Скачать или смотреть Google DeepMind's Experimental Platform for Humans and LLM Agents | The Frontier Series: Episode 1

Google DeepMind's PAIR (People + AI Research) team present Deliberate Lab, an open-source platform for large-scale, real-time behavioral experiments that supports both human participants and large language model (LLM)-based agents, for conducting online research on human and LLM group dynamics.

In this conversation, Jerome Wynne, Senior AI Research Engineer at Prolific, sits down with Crystal Qian, Senior Research Scientist at Google DeepMind who led the team behind this research, to talk about creating LLM simulacra of human participants and the surprising finding that some models mirror human biases while others naturally select optimal leaders. We get into the design challenges of building AI agents that can participate in group conversations without dominating them, the negotiation study where LLMs and humans extracted similar value but through completely different strategies, and why aggregate alignment metrics can be dangerously misleading.

We also discuss the engineering challenges of synchronous online research, the video-game-lobby system they built to solve coordination problems, how a simple status indicator dramatically reduced participant attrition, and the unexpected finding that half their users didn't even want the AI features.

This is a conversation about what happens when you put humans and AI in the same room and try to make collective decisions, and what we're learning about both.

0:00 - Introduction
0:50 - Why Prolific for behavioral research
2:45 - What is Deliberate Lab
3:52 - The tooling gap in group dynamics research
7:18 - Lost at Sea: gender bias in leadership election
9:49 - Measuring confidence and competence
12:59 - Gender as a coordination mechanism
16:02 - LLM simulacra of human participants
20:33 - Where LLM conversations break down
22:00 - Mirroring vs. normative modes in models
24:35 - Solving the synchronous coordination problem
27:02 - What went wrong in early deployments
30:23 - Unexpected use cases from the research community
33:52 - AI facilitation for consensus building
38:17 - The negotiation and trading study
44:58 - Why aggregate alignment metrics are misleading
47:00 - LLMs as participants vs. tools
50:37 - Can AI make group conversations better or worse
53:17 - Designing agents for organic group interaction
57:22 - Eating your own dog food
59:43 - How human attitudes toward AI are changing

About the guest:

Crystal Qian is a Senior Research Scientist at ‪@googledeepmind‬, within the People + AI Research Group (PAIR). She leads a team investigating how LLMs can shape and improve social dynamics. Recent work includes simulating voting patterns in group elections, evaluating how LLM assistance can improve bargaining outcomes and group consensus, and developing scalable evaluation methods. Her current research interests involve human-AI interaction, agentic simulations, and societal impact, grounded through the analytical lens of game mechanics and behavioral experimentation.

Read the Deliberate Lab paper on ArXiv: https://arxiv.org/pdf/2510.13011v1
Learn more about Deliberate Lab: https://deliberate-lab.appspot.com/#/

Get the quality human data you need for AI research and development: https://www.prolific.com/ai?utm_sourc...

Connect with Prolific:
🔵 X:   / prolific  
🔵 LinkedIn:   / prolific-com  
🔵 Facebook:   / joinprolific  
🔵 Instagram:   / joinprolific  
🔵 Bluesky: https://bsky.app/profile/joinprolific...

#ai #deepmind #prolific