Workshops Archive

From Candidate to Confirmed: A Verification-First Agenda for Open Autonomous Science


Workshop: 4th Annual Trillion Parameter Workshop: Building Open AI Infrastructure, Models, and Agentic Systems for Science

Authors: Rafael Ferreira da Silva (Oak Ridge National Laboratory (ORNL)); Laura Biven (Jefferson Lab); Ian T. Foster (Argonne National Laboratory); Tirthankar Ghosal (Oak Ridge National Laboratory); Tom Gibbs (NVIDIA); Benjamin Mintz, Woong Shin, and Frédéric Suter (Oak Ridge National Laboratory); and Michela Taufer (University of Tennessee, Knoxville)

Abstract: Automation, robotics, and LLMs are reshaping scientific discovery, generating hypotheses, designs, and analyses faster than humans can review them. Over the past year, multi-agent systems have produced laboratory-validated hypotheses, and interface standards like MCP and A2A have begun connecting agents to scientific tools and to one another. Progress on generation has outrun the infrastructure needed to trust it. Agents nearing expert performance on closed-ended questions complete only a small fraction of open-ended research tasks, no system has self-validated a novel discovery end to end, and a widely cited autonomous-discovery result was corrected after its novelty claims were contested. Verification, not generation, is now the binding constraint on autonomous science, and the open-model community is well positioned to address it. We propose a verification-first agenda of open, reproducible evaluation, end-to-end provenance for agentic workflows, and interoperable, accountable agent interfaces, artifacts that belong in shared open infrastructure alongside the models themselves.


Back to 4th Annual Trillion Parameter Workshop: Building Open AI Infrastructure, Models, and Agentic Systems for Science Archive Listing

Back to Full Workshop Archive Listing