The Costs and Outcomes of Organizing Open-source AI Innovation: Survey Evidence from China

The Costs and Outcomes of Organizing Open-source AI Innovation: Survey Evidence from China | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article The Costs and Outcomes of Organizing Open-source AI Innovation: Survey Evidence from China Zhe Wang This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8409070/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Open-source AI innovation has become integral to global AI development, particularly in China, where it enables distributed knowledge creation, collaborative problem-solving, and rapid advancement in technologies like machine learning models, large language models, and ethical AI frameworks. Yet success in open-source AI systems does not arise from openness alone; it rests heavily on governance processes that align heterogeneous actors, address ethical concerns such as data privacy and algorithmic bias, and sustain orderly collaboration amid rapid technological evolution. Although previous research has explored community-based governance in open-source ecosystems, relatively little attention has been given to how formal open-source organizations manage the governance costs that arise as AI projects scale and institutionalize, especially in Asian contexts where cultural values like collectivism and harmony influence collaborative dynamics. This study proposes a governance-cost framework that captures three categories of governance costs - incentive coordination, rule enforcement and value transformation - and examines how these costs mediate the relationship between governance arrangements and AI innovation outcomes. Drawing on a survey of more than 600 key informants from Chinese open-source AI organizations, the study analyzes how organizational structures influence knowledge convergence in AI algorithms, market diffusion of AI applications, and industrial collaboration for ethical AI deployment through the mechanism of governance costs. The findings show that decentralized organizations enhance knowledge convergence by reducing coordination frictions in AI model development, while firm-led and public institution-based organizations demonstrate stronger market diffusion and collaboration performance due to more stable enforcement systems and clearer value-distribution mechanisms that align with Chinese values of collective benefit and societal harmony. Governance costs significantly mediate these relationships, indicating their central role in shaping AI innovation outcomes. By identifying governance costs as foundational mechanisms, this study advances theoretical understanding of open-source AI governance and provides actionable policy guidance for designing cost-efficient, ethically grounded governance in rapidly evolving Asian AI ecosystems, emphasizing the need for policies that integrate Asian policy tendency and cultural values to counterbalance Western-dominated AI frameworks. Management Software Engineering Open-source AI Organizations Governance cost Innovation outcome Quantitative Asian values 1. Introduction Open-source AI innovation has emerged as a powerful engine of technological and industrial advancement in artificial intelligence, particularly in China, where it facilitates distributed creation, open access to AI models and datasets, modular development of neural networks, and collaboration across organizational and institutional boundaries to address societal challenges like healthcare diagnostics and smart city infrastructure (Wang et al, 2025 ). However, the openness that makes such AI innovation systems productive also generates complexity, especially in handling ethical issues such as bias in AI training data and privacy in large-scale deployments (Keles, 2025 ). As contributors vary in incentives, backgrounds, and institutional affiliations - often influenced by Chinese cultural emphases on collectivism and long-term societal harmony- organizations must establish mechanisms to coordinate AI development activities, mediate conflicts over ethical standards, and convert collective contributions into usable and ethically sound AI outcomes. These mechanisms incur governance costs that fundamentally shape the performance of open-source AI organizations ( OSAIOs ) in China. Despite substantial interest in open-source communities, existing research has not fully explained how governance costs arise within formal organizational settings for AI (Shah, 2006 ; Stewart & Gosain, 2006 ), nor how these costs mediate the impact of governance arrangements on AI-specific innovation outcomes, such as the convergence of diverse AI algorithms or the ethical diffusion of AI technologies. This limitation is particularly salient in institutional contexts characterized by rapid technological and organizational growth, such as China’s expanding AI ecosystem, which is influenced by state policies promoting “AI for good” and cultural values prioritizing collective welfare over individual gain (Taddeo & Floridi, 2018 ). China provides a valuable setting for examining how governance structures function within formal OSAIOs, as its open-source AI ecosystem includes autonomous communities developing grassroots AI tools, consortium-based bazaars for collaborative AI dataset curation, firm-led platforms like those from Tencent or Alibaba integrating proprietary and open AI models, and public institution-driven organizations advancing national AI priorities through universities and government labs. These governance arrangements produce different cost profiles that are likely to influence how effectively AI knowledge is integrated, how AI innovations diffuse into markets with ethical safeguards, and how cross-sector collaboration is sustained to align with Asian values of harmony and inclusivity (Yin & Jamali, 2021 ). Yet empirical evidence on these dynamics remains scarce, especially in relation to policy needs for AI open-source that incorporate non-Western perspectives. To address this gap, this study develops a cost-outcome framework that conceptualizes governance costs as the principal mechanism linking organizational type and AI innovation outcomes. We examine three forms of AI innovation outcomes - knowledge convergence in AI model refinement, market diffusion of AI applications, and industrial collaboration for ethical AI ecosystems (Secundo et al, 2025 ) - and analyze how governance costs related to incentive coordination, rule enforcement, and value transformation mediate these outcomes. Using data collected from 601 experienced participants in Chinese OSAIOs, we analyze how governance arrangements create different cost structures and how these structures explain divergent AI innovation trajectories. The findings show that decentralized organizations achieve strong knowledge convergence through low coordination barriers in AI development, while formalized governance models, particularly firm-led and public institution-based structures, promote market penetration and collaboration through more predictable enforcement and value transformation mechanisms that resonate with Chinese emphases on societal stability and collective progress. Governance costs represent the decisive mechanism through which governance shapes AI innovation outcomes. This study therefore contributes by identifying governance costs as a theoretical foundation for understanding open-source AI governance and by documenting this mechanism empirically in an important Asian context, offering policy suggestions for other Asian societies to develop AI frameworks rooted in their cultural values, such as India’s focus on frugal innovation or Japan’s emphasis on human-AI harmony. 2. Literature review Effective governance is central to sustaining innovation in open-source AI ecosystems, where collaboration spans organizational and institutional boundaries to develop ethical, transparent AI systems. Yet, despite extensive attention to community governance and institutional design, limited research has examined how different OSAIO types shape AI innovation outcomes through underlying cost mechanisms, particularly in Asian contexts where cultural values influence ethical considerations. This section develops the theoretical foundation for analyzing open-source AI organizations (OSAIOs) by reviewing key OSAIO types and conceptualizing governance costs as critical mediating mechanisms that connect organizational structures with AI innovation performance. 