Authors: Rafael Ferreira da Silva (Oak Ridge National Laboratory (ORNL)), Michela Taufer (University of Tennessee, Knoxville), Tom Gibbs (NVIDIA Corporation), Ravinder Kapoor (U.S. Department of Energy), Arvind Ramanathan (Argonne National Laboratory)
Abstract: Autonomous experiment operation is demonstrating the ability to dramatically improve the pace of scientific discovery. This BoF convenes a community to continue building on the early two-year roadmap, discussing how different teams are using agentic workflows to improve the potential for new discovery and to address the related need to verify the new results.
Long Description: Autonomous science is reshaping discovery across materials, energy, climate, and the life sciences by closing the loop between hypothesis, experiment, and analysis. A year ago, the organizers of this BoF argued that autonomous laboratories operated as isolated islands and proposed the Autonomous Interconnected Science Lab Ecosystem (AISLE), a grassroots network organized around five dimensions, with a set of milestones to guide community efforts.
This BoF builds on our SC25 session, which drew more than 80 participants, and on the Advancing Autonomous Scientific Discovery (A2SD) workshops at ISC High Performance. The interactive, roadmap-driven format remains the same; the content advances to the trust-and-verification agenda below.
The intervening year moved faster than that roadmap anticipated. Multi-agent systems produced experimentally validated hypotheses, from SARS-CoV-2 nanobody design to drug-repurposing candidates. Self-driving laboratories entered a second generation that is more interoperable and orchestrated. The interface layer consolidated around the Model Context Protocol and the Agent2Agent protocol, now joined by an emerging skills layer. Reasoning and scientific foundation models matured into a new substrate. And the Genesis Mission placed robotic laboratories and autonomous experimentation at the center of U.S. federal science strategy, while well-capitalized industry entrants became primary actors.
Progress met a sobering countercurrent. A correction to a flagship autonomous-discovery result walked back its novelty claims and removed a training-data leak. Benchmarks show that agents rivaling experts on closed-ended questions still complete only a small fraction of open-ended research, and fabricated citations have surfaced at leading venues. No system has yet to make and experimentally self-validate a genuinely novel discovery end-to-end. We read this as the defining tension of the field: producing a candidate discovery is no longer the hard part; verifying it is.
This BoF convenes stakeholders from science, computing, industry, and policy to confront that asymmetry. We will present a two-year community roadmap that revisits the five original dimensions, scores each milestone as achieved, partially achieved, reframed, or open, and elevates two concerns to first-class dimensions: trust, verification, and reproducibility; and safety, security, integrity, and governance. We will use two shared lenses: an autonomy ladder distinguishing tool, analyst, and scientist, and the capability-reliability gap that separates what agents can propose from what can be trusted.
Breakout discussion will surface infrastructure gaps and research priorities across the seven dimensions, including vendor-neutral instrument access with inference at the edge, provenance enforced at the point of capture, verification and validation across the experiment lifecycle, zero-trust cross-institution agent coordination, screening at the digital-to-physical interface, and workforce judgment that resists the homogenizing pull of automation.
We will position the grassroots network within the emerging national and global landscape, spanning the U.S. DOE Genesis Mission, the NSF programmable cloud laboratory testbed, the European strategy for AI in science, and the Acceleration Consortium. We argue that top-down mobilization and bottom-up coordination are complementary, and that an interconnected network is the interoperability fabric letting these programs connect rather than re-silo.
Website: https://autonomousscience.org/bofs/sc26/