AI for Science Strategic Compass: Aligning Discovery Tensions with Core AI Functions

Ran Liu, Zhibin Lin, Xiaowei Huang

Additional Submission Instructions: For the camera-ready version, please include the author names and affiliations, funding disclosures, and acknowledgements.

Track: Track 1: Original Research/Position/Education/Attention Track

Keywords: AI for science, scientific-discovery tensions, function-tension alignment, function-based AI taxonomy, strategy matrix, atomic triad, AI integration strategy, decision support

TL;DR: A 6×4 function-based strategy matrix mapping discovery tensions to six AI functions; each cell specifies three mitigation mechanisms and method-family exemplars, anchored to MECE atomic categories.

Abstract: AI is transforming scientific discovery, yet researchers face a fragmented, fast-moving field of AI that lacks stable, strategy-level guidance for method selection and integration. In this study, we introduce the AI for Science Strategic Compass (AFSC), a compact decision framework that aligns four cross-domain scientific-discovery tensions (Complexity, Constraint, Scarcity, Explosion) with six core AI functions (Represent; Reason & Infer; Optimize & Control; Simulate & Emulate; Generate & Create; Autonomize & Orchestrate) via a 6×4 Strategy Matrix. We adopt a function-based typology that is domain-agnostic and comparatively stable under ongoing methodological change, enabling direct alignment with these tensions and yielding decision-relevant guidance. Each cell is labeled with a keyword that captures the shared mitigation logic and lists three strategic pathways linked to representative method families. Pathways are anchored to a function-internal atomic triad, stabilizing the vocabulary as techniques change. Automated corpus audits validate the framework’s scope: the four tensions collectively cover all sampled abstracts across six natural science domains, and the six functions account for 98.9% of capabilities reported in recent AI papers. AFSC shifts selection from tool-driven browsing to strategy-first planning, lowering cognitive load and remaining portable across domains. We illustrate its use with an exoplanet spectral retrieval case study that demonstrates systematic integration of complementary AI approaches across functions to address multiple research tensions.

Submission Number: 131