University–industry AI patents in Korea are not inherently better — only firms with strong R&D and deeper, experienced collaborations convert academic ties into higher-quality patents, with absorptive capacity reducing the harms of technological distance.

Main Finding

University–industry (UI) collaboration does not automatically raise AI patent quality on average. Instead, higher-quality patent outcomes from collaboration depend on the partner firm’s internal R&D capability and on qualitative features of the collaboration: deeper, experienced collaborations and higher absorptive capacity increase patent quality, and absorptive capacity mitigates the harms of large technological cognitive distance.

Key Points

• Dataset: 90,782 AI patents filed at the Korean Intellectual Property Office (KIPO), 2013–2023.
• Patent Quality Index (PQI): constructed by combining forward citations, patent-family size, and number of claims using min–max normalization.
• Average effect: UI collaboration alone shows no significant average improvement in PQI.
• Moderation: Firms with stronger R&D capabilities obtain higher-quality patents from UI collaboration (positive interaction effect).
• Collaboration structure: Greater collaboration depth and accumulated prior UI experience significantly raise PQI.
• Cognitive distance: Large technological/cognitive distance between partners can harm patent quality, but that negative effect is mitigated when the firm (or partners) have greater absorptive capacity.
• Theoretical framing: Links open innovation (value of knowledge inflows) with absorptive capacity (ability to recognize, assimilate, and exploit external knowledge) to explain when UI collaboration produces sustainable, high-quality innovation.

Data & Methods

• Scope: 90,782 AI-related patent applications filed with KIPO over 2013–2023.
• PQI construction: integrated three patent-quality indicators (forward citations, family size, number of claims), normalized via min–max scaling into a composite index.
• Empirical approach: regression analyses testing the effect of UI collaboration on PQI, including interaction/moderation terms for firm R&D capability and absorptive capacity; additional regressors capture collaboration depth and prior collaboration experience. (Paper reports significance patterns described above.)
• Key variables of interest: binary/continuous measures of UI collaboration, firm R&D capability, collaboration depth, accumulated experience, technological cognitive distance, absorptive capacity.
• Identification: comparative/statistical assessment of moderation effects rather than claiming pure causal identification from experimental or instrumental-variable designs (as described in the summary).

Implications for AI Economics

• Quality over quantity: Policymakers and firms should shift emphasis from counting collaborative projects or patent counts to fostering conditions that raise patent quality (e.g., long-term partnerships, deeper engagement).
• R&D capacity matters: Public support and private investment to strengthen firms’ internal R&D and absorptive capacity amplify the gains from UI collaboration and improve returns to collaborative innovation spending.
• Targeted collaboration policy: Encourage sustained, repeated UI partnerships and capacity-building programs (training, joint labs, researcher mobility) rather than short-term or superficial linkages.
• Managing technological distance: Collaborations across distant technological domains can be productive if partners (especially firms) have the absorptive capacity to integrate unfamiliar knowledge; policy should combine matchmaking with capacity-enhancement.
• Measurement and evaluation: Use composite quality metrics (like PQI) rather than raw patent counts when assessing national competitiveness in AI and the effectiveness of collaboration policies.
• Ecosystem design: For a sustainable AI innovation ecosystem, coordinate university incentives, firm R&D investment, and intermediary support (technology transfer offices, industry consortia) to foster deep, experience-rich collaborations that translate into high-value IP.
• Research agenda: Future work could explore causal mechanisms (e.g., panel methods, instruments), heterogeneity across AI subfields, and international generalizability beyond Korea.