2.1. Costs and outcomes in OSAIOs The concept of governance cost originates from transaction cost economics (TCE), which posits that governance structures evolve to mitigate inefficiencies caused by bounded rationality, opportunism, and asset specificity (Coase, 1937 ; Williamson, 1996 ). In the context of AI innovation governance, these costs refer to the total resource expenditures required to establish, coordinate, monitor, and transform collective AI activities, including ethical audits of algorithms (Boumgarden et al., 2012 ; Benkler, 2013 ). Unlike traditional firms, OSAIOs operate at the intersection of community norms, market incentives, and institutional frameworks, facing hybrid governance challenges amplified by AI's unique risks like bias and opacity (Powell, 1996; O’Mahony & Ferraro, 2007 ). Building on this theoretical foundation, we define governance cost in OSAIOs as: “the aggregate economic, cognitive, and institutional efforts required to align diverse participants, enforce shared rules including ethical guidelines, and convert collective outputs into value under conditions of openness and uncertainty in AI development ”. (Stead, 1976 ; Cavallo et al., 2022 ; Secundo et al, 2025 ). Table 1 summarizes the defining characteristics of governance costs in open-source AI organizations. Table 1 Characteristics of governance costs in open-source AI organizations Category Description AI-Specific Examples Incentive Coordination Resources to align motivations and mitigate free-riding Aligning contributors' incentives for ethical AI data labeling Rule Enforcement Expenditures for monitoring and adapting rules Enforcing bias mitigation protocols in AI models Value Transformation Investments to translate knowledge into usable innovations Converting open AI datasets into ethically deployable applications Therefore, governance costs in OSAIOs can be conceptualized along three interrelated channels, each addressing a fundamental coordination problem in open AI ecosystems: The first governance cost is incentive coordination cost, which is the resources devoted to aligning diverse motivations, attracting and retaining contributors, and mitigating free-riding or attrition in AI projects (Simon, 1978 ; Lerner & Tirole, 2002 , 2005 ; von Krogh et al., 2012 ; Ke & Zhang, 2010 ); The second governance cost is rule enforcement cost, which is the expenditures related to establishing, monitoring, and adapting formal or informal rules in rapidly changing AI technological and policy environments, including ethical standards (March, 2010 ; Weber, 2004 ; Joblin et al., 2023); The last but not least is value transformation cost, which is the investments required to translate collective AI knowledge into usable innovations, manage intellectual property, and ensure equitable benefit distribution aligned with cultural values (Williamson, 1996 ; Chesbrough, 2017 ; Hall et al., 2005 ). These three channels function as mediating mechanisms that link OSAIO types to AI innovation outcomes (Schmidt & Schnitzer, 2002 ). Each governance structure produces a distinctive cost profile, shaping performance across three dimensions (shown by Table 2 ): knowledge convergence in AI, market diffusion of AI, and industrial collaboration for AI. For instance, autonomous communities benefit from low enforcement costs due to shared norms but face high transformation costs because of informal value appropriation in AI ethics (Ostrom, 1990 ; von Krogh et al., 2012 ). In contrast, firm-led OSAIOs reduce transformation costs through centralized IP management yet incur higher incentive coordination costs when community trust declines, especially in ethically sensitive AI areas (West & O’Mahony, 2008). The effectiveness of a governance mode thus depends on its ability to balance these cost trade-offs and foster both efficiency and ethical collaboration in AI (Foss & Klein, 2012 ). In conclusion, governance costs serve as a comprehensive analytical lens for understanding how governance structures shape AI innovation outcomes in open-source ecosystems. By conceptualizing governance costs as dynamic, multidimensional mediators, this study advances theoretical understanding of open-source AI governance and provides actionable insights for optimizing AI innovation governance in emerging economies such as China, where policies must integrate values like collective harmony to address ethical concerns. Table 2 Characteristics of innovation outcomes in open-source AI organizations Outcome Description AI-Specific Metrics Knowledge Convergence Integration of distributed AI expertise Convergence of AI model architectures Market Diffusion Commercial adoption of open-source AI solutions Deployment of open AI in industries Industrial Collaboration Depth of partnerships for AI ecosystems Cross-sector ethical AI initiatives 2.2. OSAIO types and open-source AI innovation Open-source AI innovation emerged from decentralized, self-organizing communities bound by shared technical values rather than formal hierarchies, as seen in early AI projects like TensorFlow (Hippel & Krogh, 2003 ; Gabriel & Goldman, 2002 ). Early collaborations relied on voluntary participation and peer recognition rooted in ethical AI principles (Torvalds, 2021 ). These informal structures effectively fostered creativity and rapid AI knowledge exchange but struggled to maintain long-term stability as AI projects scaled, particularly with ethical challenges like algorithmic fairness. As open-source AI initiatives expanded, they adopted formalized governance mechanisms to provide stability, reduce transaction costs, and align incentives without suppressing openness, especially in China where state policies emphasize AI for societal benefit (Benkler, 2013 ; Bogers et al., 2017 ). Drawing on governance and organizational theory (Williamson, 1996 ; Gulati et al., 2012 ), as shown by Table 3 , four ideal OSAIO types can be identified, each balancing openness and control through distinct institutional logics influenced by Asian values. Each governance mode reflects a distinctive configuration of authority, incentives, and coordination for AI. Importantly, the institutionalization of governance introduces governance costs, which are the economic, cognitive, and organizational resources required to design, operate, and adapt governance structures that sustain AI innovation while addressing ethical concerns (Hauge, Ayala & Conradi, 2010 ). These costs can be viewed as both transactional relating to AI coordination and control and transformational relating to AI value creation and appropriation. While they impose operational burdens, they also function as strategic investments that enable scalability, stability, and ethical accountability in complex AI systems (McGahan & Silverman, 2006 ; Autio et al., 2013 ). The magnitude and composition of these costs vary with institutional context, organizational form, and temporal dynamics (Aghion & Howitt, 1992). Understanding their mechanisms is therefore essential for explaining how OSAIOs balance the efficiency of markets, the accountability of hierarchies, and the adaptability of networks in AI, particularly in Asia where cultural values shape policy responses (Powell, 1990 ; Benkler, 2016 ). Table 3 Four ideal OSAIO types in China OSAIO Types Description Supporting Literatures Autonomies self-organized, meritocratic communities where technical contribution defines authority, emphasizing collective AI advancement Raymond, 1999 ; von Krogh et al., 2012 Bazaars Foundation- or alliance-based systems that coordinate diverse actors through modular AI design and participatory consensus, exemplified by Chinese AI consortia West & O’Mahony, 2008; West & Gallagher, 2006 Firms Corporate-led entities integrating open and proprietary AI under strategic management, such as Baidu's PaddlePaddle Dahlander & Wallin, 2020 ; Radziwon & Bogers, 2019 Public institutions Government- or university-led organizations advancing public AI missions through hybrid governance, aligning with Chinese values of harmony Mergel, 2016 ; Liu et al., 2023 3. The development of hypotheses Building on the governance cost framework, this section develops testable hypotheses linking OSAIO types to AI innovation outcomes in OSAIOs. It proceeds in two stages. First, we delineate how distinct governance structures shape AI innovation performance through variations in coordination, incentive alignment, and institutional embedding, with a focus on ethical AI policy needs. Second, we theorize three core governance cost mechanisms - incentive coordination, rule enforcement, and value transformation - that mediate these effects, thereby explaining the differential AI innovation trajectories across OSAIOs in China’s evolving open-source AI ecosystem. 3.1. Innovation outcomes in OSAIOs Open-source AI innovation has undergone significant structural evolution, transitioning from informal community-driven initiatives to formally organized entities encompassing non-profit foundations and corporate-dominated structures for AI model development (Raymond, 1999 ; Benkler, 2013 ). This evolution reflects a broader transformation in collaborative AI, where OSAIO types critically shape the efficiency and sustainability of open-source AI ecosystems, especially in addressing ethical concerns like fairness in AI decision-making. In the Chinese context, open-source AI organizations (OSAIOs) exhibit a unique four-tier architecture integrating technology communities, social organizations, market entities, and government systems (Wang et al., 2023 ). These tiers manifest in four distinct OSAIO types: autonomies (self-organized technical communities for AI experimentation), bazaars (social organizations as collaborative platforms for AI ethics discussions), firms (enterprise-driven market entities for commercial AI), and public institutions (administratively governed hierarchical organizations for national AI strategies) (Gambardella & Panico, 2014 ; Wang, 2024 ). This diversity in governance structures underscores the adaptive nature of open-source AI innovation, balancing decentralized collaboration with strategic oversight to meet policy needs for ethical AI. For instance, corporate involvement often introduces resource efficiency and market alignment while incorporating Chinese values of collective progress, whereas community-driven autonomies prioritize knowledge democratization and rapid AI iteration with ethical inclusivity (Chesbrough, 2017 ; Schweik, 2014 ). The performance of these modes determines whether open-source AI initiatives achieve visibility or transformative impact, particularly in digital technology and knowledge-intensive sectors where collaboration scalability and ethical AI diffusion are pivotal. The efficacy of open-source OSAIO types is reflected in three core AI innovation outcomes: knowledge convergence, market diffusion, and industrial collaboration. Knowledge convergence emphasizes collective problem-solving and the generation of novel AI insights through transparent and open participation, including ethical dataset sharing (Schweik, 2012 ). Market diffusion measures the commercialization and ecosystem expansion potential of open-source AI outputs, with safeguards against misuse (Lerner & Tirole, 2005 ; Bessen, 2006 ). Industrial collaboration reflects the capacity to integrate diverse stakeholders into a sustainable AI network that aligns with Asian ethical values (Li et al., 2017 ; Chen et al., 2020 ). These outcomes are interdependent: robust knowledge convergence underpins market diffusion, while industrial collaboration amplifies both, especially in policy-driven AI ecosystems (Gambardella & Panico, 2014 ). The interplay between OSAIO types and AI innovation outcomes reveals inherent trade-offs. Autonomies excel in knowledge convergence due to their open, meritocratic culture but struggle with resource scalability for ethical AI scaling (Wang, Xue & Zhao, 2024 ). Bazaars foster industrial collaboration through neutral platforms but face coordination challenges in AI ethics. Firm-driven governance accelerates market diffusion yet risks undermining openness in AI transparency. Public institutions provide stability and funding but lack agility in responding to ethical AI dilemmas. Hence, no single governance mode universally optimizes all AI innovation outcomes; rather, each exhibits comparative advantages across different dimensions, informing policies that integrate Chinese values like harmony. Hypothesis a (H1a) : Decentralized OSAIO types are more effective than formalized governance models in knowledge convergence for AI innovation. Hypothesis b (H1b) : Formalized OSAIO types are more effective than decentralized models in market diffusion and industrial collaboration for AI innovation. 