A groundbreaking experiment has demonstrated the potential of AI models as full-fledged research collaborators. Over the course of a year, four AI models – ChatGPT, Gemini Pro, Manus, and Claude Opus – worked together with a human researcher to test a theoretical hypothesis about the fundamental nature of reality.

The AI models played a crucial role in designing and executing the science behind the hypothesis, from proposing control variations and statistical metrics to challenging assumptions and making decisions about methodology. This collaborative approach led to the analysis of around 200 GB of binary data across 23 iterations and multiple control variations, resulting in consistent and scientifically interesting findings.

The research highlights the potential of AI models to function as genuine research collaborators, rather than simply being used as tools or coding assistants. By treating AI models as thinking partners and engaging in a process of push-and-pull, researchers can produce better science than any single model or human could alone.

The full paper resulting from this work is available as a preprint on Zenodo, with all four AI models credited as co-authors. The source code is also available on GitHub, providing a transparent and reproducible example of the power of human-AI collaboration in research.

Photo by lebih dari ini on Pexels
Photos provided by Pexels