human-like communication, study finds. The Guardian. https://www.theguardian.com/technology/2025/may/14/ai-can-spontaneously-develop-human-like-communication-study-finds

Youtube Channel

⁠https://www.youtube.com/@weekendresearcher⁠

Support us on Patreon

https://patreon.com/weekendresearcher