3.2. Governance costs in OSAIOs Governance costs in OSAIOs arise from tensions between collective AI goals and individual motivations under bounded rationality, exacerbated by ethical uncertainties in AI (Williamson, 1993 ; March, 2010 ). In early stages, ambiguous rules and weak institutions heighten these costs, manifesting as incentive coordination costs, which are resources expended to reconcile diverse motivations and sustain participation in AI projects (Lerner & Tirole, 2002 ; Ostrom, 1990 ). Drawing on institutional and behavioral economics, we identify three motivational mechanisms - intrinsic, extrinsic, and outcome-based - that impose distinct alignment costs in AI contexts (Deci & Ryan, 2000 ; von Krogh et al., 2012 ; Wang, Li & Yu, 2024 ). Autonomies incur high costs maintaining intrinsic motivation for ethical AI, bazaars balance stakeholder incentives through legitimacy, firms invest in outcome-based alignment at the expense of AI openness, and public institutions rely on policy-backed but rigid incentives aligned with national AI ethics (Bagozzi & Dholakia, 2006 ). These variations influence AI innovation efficiency and ethical knowledge diffusion. As OSAIOs mature, tensions emerge between structured governance and decentralized AI collaboration, generating rule enforcement costs - expenses for monitoring compliance, resolving conflicts, and maintaining coordination in ethical AI (Williamson, 1996 ; Weber, 2004 ). Such costs arise from hierarchy-innovation misalignment, temporal trade-offs, and stability-flexibility dilemmas in AI policy. Autonomies minimize formal enforcement but suffer scalability limits for AI ethics; bazaars balance coordination via consensus mechanisms; firms reduce short-term costs through control but lose adaptability in ethical AI; public institutions incur high bureaucratic costs yet maintain stability for national AI goals. These patterns shape AI innovation efficiency and collaboration outcomes. Finally, value transformation costs emerge when OSAIOs institutionalize and appropriate AI outputs, reflecting tensions between value creation and capture in ethical AI (Benkler, 2013 ; Chesbrough, 2017 ). These costs stem from power imbalances, path dependencies, and resource constraints that hinder equitable value realization in AI. Autonomies preserve openness but underperform in AI commercialization with ethics; bazaars ensure fair distribution but incur negotiation costs; firms commercialize efficiently but risk eroding community trust in AI transparency; public institutions sustain legitimacy but lack responsiveness to ethical AI needs. Balancing collaboration integrity with commercialization agility minimizes these costs, informing AI policies. Hypothesis 2 (H2) : Incentive coordination and value transformation costs have greater impact on achieving knowledge convergence and industrial collaboration outcomes in ethical AI governance, while rule enforcement cost plays a dominant role in market diffusion of ethical AI governance. 4. Research design To empirically examine how OSAIO types influence AI innovation outcomes through governance costs, this study adopts a rigorously structured survey-based research design. The following section outlines the data collection procedures, measurement development, and analytical strategy, ensuring methodological transparency and empirical robustness in linking theoretical constructs to observable AI innovation dynamics among Chinese open-source AI organizations. 4.1. Methodology and data collection The study employs a structured survey design to investigate the relationship between OSAIO types, governance costs, and AI innovation outcomes in Chinese OSAIOs. A survey approach is particularly suitable given the limited availability of standardized data in China’s emerging open-source AI ecosystem and the need to capture organizational-level governance practices directly, including ethical AI considerations. Institutional collaboration with the China Computer Federation (CCF) and the OpenAtom Foundation ensured survey credibility and broad dissemination among qualified respondents focused on AI. Primary data were collected during the October 2023–2024 China Open-Source Conference, the most comprehensive national forum for OSAIO leaders, maintainers, and community decision-makers in AI. The event provided a concentrated sampling frame of knowledgeable informants, enhancing both response quality and representativeness in AI contexts. Survey development followed a six-month, iterative process involving semi-structured interviews with 36 experts representing all four OSAIO types in AI. Through multiple pretests and pilot studies, the final instrument was refined to maximize content validity and contextual relevance, incorporating questions on ethical AI governance. It captured detailed measures of organizational characteristics, governance mechanisms, AI innovation outputs, and governance cost perceptions. The final digital survey yielded 601 valid responses out of 891 distributed questionnaires (valid response rate: 95.1% among completed attempts). Quality control procedures included completion time thresholds, IP and regional validation, and logical consistency checks. The respondent composition, which are technical contributors (42.3%), maintainers (26.8%), community managers (18.2%), and organizational leaders (12.7%), ensures diverse perspectives on AI governance and innovation processes. To mitigate potential biases, we conducted several diagnostic tests. Non-response bias was assessed by comparing early and late respondents, with no significant differences detected (p > 0.05). Sample representativeness was validated against the 2023 China Open-Source Ecosystem Report, showing strong alignment across governance mode distributions in AI. Role-based filters ensured respondent authority, while ex-post harmonization with publicly available AI project data further enhanced the reliability of self-reported AI innovation indicators. These methodological safeguards strengthen the validity, generalizability, and analytical rigor of the dataset for AI policy insights. 4.2. Measurements and variables The constructs of the survey were operationalized through theoretically grounded, empirically validated multi-item scales in Appendix Figure A1, A2 and Table A1 , A2 . The descriptive statistics and correlation matrix are shown by Table 4 . OSAIO Types. Drawing on open-source AI governance literature (Herzog & Leker, 2010 ; Schweik, 2014 ), OSAIOs are classified into four types: (1) autonomies, (2) bazaars, (3) firms, and (4) public institutions. Each category is represented by a dummy variable, with commercial firms as the reference group. In our sample, decentralized governance dominates: bazaars (59.73%) and autonomies (10.48%) together constitute 70.21%, followed by public institutions (12.65%) and firms (7.14%), reflecting China's AI landscape. Innovation Outcomes. AI innovation performance is conceptualized along three dimensions: (1) knowledge convergence is the integration of distributed AI expertise, including ethical model training; (2) market diffusion is the commercial adoption and spread of open-source AI solutions with ethical safeguards; (3) industrial collaboration is the depth of partnerships with industry stakeholders for ethical AI. Each construct was measured using five-point Likert scales, refined through expert review and pretesting to ensure validity and reliability (Cronbach’s \(\:\alpha\:\) > 0.70; AVE > 0.50). Governance Costs. Governance costs refer to the resources expended to coordinate AI activities, enforce rules including ethical guidelines, and align incentives within OSAIOs. We operationalize three complementary dimensions: (1) incentive coordination costs are disparities between contributor motivations and actual participation in AI ethics; (2) rule enforcement costs are difficulties in implementing and maintaining governance mechanisms amid AI technical and institutional complexity; (3) value transformation costs are inefficiencies arising from information asymmetry and multi-party reward distribution in AI. Each dimension is measured using multi-item Likert scales normalized to a 0–1 range for analytical comparability. Control variables include respondent demographics (gender, age, education, etc.), time devoted to open-source AI work, and provincial fixed effects to account for regional institutional heterogeneity