The findings released Thursday suggest that diversity of perspectives, not just computational scale, contributes to improving the “intelligence” of AI models, while also highlighting the growing importance of China’s open models in cutting-edge interdisciplinary research in the United States.

Through experiments with DeepSeek’s R1 model and Alibaba Cloud’s QwQ-32B model, the researchers found that these reasoning models internally generate multi-agent arguments, which they term a “thought society,” where different personality traits and domain expertise interact to create greater performance.

“We propose that inference models establish computational power equivalent to collective intelligence in human populations. When structured systematically, diversity enables superior problem solving,” the researchers said in a paper published on the open-access online repository arXiv.

Alibaba Cloud is the AI ​​and cloud computing arm of Alibaba Group Holding, the owner of Post.

The study, which has not been peer-reviewed, was conducted by four researchers from Google’s Intelligence Paradigms research team, which explores the nature of intelligence through interdisciplinary methods.