analiza aiqbench
Current AI agents struggle to reliably synthesize scientific conclusions in high-stakes domains, achieving a factual F1 of only 0.337 under rigorous clean-room evaluation conditions. The study reveals that performance metrics are often inflated by data leakage, while consumer-facing tools frequently produce incomplete or contradictory outputs despite available ground truth. This underscores the critical need for standardized, leakage-free benchmarks before deploying autonomous agents for consequential decision-making.
Can AI Agents Synthesize Scientific Conclusions?
arXiv:2606.11337v1 Announce Type: cross Abstract: Scientific AI agents increasingly retrieve evidence, reason across sources, and synthesize conclusions used in consequential decisions. Yet, their ability to do so in high-stakes domains such as health remains unclear. We introduce SciConBench, a large-scale live benchmark of 9.11K questions and expert-written conclusions from systematic reviews to evaluate open-domain scientific conclusion synthesis. The benchmark draws on an expert-validated automated evaluation pipeline that decomposes conclusions into atomic facts and measures correctness and comprehensiveness via factual precision and recall. To mitigate data leakage, we further introduce SciConHarness, a clean-room evaluation harness that equips agents with controlled web interaction to ensure valid measurement. Evaluating 8 frontier models and deep research agents, we find that factual quality remains low: under clean-room settings, the best agent achieves only a factual F1 of 0.337. Our clean-room setting consistently reduces performance relative to unconstrained evaluation, suggesting that leakage inflates estimates of models' true synthesis capabilities. Finally, we audit consumer-facing agents (e.g., Google AI Overview, OpenEvidence) and find they frequently generate incomplete and sometimes contradictory conclusions, even when the ground-truth answer is available. Overall, our results show that reliable synthesis of scientific conclusions remains an open challenge, and that clean-room evaluation is essential for assessing open-domain AI agents.
Title:Can AI Agents Synthesize Scientific Conclusions?
Abstract:Scientific AI agents increasingly retrieve evidence, reason across sources, and synthesize conclusions used in consequential decisions. Yet, their ability to do so in high-stakes domains such as health remains unclear. We introduce SciConBench, a large-scale live benchmark of 9.11K questions and expert-written conclusions from systematic reviews to evaluate open-domain scientific conclusion synthesis. The benchmark draws on an expert-validated automated evaluation pipeline that decomposes conclusions into atomic facts and measures correctness and comprehensiveness via factual precision and recall. To mitigate data leakage, we further introduce SciConHarness, a clean-room evaluation harness that equips agents with controlled web interaction to ensure valid measurement. Evaluating 8 frontier models and deep research agents, we find that factual quality remains low: under clean-room settings, the best agent achieves only a factual F1 of 0.337. Our clean-room setting consistently reduces performance relative to unconstrained evaluation, suggesting that leakage inflates estimates of models' true synthesis capabilities. Finally, we audit consumer-facing agents (e.g., Google AI Overview, OpenEvidence) and find they frequently generate incomplete and sometimes contradictory conclusions, even when the ground-truth answer is available. Overall, our results show that reliable synthesis of scientific conclusions remains an open challenge, and that clean-room evaluation is essential for assessing open-domain AI agents.
79 pages, 34 figures, 17 tables. Under Submission
Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computers and Society (cs.CY)
arXiv:2606.11337 [cs.AI]
(or arXiv:2606.11337v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2606.11337 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 316 cuvinte
Subiecte conexe
Large Knowledge Model: From Papers to a Scientific Reasoning Landscape
arXiv · 24 sept. 2026
Generating Biomedical Fact-Checking Reports with RL-Enhanced Agentic Search
arXiv · 27 aug. 2026
Tracing the Cascade: A Topology-Aware Evaluation Framework for Scientific Agent Hallucinations
arXiv · 5 aug. 2026
Learning to Evaluate Before Improving: Automatic Rubric Induction for Automatic Research Agents
arXiv · 1 sept. 2026