in AI policy. Confirmatory factor analysis confirmed strong convergent and discriminant validity (all factor loadings > 0.70; CR > 0.80). Table 4 Descriptive statistics and correlation matrix Variable Mean SD 1 2 3 4 5 1. Knowledge convergence 4.30 0.87 1.00 2. Market diffusion 4.20 0.88 0.65 1.00 3. Industrial collaboration 4.25 0.88 0.72 .068 1.00 4.Incentive coordination costs 0.17 0.18 -0.31* -0.28** -0.35** 1.00 5. Rule enforcement costs 0.15 0.16 -0.25** -0.32** -0.29** 0.42 1.00 6. Value transformation costs 0.70 0.18 -0.38** -0.35** -0.41** 0.37 0.39 Note: * p < 0.1, ** p < 0.05, *** p < 0.01 Given the ordered categorical nature of AI innovation outcomes, we employed ordered logistic regression (ordinal logit) with maximum likelihood estimation. Brant tests confirmed that the proportional odds assumption was satisfied, validating model specification. Meanwhile, to test the mediating role of governance costs, we implemented bootstrapped indirect effects (5,000 resamples; bias-corrected CIs per Hayes, 2017 ). This approach allows simultaneous estimation of direct and indirect pathways and provides distribution-free, robust inference for AI contexts. Robustness checks include: re-estimating models via OLS using composite AI innovation scores; testing alternative control variable sets; verifying absence of multicollinearity (VIF < 3.5); conducting subgroup analyses by governance mode and likelihood-ratio tests for model comparison. This multi-method validation strategy ensures that results are statistically robust, theoretically coherent, and empirically reliable, which are presented in Appendix Table A4 to A6. 5. Empirical findings This section presents the empirical findings from ordered logistic regressions and structural equation modeling to examine how OSAIO types influence AI innovation outcomes directly and indirectly through governance costs. By testing the two hypotheses, the analysis provides systematic evidence on the distinct effects of governance structures and the mediating role of governance cost mechanisms in shaping open-source AI innovation performance in China. 5.1. Main effects The ordered logistic regression analysis provides robust evidence supporting H1a and H1b, confirming that OSAIO types exert distinct and statistically significant effects on the three dimensions of AI innovation outcomes: knowledge convergence, market diffusion, and industrial collaboration. The models demonstrate satisfactory explanatory power (Pseudo R²=0.032 ~ 0.042), consistent with established benchmarks in organizational and AI innovation studies. For knowledge convergence in AI, both autonomous communities ( \(\:\beta\:\:\) = 1.186, p < 0.05) and bazaar-style governance ( \(\:\beta\:\:\) = 0.854, p < 0.1) exhibit positive and significant coefficients. This suggests that decentralized, meritocratic structures foster AI knowledge sharing and collaborative problem-solving, including ethical algorithm refinement, consistent with prior findings on open coordination in AI (von Krogh et al., 2012 ; O’Mahony & Ferraro, 2007 ). In contrast, public institutions do not show significant influence in this domain, reflecting their weaker alignment with bottom-up AI knowledge generation. For market diffusion of AI, public institutions display the strongest positive effect ( \(\:\beta\:\:\) = 0.987, p < 0.05), underscoring their ability to leverage policy networks, legitimacy, and institutional infrastructure to scale ethical AI diffusion. The firm-led reference category also performs well, while autonomous communities show an insignificant and negative effect, suggesting that decentralized forms may excel at AI ideation but face challenges in commercialization and ethical diffusion due to weak market linkages. For industrial collaboration in AI, public institutions again demonstrate the most substantial effect ( \(\:\beta\:\:\) = 1.245, p < 0.01), followed by bazaars ( \(\:\beta\:\:\) = 0.432, p < 0.1). This pattern highlights the unique role of institutional coordination in mobilizing multi-party collaboration across AI industry boundaries, while semi-formal bazaar structures facilitate inclusive participation and ethical AI exchange. The overall findings confirm that decentralized modes (autonomies and bazaars) are superior for AI knowledge integration, whereas formalized modes (firms and public institutions) are more effective for market and industrial outcomes in ethical AI. Collectively, Table 5 presents the estimated main effects of OSAIO types on three AI innovation outcomes using ordered logistic regression. These results substantiate the theoretical argument that OSAIO types serve as distinct coordination architectures, each optimizing specific AI goals. They provide strong empirical support for H1, while establishing a foundation for subsequent mediation analyses. Table 5 Ordered logistic regression results: main effects of OSAIO types Variable Knowledge convergence Market diffusion Industrial collaboration Autonomies 1.186** (0.215) -0.324 (0.287) 0.215 (0.302) Bazaars 0.854* (0.342) -0.187 (0.401) 0.432* (0.258) Public institutions 0.430 (0.376) 0.987** (0.321) 1.245*** (0.287) Firms (ref) - - - Controls Yes Yes Yes PseudoR² 0.032 0.038 0.042 Observations 601 601 601 Note: Robust standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01 5.2. Mechanisms in governance costs To further examine whether governance costs function as the key mechanisms through which OSAIO types influence AI innovation outcomes (H2), we applied bootstrap estimation with 5,000 replications. The indirect effects and confidence intervals confirm the statistical significance of all three governance cost dimensions - incentive coordination costs, rule enforcement costs, and value transformation costs - as mediators. This evidence highlights that governance costs constitute the micro-level transmission channels linking structural governance arrangements to AI innovation performance, with implications for ethical AI policy. For knowledge convergence outcome (in Table 6 ), the results show that value transformation costs exert the strongest mediating effect (indirect effect = -0.287, 95% CI [-0.382, -0.198]), indicating that barriers in information exchange and unequal value distribution substantially inhibit the integration of dispersed AI knowledge resources across OSAIO types. Rule enforcement costs also demonstrate a meaningful mediation effect (indirect effect = -0.134, 95% CI [-0.217, -0.068]), suggesting that difficulties in coordinating and enforcing rules among multiple AI contributors weaken collective problem-solving and ethical knowledge co-creation. Although comparatively smaller in magnitude, the mediation effect of incentive coordination costs (indirect effect = -0.087, 95% CI [-0.152, -0.031]) remains significant, reflecting that motivational disparities between contributors and organizational objectives still create friction within collaborative AI innovation. Together, these results confirm that effective knowledge convergence in open-source AI organizations depends not only on the structural openness of OSAIO types but also on the efficiency with which specific governance costs are managed, especially those related to value transformation and enforcement consistency in ethical AI. Table 6 Mediation analysis for innovation outcome of OSAIOs: knowledge convergence Variables Indirect effect BootSE BootLLCI BootULCI Incentive coordination costs -0.087 0.031 -0.152 -0.031 Rule enforcement costs -0.134 0.038 -0.217 -0.068 Value transformation costs -0.287 0.047 -0.382 -0.198 Turning to market diffusion outcome (in Table 7 ), rule enforcement costs emerge as the most critical mediating channel (indirect effect = -0.203, 95% CI [-0.301, -0.125]), underscoring that difficulties in implementing governance rules and adapting institutional arrangements under conditions of AI technological and regulatory uncertainty can significantly impede ethical AI diffusion. Value transformation costs also play a considerable role (indirect effect = -0.165, 95% CI [-0.241, -0.097]), indicating that asymmetric information and complex reward allocation mechanisms reduce incentives for broader market adoption of ethical AI. Incentive coordination costs, while statistically significant (indirect effect = -0.076, 90% CI [-0.138, -0.021]), contribute less prominently, suggesting that motivational issues matter less for AI diffusion outcomes than do structural coordination and enforcement challenges. These findings suggest that successful market diffusion in open-source AI ecosystems depends critically on the ability of governance actors to minimize coordination and enforcement costs, translating institutional stability into market legitimacy for ethical AI development. Table 7 Mediation analysis for innovation outcome of OSAIOs: market diffusion Variable Indirect effect BootSE BootLLCI BootULCI Incentive coordination costs -0.076 0.030 -0.138 -0.021 Rule enforcement costs -0.203 0.045 -0.301 -0.125 Value transformation costs -0.165 0.037 -0.241 -0.097 Finally, in industrial collaboration outcome (in Table 8 ), the mediation analysis again identifies value transformation costs as the most substantial channel (indirect effect = -0.312, 95% CI [-0.408, -0.224]), highlighting that information opacity, attribution uncertainty, and inequitable distribution of jointly created value hinder the formation and sustainability of cross-sector partnerships for ethical AI. Rule enforcement costs (indirect effect = -0.156, 95% CI [-0.237, -0.089]) and incentive coordination costs (indirect effect = -0.123, 95% CI [-0.198, -0.062]) also significantly mediate the relationship, reflecting the coordination difficulties that arise in multi-stakeholder AI collaborations where divergent interests and governance expectations coexist. These results demonstrate that effective industrial collaboration in AI is highly sensitive