Blind Variation, Selective Retention: Improving Creative Idea Generation in AI Scientists
AuthorsGuowei Xu, Jiaru Zou, Mete Erdogan, Haotian Ye, Pan Lu, James Zou
AffiliationsStanford University*Co-first Authors
The paper finds that AI scientists may generate more creative ideas by first knowing less about a task, then using evidence to choose the strongest ones.
Key results
Creativity score with limited task information; full-information BV scored 0.462.
Selective Retention discard rate in the candidate-screening result.
Share of top-5% creativity proposals kept at the 50% discard rate.
Share of top-5% utility proposals kept at the 50% discard rate.
BVSR result on gene-expression analysis, compared with 0.906 for SimpleTES.
What the paper found
This paper proposes Blind Variation and Selective Retention, or BVSR, to make AI scientists explore more creatively without losing the benefit of experience. In Blind Variation, idea generation receives deliberately limited task information, reducing anchoring on familiar approaches; in Selective Retention, a Qwen3.5-9B selector trained with a pairwise logistic loss learns from previously tested proposals and scores to screen new candidates. Across 12 scientific tasks and five proposer models, including DeepSeek-V4 and Kimi-K3, limited-information generation achieved a creativity score of 0.530, compared with 0.462 when the same approach received the full task description. At a 50% discard rate, the selector retained 79% of top-5% creativity candidates and 82% of top-5% utility candidates. Integrated into SimpleTES, BVSR improved results across five discovery tasks; for gene-expression analysis, it reached 0.918 versus 0.906 for the original workflow. Claude Opus 5 generated task contexts, while OpenAI models helped judge proposal originality and non-obviousness. The central finding is that information should play different roles at different stages: constrain it during exploration, then use empirical outcomes to guide selection.
Original abstract
Scientific discovery requires ideas that are original, useful, and non-obvious. Yet AI scientists are typically given as much task information as possible during idea generation, which can anchor generation toward familiar approaches. Inspired by the cognitive framework of Blind Variation and Selective Retention (BVSR), we propose a discovery framework that separates these two roles of information: it generates proposals under deliberately limited task information to broaden exploration, then learns from proposals previously verified on the same task and their outcomes to retain promising candidates for costly empirical evaluation. We evaluate BVSR across five proposer models and 12 scientific tasks. Limiting task information substantially increases proposal originality and non-obviousness, and produces a larger share of highly creative and high-utility proposals than existing state-of-the-art prompting-based baselines. Learning from prior proposal outcomes on the task further improves candidate selection, discarding half of the candidates while retaining most top-performing proposals. When incorporated into existing AI-scientist workflows, BVSR achieves stronger discoveries across gene expression analysis, drug response prediction, anomaly detection, scaling law discovery, and brain activity prediction. These results suggest that effective AI scientists should use information differently across stages of discovery: constrain task-specific information during exploration while exploiting accumulated empirical experience on the task.
Read the original paperMore in AI for Science
Browse all 58 papers →DrugCoding: A Scientific Programming Agent for Drug Discovery
Junwei Hu, Zhuohao Lin, Kun Li, Hengnan Zhang
DrugCoding is an AI agent that plans and executes multi-step computational tasks to help automate drug discovery.
Reliability by Architecture: Agentic Workflows for Validated Multi-Physics Simulation from Natural Language
Søren Taverniers, Eric Darve, Hamdi A. Tchelepi
An LLM-assisted workflow prepares and repairs multi-physics simulations, while deterministic checks help ensure the results match both the physics and the user’s request.
Sample-Efficient Control of Magnetoconvective Flows via Differentiable Simulation
Jannis Becktepe, Dmitry Krasnov, Nils Thuerey, Jörg Schumacher, Sebastian Peitz
A differentiable simulator helps researchers control magnetoconvective flows more efficiently and design better heat-transfer outcomes.