Proposal Competitions
We invite proposals for two competitions that explore how AI systems can improve scientific discovery and the processes that support it. Submissions to each competition are limited to two pages of main text.
(A) Iterative Improvement of AI Scientists on Specific Tasks
Design a framework that enables an AI scientist to iteratively improve at a meaningful scientific task using existing scientific data and knowledge, without requiring new experiments to provide the feedback signal.
We are particularly interested in tasks where there is no obvious train/test split or ground truth, scientific evidence is heterogeneous or incomplete, and expert scientists nevertheless become substantially better through experience.
Proposals should:
- Define the scientific task and the scientific ability the system should develop.
- Construct a source of experience and feedback from existing information.
- Describe the iterative learning mechanism. This may include changing the model’s weights.
- Propose an evaluation that demonstrates generalizable scientific ability rather than memorized answers.
Implementation is not required for this competition; submissions may describe a proposed framework.
(B) AI Credit Assignment in Science
Credit assignment is a central part of science and is already being affected by AI. We invite proposals that use AI systems to improve credit assignment, including:
- Ensuring that AI scientists and mathematicians comprehensively cite appropriate studies.
- Preventing AI models from ingesting and inadvertently sharing unpublished work.
- Determining the novelty and impact of scientific papers.
- Improving the objectivity, accuracy, and throughput of initial and final manuscript or grant review.
- Developing guardrails that ensure these processes are performed with high fidelity.
Proposals addressing other problems in AI credit assignment are also welcome. Submissions will be evaluated on technical feasibility, evaluation methodology, rigor, and potential to transform AI credit assignment.
Submission and Presentation
- Length: At most two pages of main text for each competition proposal.
- Submission platform: OpenReview. The workshop submission link and final formatting instructions will be announced.
- Presentations: The two highest-ranked proposals in each competition will receive 10-minute podium presentations. Accepted proposals will be presented in interactive poster sessions.
- Awards: Sponsors and cash award amounts are TBA.
Important Dates (Anywhere on Earth)
- Submission deadline (papers and competition proposals): February 1, 2027 AoE
- Acceptance notifications: February 26, 2027 AoE
- Workshop date: To be announced