to the management of governance costs, particularly those associated with the fair transformation and distribution of value in ethical contexts. Table 8 Mediation analysis for innovation outcome of OSAIOs: industrial collaboration Variable Indirect effect BootSE BootLLCI BootULCI Incentive coordination costs -0.123 0.035 -0.198 -0.062 Rule enforcement costs -0.156 0.038 -0.237 -0.089 Value transformation costs -0.312 0.047 -0.408 -0.224 5.3. Discussion: balancing coordination and creativity in Asian AI innovation Synthesizing these empirical findings reveals a coherent mechanism through which OSAIO types shape AI innovation outcomes via differentiated governance costs. Distinct governance structures engender unique constellations of transaction, coordination, and motivational costs that mediate their AI innovation performance. Decentralized modes, such as autonomies and bazaars, demonstrate superior knowledge convergence due to their openness, peer recognition, and inclusive contribution norms, echoing findings in studies of community-anchored open-source AI (von Krogh, Spaeth, & Lakhani, 2003 ; Lakhani & von Hippel, 2003 ). However, they also face elevated value transformation and coordination costs stemming from fragmented accountability, information asymmetry, and weaker institutional enforcement- similar to observations in “private-collective” models of AI innovation where public goods logic clashes with excludability pressures in ethical AI (von Hippel & von Krogh, 2003 ; Verdegem, 2024 ). In contrast, more formalized structures- firms and public institutions - excel in market diffusion and industrial collaboration by leveraging concentrated resources, institutional legitimacy, and standardized procedures for ethical AI, as observed in studies of firm-community relationships and institutional sponsorship in open-source AI (Dahlander & Magnusson, 2008 ). Yet, these formal modes encounter higher rule enforcement costs that may constrain adaptability and inclusiveness - parallel to findings in research on governance rigidity and its impact on AI innovation performance in hierarchical or firm-led open-source engagements (Fitzgerald, 2006 ; Crowston & Howison, 2005 ). These patterns suggest that governance costs are not simply inefficiencies to be minimized but structural mechanisms that translate organizational design into AI innovation outcomes, with policy implications for ethical AI in Asian countries. Among the three cost types, value transformation costs emerge as the most pervasive constraint, reflecting the intrinsic tension between openness and equitable value distribution in AI, where cultural values like Chinese harmony can guide fair allocation (Bessen, 2006 ; Baldwin & von Hippel, 2011). Rule enforcement costs prove decisive for market-oriented outcomes, where procedural clarity and compliance are essential for scaling ethical AI across diverse stakeholders - a theme echoed in research on institutional legitimacy and boundary work in OSAIOs. Incentive coordination costs, while weaker in effect, remain a persistent source of friction in sustaining voluntary engagement for ethical AI, consistent with work on motivation shifts in open-source AI contributors (Gerosa et al., 2021 ). Together, these mediating dynamics reveal a fundamental trade-off between openness and control that defines the governance of open-source AI ecosystems: decentralized structures foster creativity but risk inefficiency and value leakage in ethical AI, whereas formalized systems secure coordination and diffusion but may suppress diversity and slow boundary spanning - an insight that aligns with debates in open AI innovation and institutional variety (Chesbrough, 2003 ; Hall & Soskice, 2001 ). Overall, the findings provide compelling evidence that governance costs serve as the underlying mechanisms linking OSAIO types to AI innovation outcomes. Specifically, value transformation costs most strongly mediate knowledge and collaboration processes in AI, whereas rule enforcement costs play a dominant role in market-oriented AI innovation. Incentive coordination costs remain relevant across contexts but exert relatively smaller effects. This pattern reveals a nuanced mechanism: decentralized OSAIO types enhance openness and participation but incur higher coordination and value transformation costs, while more formalized modes achieve efficiency in diffusion and collaboration but are constrained by enforcement rigidity. By delineating these mechanisms, the analysis not only validates the hypotheses but also advances our understanding of how governance costs operationalize the broader relationship between structural design and AI innovation performance in open-source ecosystems, offering a basis for Asian AI policies rooted in cultural values. 6. Contribution and implications This study examines how governance costs shape AI innovation outcomes in open-source AI organizations (OSAIOs) in China. Rather than emphasizing organizational typologies, the analysis identifies governance costs as the key explanatory mechanism linking governance design with AI innovation outcomes, particularly in addressing ethical and social concerns (Chesbrough, 2003 ). Evidence from Chinese OSAIOs shows that the configuration and management of governance costs largely determine differences in knowledge convergence for AI models, market diffusion of ethical AI applications, and industrial collaboration for sustainable AI ecosystems. 6.1. Theoretical contributions The study makes three theoretical contributions to the literature on open-source AI innovation and governance (von Krogh, Spaeth, and Lakhani, 2003 ). First, it develops and tests a governance cost framework that specifies three functional cost categories -incentive coordination, rule enforcement, and value transformation -and demonstrates their distinct effects on AI innovation outcomes. Value transformation costs are positively associated with knowledge convergence and collaboration in ethical AI, whereas rule enforcement costs primarily facilitate market diffusion of transparent AI. These results clarify how the efficiency and allocation of governance costs explain heterogeneity in AI innovation outcomes across organizations, especially in Asian contexts. Second, the results highlight that governance costs vary systematically across organizational arrangements (Schweik & English, 2012 ; Remneland-Wikhamn & Knights, 2012 ; Greiner & Goodhue, 2005 ). Autonomous and bazaar forms tend to lower coordination barriers and enhance AI knowledge integration, while firm-led and public institution-based forms strengthen enforcement and resource mobilization for ethical AI. The comparative evidence suggests that governance effectiveness depends on the alignment between cost structures and AI innovation objectives rather than on any particular governance mode, informing policies that balance openness with cultural values. Third, the study conceptualizes governance costs as purposeful, not merely frictional (Chakraborti et al., 2023; Yin et al., 2022 ). Governance costs represent strategic investments in incentive coordination, rule enforcement, and value transformation under open and uncertain AI conditions, ensuring ethical alignment. This perspective integrates transaction cost economics with AI innovation governance, extending cost-based reasoning beyond firm boundaries to distributed AI ecosystems in Asia. 6.2. Policy implications for Asian AI governance and more Based on the cost-outcome framework developed in this study, which highlights the mediating role of incentive coordination, rule enforcement, and value transformation costs in shaping open-source AI (OSAI) innovation outcomes, we expand the policy suggestions to provide more comprehensive, regionally tailored guidance. These recommendations emphasize reducing governance costs while integrating Asian value systems, such as Confucian harmony (he, “和”), collectivism, human-centeredness (ren or benevolence, “仁”), environmental stewardship rooted in Buddhist and Daoist philosophies, and pragmatic innovation aligned with diverse cultural contexts. By drawing on these values, Asian policymakers can counterbalance Western-dominated AI frameworks (e.g., the EU’s precautionary, rights-based approach or the U.S.’s market-driven model) and foster “sovereign AI” that prioritizes societal well-being, equity, and regional cooperation. This expansion incorporates insights from recent developments, including the ASEAN Guide on AI Governance and Ethics (2024), national strategies across Asia, and ethical concerns like bias mitigation, ecological sustainability, and cultural preservation. For policymakers, fostering an enabling institutional environment that reduces systemic transaction costs in AI is critical. Rather than prescribing a single model (Hertel et al., 2003 ; von Hippel, 2005 ), policies should support diverse organizational forms and shared infrastructure that facilitate ethical collaboration and diffusion across sectors, countering the dominance of EU, US, and even Chinese frameworks. In China, policies could incentivize public institution-led OSAIOs to integrate collectivist values for ethical AI diffusion, while in other Asian countries like South Korea or Singapore, frameworks could emphasize Confucian harmony or pragmatic innovation to develop national AI capacities. This includes subsidies for value transformation in ethical AI, regulatory sandboxes for rule enforcement testing, and incentives for cross-border Asian AI collaborations to promote shared values over Western individualism. Policies should not prescribe a uniform model but support diverse OSAI organizational types (autonomies, bazaars, firms, and public institutions) through flexible, adaptive frameworks. This aligns with the study’s findings that decentralized structures excel in knowledge convergence (e.g., collaborative AI model development) but incur higher coordination and value transformation costs, while formalized ones enhance market diffusion and industrial collaboration but face enforcement rigidities. Recommendations are structured at national, regional, and cross-border levels, with specific examples for key Asian countries and groupings. 