Forecasting Scientific Progress with Artificial Intelligence

Sean Wu, Pan Lu, Yupeng Chen, Jonathan Bragg, Yutaro Yamada, Peter Clark, David Clifton, Philip Torr, James Zou, Junchi Yu

arXiv:2605.22681 · 2026-05-20 공개 · arXiv · PDF

uncertainty-estimation model-overconfidence temporal-prediction cusp-benchmark knowledge-constraints domain-dependence pre-cutoff-knowledge scientific-progress

Abstract

Artificial intelligence (AI) is increasingly embedded in scientific discovery, yet whether it can anticipate scientific progress remains unclear. To study this question, we introduce a temporally grounded evaluation framework for forecasting scientific progress under controlled knowledge constraints. We present CUSP (Cutoff-conditioned Unseen Scientific Progress), a multi-disciplinary and event-level benchmark that evaluates scientific forecasting in AI systems through feasibility assessment, mechanistic reasoning, generative solution design, and temporal prediction. Across 4,760 scientific events, we observe systematic and domain-dependent limitations in current frontier models. While models can identify plausible research directions from competing candidates, they fail to reliably predict whether scientific advances will be realized and systematically misestimate when they will occur. Performance is highly heterogeneous across domains, with the timing of AI progress more predictable than advances in biology, chemistry, and physics. Performance is largely insensitive to whether events occur before or after the training cutoff, suggesting these limitations cannot be explained solely by knowledge exposure in training data. Under controlled information access, additional pre-cutoff knowledge improves performance but does not close the gap to full-information settings, which becomes more pronounced for high-citation advances. Models also exhibit systematic overconfidence and strong response biases, indicating unreliable uncertainty estimation. Taken together, current AI systems fall short as predictive tools for scientific progress. Access to prior knowledge does not translate into reliable forecasting, and performance benefits more from post-event information than from forward-looking prediction.

한국어 요약

한 줄 요약

**[Forecasting Scientific Progress / Benchmark]** CUSP 벤치마크가 4,760 과학 이벤트로 LLM의 과학 진보 예측 평가 -- frontier 모델이 실현 여부·시기 예측에 systematic 실패, 영역간 이질·overconfidence·약한 uncertainty 추정.

핵심 기여도

핵심 아이디어

LLM의 과학 진보 예측 능력은 단순 지식 노출이 아닌 forward-looking inference 자체에 결정적 결함이 있으며, AI 발전의 timing이 자연과학보다 예측 쉽고 high-citation 진보일수록 격차가 커지는 도메인 의존 패턴이 존재한다.

기술적 접근법

주요 결과

의의 및 한계

**의의**: 과학 진보 예측 능력을 controlled로 측정하는 첫 multi-disciplinary 벤치마크, "현재 AI가 과학 forecasting 도구로서 부적합"이라는 명확한 결론, training data 노출과 진정한 forward-looking 추론의 분리. **한계**: 4,760 이벤트 큐레이션의 도메인 균형, event-level 평가의 fine-grained outcome 비교 한계, model family 일반화 추가 검증.

실용적 활용