Generalization and the Rise of System-level Creativity in Science

Innovation ecosystems require careful policy stewardship to drive sustained advance in human health, welfare, security and prosperity. We develop new measures that reliably decompose the influence of innovations in terms of the degree to which each represents a field-level foundation, an extension of foundational work, or a generalization that synthesizes and modularizes contributions from distant fields to catalyze combinatorial innovation. Using 23 million scientific works from OpenAlex and 19 million works from Web of Science, we demonstrate that while foundational and extensional work within fields has declined in recent years-a trend garnering much recent attention-generalizations across fields have increased and accelerated with the rise of the web, social media, and artificial intelligence, shifting the locus of innovation from within fields to across the system as a whole. We explore implications for science policy.

💡 Research Summary

The paper introduces a novel analytical framework that decomposes scientific innovation into three distinct categories: field‑level foundational work, field‑level extensional work, and system‑level generalizations. Foundational papers introduce new paradigms or methodologies that reshape the knowledge base of a specific discipline. Extensional papers build on these foundations, delivering incremental or applied advances within the same field. Generalizations, by contrast, synthesize concepts, data, or methods from distant fields, modularizing them into new combinatorial forms that can spark further interdisciplinary breakthroughs.

To operationalize this taxonomy, the authors harvested a massive corpus of scholarly output—approximately 23 million records from OpenAlex and 19 million from the Web of Science, totaling over 40 million works spanning 1990–2022. They combined topic modeling (LDA) with citation network analysis to generate a continuous “innovation score” for each paper, ranging from 0 (purely foundational) to 1 (purely generalizing). Papers with intermediate scores are classified as extensional.

Temporal analysis reveals a striking shift in the composition of scientific innovation. Up to the early 2000s, foundational and extensional research together accounted for roughly 70 % of all high‑scoring innovations, a pattern that aligns with long‑standing narratives about the dominance of intra‑disciplinary competition. Starting around 2010, however, the proportion of generalizations began to rise, accelerating dramatically after 2015. By 2020, generalizations comprised about 38 % of all identified innovative works—more than three times their share a decade earlier. The authors attribute this surge to the proliferation of digital infrastructure: open‑access repositories, pre‑print servers, social‑media platforms for scholarly exchange, and AI‑driven text and data mining tools that lower the barriers to cross‑disciplinary discovery.

Citation dynamics further differentiate the three categories. Generalization papers receive citations from a broader set of fields (average multi‑field citation rate of 65 %) and maintain higher long‑term impact, with a median of 25 citations persisting beyond ten years. In contrast, foundational and extensional papers exhibit sharper early citation peaks but decay more quickly, averaging fewer than 15 citations after the same period. This pattern suggests that generalizations act as “combinatorial hubs,” reshaping the structure of the scientific knowledge network and sustaining influence over longer horizons.

Policy implications are drawn explicitly. First, the authors argue for a re‑orientation of funding mechanisms away from strictly discipline‑centric allocations toward “connectivity grants” that explicitly reward cross‑field synthesis. Second, they call for sustained investment in open data, code, and AI‑enabled meta‑analysis platforms to further lower the friction of interdisciplinary work. Third, they recommend expanding cross‑disciplinary training programs and post‑doctoral tracks designed to cultivate scholars capable of navigating and integrating multiple knowledge domains. Implementing these measures, the authors contend, will amplify system‑level creativity, thereby accelerating progress in health, security, welfare, and economic prosperity.