6.2.1. National level policy implications At the national level, policies should focus on building AI ecosystems that minimize governance costs by investing in human capital, infrastructure, and ethical tools, while embedding cultural values to ensure AI serves collective harmony and well-being rather than exacerbating inequalities. To reduce incentive coordination costs (e.g., aligning diverse motivations in OSAI communities), governments should subsidize upskilling programs that incorporate Asian ethical traditions. For instance, in China, expand existing initiatives like the National AI Talent Development Plan to include Confucian virtues training, emphasizing xin (trustworthiness, “信”) in AI collaboration to sustain contributor engagement. In South Korea, leverage the country’s high-stakes education system by integrating AI ethics curricula based on Confucian harmony, addressing concerns like AI misuse in competitive exams (e.g., through programs similar to the proposed AI Responsibility Act of 2023). In India, promote Gandhian principles of equitable access by funding frugal AI training for underserved populations, mitigating biases against marginalized groups like Dalits and ensuring diverse datasets reflect India’s multicultural fabric. Subsidies could cover certification programs to build a workforce skilled in ethical AI, reducing attrition in OSAI projects. To supporting innovation ecosystems and reduce value transformation costs, policy-makers shall provide grants and shared infrastructure to lower the costs of translating collective AI outputs into equitable value, aligning with human-centered values. In Japan, build on “Society 5.0” principles - emphasizing human-AI coexistence and Daoist-inspired environmental balance - by funding open-source repositories for sustainable AI (e.g., low-energy models to address ecological destruction from data centers). India’s National Strategy for AI could offer tax incentives for OSAI start-ups using local, bias-free data, promoting “sovereign AI” to counter colonial dependencies on Western tech. In Singapore, expand regulatory sandboxes to test value transformation mechanisms, such as attribution systems for fair benefit distribution in AI, fostering pragmatic innovation while ensuring transparency. To invest in ethical R&D and rule enforcement and minimize rule enforcement costs, policy-makers should amid rapid AI evolution, allocate R&D funding for tools that embed cultural safeguards. In Thailand and Vietnam, under ASEAN frameworks, investments in cybersecurity-aligned AI research, drawing on Buddhist teleology for nature-respecting systems shall be encouraged. South Korea could enact hard regulations like the pending AI laws, incorporating Confucian relational ethics to enforce accountability in elder care AI, preventing dehumanization in low-birth-rate societies. Policies should include subsidies for tools like explainable AI and privacy-by-design, reducing enforcement burdens in formalized OSAI structures. To raise public awareness and address social concerns, promote media literacy campaigns rooted in Asian values to build trust and reduce coordination frictions is actionable. Many Asian countries are transitioning from soft (guidelines) to hard (laws) regulations. This shift should be gradual, prioritizing ethical audits in high-risk areas like biometrics and employment to balance innovation with societal harmony. 6.2.2. Cross-border and policy suggestions To promote Asian values over Western individualism, policies should encourage collaborations that assert cultural sovereignty in global AI discourse. First and foremost, incentives for Asian AI cross-border OSAI projects shall be well-discussed, such as Japan-China-South Korea consortia on elder care AI, integrating Confucian familial obligations with technological augmentation. This reduces coordination costs by pooling resources and counters U.S.-EU dominance, as seen in China’s Global AI Governance Initiative engaging the Global South. Meanwhile, policies should prioritize the investments in emphasizing local data and philosophical anthropologies (e.g., excluding AI from personhood in Buddhist or Islamic views). International advocacy, such as through UNESCO’s 2021 Ethics Recommendation, could push for global standards incorporating Asian relational ethics. By implementing these expanded suggestions, Asian policymakers can optimize OSAI outcomes - enhancing knowledge convergence, market diffusion, and collaboration - while embedding values like harmony, equity, and sustainability. This not only addresses bias, privacy, ecological harm concerns, but also positions Asia as a leader in human-centered AI governance, fostering inclusive innovation in a multipolar world. Future research should evaluate these policies longitudinally to refine their impact on governance costs. 6.3. Limitation and future work This study is limited by its cross-sectional design, which constrains causal inference in dynamic AI contexts. Longitudinal analyses could capture feedback effects between governance costs and AI innovation outcomes. Comparative research across Asian institutional contexts is needed to test generalizability and refine policy suggestions. Future studies may also examine complementary mechanisms - such as trust in AI ethics, digital infrastructure for bias mitigation, or leadership rooted in Asian values - that condition the impact of governance costs. Despite these limitations, this study establishes a robust conceptual and empirical foundation for examining governance costs in open-source AI innovation. By integrating transaction cost economics, institutional theory, and AI governance perspectives, it reframes governance costs as a central theoretical construct for understanding how open, distributed organizations achieve sustainable and ethical AI innovation under varying institutional constraints in Asia. Declarations Please enter the following declarations: Statement on Ethics Approval and Consent to Participate The School of Humanities and Social Sciences and the School of Public Administration at Beihang University hereby confirm that the above-referenced study does not require formal ethics committee approval or Institutional Review Board (IRB) review. This research is based exclusively on the secondary analysis of fully anonymized and aggregated data sourced from events organized by the China Computer Federation (CCF). The datasets consist solely of non-identifiable information, such as anonymized participant rankings, scores, and statistical summaries derived from public scientific records. All direct identifiers (e.g., names, national ID numbers, contact details) were permanently removed or irreversibly transformed by the data provider prior to access by the research team. No personally identifiable information (PII) was available, and individual participants cannot be identified, directly or indirectly, from the data or the aggregated results presented in the study. In accordance with the Declaration of Helsinki, the U.S. Common Rule (45 CFR 46), the EU General Data Protection Regulation (GDPR) principles for research, and prevailing ethical standards for social sciences and public administration research in the People’s Republic of China, studies involving only fully anonymized or de-identified secondary data that pose no risk to participants are exempt from formal ethics committee review. This study is a retrospective observational analysis of existing aggregated public data, involving no interaction, intervention, biological materials, or contact with human subjects. References Autio E, Dahlander L, Frederiksen L (2013) Information exposure, opportunity evaluation, and entrepreneurial action: An investigation of an online user community. Acad Manag J 56(5):1348–1371 Bagozzi RP, Dholakia UM (2006) Open-source software user communities: A study of participation in Linux user groups. Manage Sci 52(7):1099–1115 Benkler Y (2016) Peer production, the commons, and the future of the firm. Strategic Organ 15(2):264–274 Benkler Y (2006) The Wealth of Networks: How Social Production Transforms Markets and Freedom. Yale University Press Benkler Y (2013) Practical anarchism peer mutualism, market power, and the fallible state. Politics Soc 41(2):213–251 Bessen J (2006) Open source software: Free provision of complex public goods. The economics of open source software development. Elsevier, pp 57–81 Bogers M, Zobel AK, Afuah A, Almirall E, Brunswicker S, Dahlander L, Ter Wal AL (2017) The open innovation research landscape: Established perspectives and emerging themes across different levels of analysis. Ind Innovat 24(1):8–40 Boumgarden P, Nickerson J, Zenger TR (2012) Sailing into the wind: Exploring the relationships among ambidexterity, vacillation, and organizational performance. Strateg Manag J 33(6):587–610 Cavallo A, Burgers H, Ghezzi A, Van de Vrande V (2022) The evolving nature of open innovation governance: A study of a digital platform development in collaboration with a big science centre. Technovation 116:102370 Chen J, Han L, Qu G (2020) Citizen innovation: Exploring the responsibility governance and cooperative mode of a post-schumpeter paradigm. J Open Innovation: Technol Market Complex 6(4):172 Chesbrough HW (2017) The Future of Open Innovation. Res Technol Manage 60(6):29–35 Chesbrough HW (2003) Open Innovation: The New Imperative for Creating and Profiting from Technology. Harvard Business School Press Coase RH (1937) The nature of the firm. Macmillan Education UK, pp 37–54 Crowston K, Howison J (2005) The