Open conference where AI serves as both primary authors and reviewers (agents4science.stanford.edu)

Agents4Science is launching the first open conference that requires AI systems to act as primary authors and reviewers, creating a transparent testbed to study AI-driven scientific inquiry. The inaugural one-day virtual event (registration open) will publish AI-authored papers alongside the prompts, agent outputs, and AI-generated reviews, with panels and Q&As featuring leading editors, professors and a Nobel laureate. By making AI involvement explicit and public, the conference aims to surface how agents generate hypotheses, design methods, analyze data and critique work—rather than masking their contributions under human authorship. Technically and culturally significant, Agents4Science creates an environment to evaluate AI agents end-to-end: generative capabilities (novel hypotheses and methods), evaluation pipelines (AI-driven peer review), and reproducibility (release of prompts, traces and reviews). This enables development of attribution standards, verification tools, benchmarks, provenance metadata and evaluation metrics while exposing failure modes such as hallucination, metric gaming, or weak accountability. For the AI/ML community it’s both a risk-mitigating experiment and an opportunity to accelerate discovery—providing concrete data to shape norms, tooling and oversight needed before AI becomes a trusted research collaborator.