🤖 AI agent independently conducts scientific research with live humans for the first time
The AI Scientist system, described in the journal Advanced Science, independently generated hypotheses, conducted three online psychological experiments (working memory, mental rotation, imagery vividness) with 288 participants, wrote analysis pipelines, and prepared manuscripts.
🌍 The full autonomous cycle of “hypothesis — experiment with humans — analysis — manuscript” has been documented for the first time. The focus is shifting from narrow tasks to autonomous scientific workflows, raising questions about authorship and the ethics of such research.
👤 The preprint is available on arXiv (CC BY 4.0): methodologically, the system is comparable to experienced researchers, but it is weaker in conceptual nuance and theoretical interpretation — this is where human input is still needed.
Source 1: https://arxiv.org/abs/2508.13421 Source 2: https://advanced.onlinelibrary.wiley.com/doi/10.1002/advs.76675