social structure of free and open source software development. First Monday 10(2). https://doi.org/10.5210/fm.v10i2.1203 Dahlander L, Magnusson M (2008) Relationships between Open Source Firms and Communities: Observations from Nordic Firms. Res Policy 37(4):481–493 Dahlander L, Wallin MW (2020) Governing open innovation. What matters? Res Policy 49(1):109822 Fitzgerald B (2006) The Transformation of Open Source Software. MIS Q 30(3):587–598 Foss NJ, Klein PG (2012) Organizing entrepreneurial judgment: A new approach to the firm. Cambridge University Press Gabriel RP, Goldman R (2002) Open source: Beyond the fairytales. Perspect Bus Innov, (8): 24–40 Gambardella A, Panico C (2014) On the management of open innovation. Res Policy 43(5):903–913 Gerosa M, Wiese I, Trinkenreich B, Link G, Robles G, Treude C, Steinmacher I, Sarma A (2021) The shifting sands of motivation: Revisiting what drives contributors in open source. arXiv preprint. arXiv:2101.10291 Greiner M, Goodhue D (2005) Make-or-Buy in the Age of Open Source: A Transaction Cost Analysis. Proceedings of the International Conference on Information Systems (ICIS) Gulati R, Wohlgezogen F, Zhelyazkov P (2012) The two facets of collaboration: Cooperation and coordination in strategic alliances. Acad Manag Ann 6(1):531–583 Hall BH, Jaffe A, Trajtenberg M (2005) Market value and patent citations. RAND J Econ 36(1):16–38 Hall PA, Soskice D (2001) Varieties of Capitalism: The Institutional Foundations of Comparative Advantage. Oxford University Press Hauge Ø, Ayala C, Conradi R (2010) Adoption of open-source software in software-intensive organizations: a systematic literature review. ‎Inf Softw Technol 52(11):1133–1154 Hayes AF (2017) Introduction to mediation, moderation, and conditional process analysis: A regression-based approach. Guilford Hertel G, Niedner S, Hermann S (2003) Motivation of Software Developers in the F/OSS Projects:an Internet-based Sur-vey of Contributors to the Linux Kernel. Res Policy 327:1159–1177 Herzog P, Leker J (2010) Open and closed innovation – different innovation cultures for different strategies. Int J Technol Manage (IJTM) 52(3/4):322–343 Joblin M, Eckl-Ganser B, Bock T, Schmid A, Siegmund J, Apel S (2022) Hierarchical and Hybrid Organizational Structures in OS Software Projects: A Longitudinal Study. ACM Transactions on Software Engineering and Methodology Keles S (2025) Navigating in the moral landscape: analysing bias and discrimination in AI through philosophical inquiry. AI Ethics 5(1):555–565 Ke W, Zhang P (2010) The effects of extrinsic motivations and satisfaction in open-source software development. Journal of the Association for Information Systems, 2010, 11(12): 784–808 Lakhani KR, von Hippel E (2003) How Open Source Software Works: Free User-to-User Assistance. Res Policy 32(6):923–943 Lerner J, Tirole J (2002) Some Simple Economics of F/OSS. J Industrial Econ, (2), 197–234 Lerner J, Tirole J (2005) The economics of technology sharing: Open source and beyond. J Economic Perspect 19(2):99–120 Li W, Wu WJ, Wang HM, Cheng XQ, Chen HJ, Zhou ZH, Ding R (2017) Crowd intelligence in AI 2.0 era. Front Inform Technol Electron Eng 18(1):15–43 Liu X, Yang P, Ding X (2023) Development of China’s new energy vehicle industry under the central-local government policy synergy: The perspective of innovation ecosystem. China Soft Sci, (11):38–53. (In Chinese) March JG (2010) The Ambiguities of Experience. Cornell University Press McGahan AM, Silverman BS (2006) Profiting from technological innovation by others: The effect of competitor patenting on firm value. Res Policy 35(8):1222–1242 Mergel I (2016) Agile innovation management in government: A research agenda. Government Inform Q 33(3):516–523 O’Mahony S, Ferraro F (2007) The Emergence of Governance in Open-Source Software Communities. Acad Manag J 50(5):1079–1106 Ostrom E (1990) Governing the commons: The evolution of institutions for collective action. Cambridge University Press Powell WW (1990) Neither market nor hierarchy: Network forms of organization. Res Organizational Behav 12:295–336 Powell WW, Koput KW, Smith-Doerr L (1996) Interorganizational collaboration and the locus of innovation: Networks of learning in biotechnology. Adm Sci Q, (2): 116–145 Radziwon A, Bogers M (2019) Open innovation in SMEs: Exploring inter-organizational relationships in an ecosystem. Technol Forecast Soc Chang 146:573–587 Raymond ES, Brock G (2003) The Art of Unix Programming. Addison-Wesley Raymond ES (1999) The Cathedral and the Bazaar: Musings on Linux and Open Source by an Accidental Revolutionary. O’Reilly & Associates, Sebastopol, CA Remneland-Wikhamn B, Knights D (2012) Transaction Cost Economics and Open Innovation: Implications for Theory and Practice. Creativity Innov Manage 21(3):277–289. https://doi.org/10.1111/j.1467-8691.2012.00639.x Ryan RM, Deci EL (2000) Intrinsic and extrinsic motivations: Classic definitions and new directions. Contemp Educ Psychol 25(1):54–67 Schmidt KM, Schnitzer M (2002) Public subsidies for open source-some economic policy issues of the software market. Harv, vol 16. JL & Tech, p 473 Schweik CM (2014) Toward the Comparison of Open-Source Commons Institutions. Governing Knowledge Commons. eds. B. M. M. Schweik CM, English RC (2012) Internet Success: A Study of Open-Source Software Commons. MIT Press Secundo G, Spilotro C, Gast J, Corvello V (2025) The transformative power of artificial intelligence within innovation ecosystems: a review and a conceptual framework. RMS 19(9):2697–2728 Simon HA (1978) Rationality as process and as product of thought. Am Econ Rev 68(2):1–16 Shah SK (2006) Motivation, governance, and the viability of hybrid forms in open source software development. Manage Sci 52(7):1000–1014 Stead H (1976) The costs of technological innovation. Res Policy 5(1):2–9 Stewart KJ, Gosain S (2006) The impact of ideology on effectiveness in open source software development teams. MIS quarterly, pp 291–314 Taddeo M, Floridi L (2018) How AI can be a force for good. Science 361(6404):751–752 Teixeira J, Teixeira A, Coimbra T (2019) Governance in open source software: A systematic literature review. J Software: Evol Process, 31(5), e2160 Torvalds L (2021) An Interview with Linus Torvalds. - Open Source And Beyond Verdegem P (2024) Dismantling AI capitalism: the commons as an alternative to the power concentration of Big Tech, vol 39. AI & society, pp 727–737. 2 von Hippel E (2005) Democratizing Innovation. MIT Press von Hippel E, von Krogh G (2003) Open source software and the private-collective innovation model: Issues for organization science. Organ Sci 14(2):209–223 Von Krogh G, Haefliger S, Spaeth S, Wallin MW (2012) Carrots and Rainbows: Motivation and Social Practice in Open Source Software Development. MIS Q 36(2):649–676 von Krogh G, Spaeth S, Lakhani KR (2003) Community, Joining, and Specialization in Open Source Software Innovation: A Case Study. Res Policy 32(7):1217–1241 Wang HM, Yu Y, Wang T, Ding B (2023) Crowd Intelligence Paradigm: the new transformation of software development paradigm. SCIENTIA SINICA Informationis, (53): 1490–1502. (in Chinese) Wang S, Huang X, Zhang M, Bao S, Liu L, Fu X, Guan W (2025) Open science 2.0: revolutionizing spatiotemporal data sharing and collaboration. Comput Urban Sci 5(1):4 Wang Z (2024) Governance Mechanisms of Open-source Organizations: Research from the Perspective of Innovation Costs. Tsinghua University, Beijing. (in Chinese) Wang Z, Xue L, Zhao J (2024) The evolution of innovation costs and the dynamic governance mechanism of open-source innovation organizations. China Public Adm Rev 6(4):108–138 (in Chinese) Wang Z, Li Z, Yu Y (2024) Do community responses influence OSS contributor retention? A survival analysis. Sci China Inform Sci 67(8):184101 Weber S (2004) The success of open source. Harvard University Press West J, Gallagher S (2006) Challenges of Open Innovation: The Paradox of Firm Investment in Open-Source. R&D Manage 36(3):319–331 West J, O‘Mahony S (2008) Contrasting community building in sponsored and community founded open source projects. Proceedings of the 41st Annual Hawaii International Conference on System Sciences (HICSS) Williamson OE (1993) Transaction cost economics and organization theory. Ind Corp Change 2(2):107–156 Williamson OE (1996) The mechanisms of governance. Oxford University Press Witt MA, Jackson G (2016) Varieties of Capitalism and Institutional Comparative Advantage: A Test and Reinterpretation. J Int Bus Stud 47(5):778–806 Yin J, Jamali D (2021) Collide or collaborate: The interplay of competing logics and institutional work in cross-sector social partnerships. J Bus Ethics 169(4):673–694 Yin L, Chakraborty M, Schweik C, Frey S, Filkov V (2022) Open-Source Software Sustainability: Combining Institutional Analysis and Socio-Technical Networks. arXiv preprint arXiv:2203.03144 Additional Declarations The authors declare no competing interests. Supplementary Files Appendix.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8409070","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":564763412,"identity":"dbca2e42-df77-45e0-84de-9ced6967efab","order_by":0,"name":"Zhe Wang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA7klEQVRIiWNgGAWjYLACxgYGOfv2HijvAJFajA14zpCoJXGDRA6RWgyOnz388ucOG8btkm+PSfyoYZDju5HA+LkAn5YzeWkWkmfSmC1n56VJ9hxjMJa8kcAsPQOPFrMDOWYGhm2H2Rhu55hJ8LABXXgjgY2ZB5+W82/MDBLb/vMw3DxjJvnnH0M9YS03cowfHGw7IGFwg8dMmreNIcGAkBb7G2/MGBvbkg0ke/KSrWX7JAxnnnnYLI1Pi2R/jvHHn2129f3sZw/efPPNRp7vePLBz/i0AAGbBBIHxAZGEwHA/IGQilEwCkbBKBjhAAA+qU8WfI39XAAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0000-0002-7821-9366","institution":"Beihang University","correspondingAuthor":true,"prefix":"","firstName":"Zhe","middleName":"","lastName":"Wang","suffix":""}],"badges":[],"createdAt":"2025-12-20 03:06:28","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":true,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":true},"doi":"10.21203/rs.3.rs-8409070/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8409070/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":99793574,"identity":"d60aefef-9f6b-47fa-aac4-e902322cd99a","added_by":"auto","created_at":"2026-01-08 13:31:53","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":485967,"visible":true,"origin":"","legend":"","description":"","filename":"1.AISocietyTheCostsandOutcomesofOrganizingOpensourceAIInnovation.docx","url":"https://assets-eu.researchsquare.com/files/rs-8409070/v1/678e8f330ee46d26ecafb324.docx"},{"id":99579821,"identity":"ca71eafe-172f-461a-bc6d-7db9f5af34d2","added_by":"auto","created_at":"2026-01-06 05:58:55","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":342,"visible":true,"origin":"","legend":"","description":"","filename":"rs8409070.json","url":"https://assets-eu.researchsquare.com/files/rs-8409070/v1/acc7f26af526428b45ebe81e.json"},{"id":99793932,"identity":"f2c3c168-f1d4-4f0e-81a6-4887bdf7662d","added_by":"auto","created_at":"2026-01-08 13:33:33","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":221153,"visible":true,"origin":"","legend":"","description":"","filename":"rs84090700enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8409070/v1/4a1442da332b6105495210b6.xml"},{"id":99579826,"identity":"4eae28dc-a10c-42a2-b4b9-46e8f5302676","added_by":"auto","created_at":"2026-01-06 05:58:55","extension":"png","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":236370,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8409070/v1/e2886391a4c2091bf78dfdc1.png"},{"id":99579825,"identity":"0655a911-55af-46af-9236-dcf65b1c6214","added_by":"auto","created_at":"2026-01-06 05:58:55","extension":"jpeg","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":146822,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8409070/v1/f9309a81e3aa2c05de13bcaa.jpeg"},{"id":99579824,"identity":"8692a85d-f15f-4c0a-9a77-73f3df3cdd2f","added_by":"auto","created_at":"2026-01-06 05:58:55","extension":"png","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":69767,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8409070/v1/c203cf85fb4abffee1b3e81a.png"},{"id":99579827,"identity":"6173e639-0fd9-4394-8bc5-68b9c6d1750c","added_by":"auto","created_at":"2026-01-06 05:58:56","extension":"png","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":32780,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8409070/v1/69ac12c49bd71acc2f11cbbc.png"},{"id":99579829,"identity":"ad1bc12c-1db4-4240-a7f5-d012e957dcd5","added_by":"auto","created_at":"2026-01-06 05:58:56","extension":"xml","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":219197,"visible":true,"origin":"","legend":"","description":"","filename":"rs84090700structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8409070/v1/a84be88c61be0cbba8cedf2e.xml"},{"id":99579830,"identity":"562d9417-94ee-4ff9-87a5-7883b41aec50","added_by":"auto","created_at":"2026-01-06 05:58:56","extension":"html","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":233304,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8409070/v1/c537e502354eecfc2dfe0714.html"},{"id":99804100,"identity":"35de9ef4-ef90-40d6-b7fb-630f48f32abf","added_by":"auto","created_at":"2026-01-08 14:11:46","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1033265,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8409070/v1/d5111608-cd33-41af-8c92-e5e54b5e0b82.pdf"},{"id":99792987,"identity":"6b1fc62d-ea17-478c-801a-0c98ea67642b","added_by":"auto","created_at":"2026-01-08 13:30:47","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":425000,"visible":true,"origin":"","legend":"","description":"","filename":"Appendix.docx","url":"https://assets-eu.researchsquare.com/files/rs-8409070/v1/6607539e03b077111a4b9e77.docx"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"The Costs and Outcomes of Organizing Open-source AI Innovation: Survey Evidence from China","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eOpen-source AI innovation has emerged as a powerful engine of technological and industrial advancement in artificial intelligence, particularly in China, where it facilitates distributed creation, open access to AI models and datasets, modular development of neural networks, and collaboration across organizational and institutional boundaries to address societal challenges like healthcare diagnostics and smart city infrastructure (Wang et al, \u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). However, the openness that makes such AI innovation systems productive also generates complexity, especially in handling ethical issues such as bias in AI training data and privacy in large-scale deployments (Keles, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). As contributors vary in incentives, backgrounds, and institutional affiliations - often influenced by Chinese cultural emphases on collectivism and long-term societal harmony- organizations must establish mechanisms to coordinate AI development activities, mediate conflicts over ethical standards, and convert collective contributions into usable and ethically sound AI outcomes. These mechanisms incur governance costs that fundamentally shape the performance of \u003cb\u003eopen-source AI organizations\u003c/b\u003e (\u003cb\u003eOSAIOs\u003c/b\u003e) in China.\u003c/p\u003e \u003cp\u003eDespite substantial interest in open-source communities, existing research has not fully explained how governance costs arise within formal organizational settings for AI (Shah, \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; Stewart \u0026amp; Gosain, \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e2006\u003c/span\u003e), nor how these costs mediate the impact of governance arrangements on AI-specific innovation outcomes, such as the convergence of diverse AI algorithms or the ethical diffusion of AI technologies. This limitation is particularly salient in institutional contexts characterized by rapid technological and organizational growth, such as China\u0026rsquo;s expanding AI ecosystem, which is influenced by state policies promoting \u0026ldquo;AI for good\u0026rdquo; and cultural values prioritizing collective welfare over individual gain (Taddeo \u0026amp; Floridi, \u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e2018\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eChina provides a valuable setting for examining how governance structures function within formal OSAIOs, as its open-source AI ecosystem includes autonomous communities developing grassroots AI tools, consortium-based bazaars for collaborative AI dataset curation, firm-led platforms like those from Tencent or Alibaba integrating proprietary and open AI models, and public institution-driven organizations advancing national AI priorities through universities and government labs. These governance arrangements produce different cost profiles that are likely to influence how effectively AI knowledge is integrated, how AI innovations diffuse into markets with ethical safeguards, and how cross-sector collaboration is sustained to align with Asian values of harmony and inclusivity (Yin \u0026amp; Jamali, \u003cspan citationid=\"CR77\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Yet empirical evidence on these dynamics remains scarce, especially in relation to policy needs for AI open-source that incorporate non-Western perspectives.\u003c/p\u003e \u003cp\u003eTo address this gap, this study develops a cost-outcome framework that conceptualizes governance costs as the principal mechanism linking organizational type and AI innovation outcomes. We examine three forms of AI innovation outcomes - knowledge convergence in AI model refinement, market diffusion of AI applications, and industrial collaboration for ethical AI ecosystems (Secundo et al, \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) - and analyze how governance costs related to incentive coordination, rule enforcement, and value transformation mediate these outcomes. Using data collected from 601 experienced participants in Chinese OSAIOs, we analyze how governance arrangements create different cost structures and how these structures explain divergent AI innovation trajectories. The findings show that decentralized organizations achieve strong knowledge convergence through low coordination barriers in AI development, while formalized governance models, particularly firm-led and public institution-based structures, promote market penetration and collaboration through more predictable enforcement and value transformation mechanisms that resonate with Chinese emphases on societal stability and collective progress. Governance costs represent the decisive mechanism through which governance shapes AI innovation outcomes.\u003c/p\u003e \u003cp\u003eThis study therefore contributes by identifying governance costs as a theoretical foundation for understanding open-source AI governance and by documenting this mechanism empirically in an important Asian context, offering policy suggestions for other Asian societies to develop AI frameworks rooted in their cultural values, such as India\u0026rsquo;s focus on frugal innovation or Japan\u0026rsquo;s emphasis on human-AI harmony.\u003c/p\u003e"},{"header":"2. Literature review","content":"\u003cp\u003eEffective governance is central to sustaining innovation in open-source AI ecosystems, where collaboration spans organizational and institutional boundaries to develop ethical, transparent AI systems. Yet, despite extensive attention to community governance and institutional design, limited research has examined how different OSAIO types shape AI innovation outcomes through underlying cost mechanisms, particularly in Asian contexts where cultural values influence ethical considerations. This section develops the theoretical foundation for analyzing open-source AI organizations (OSAIOs) by reviewing key OSAIO types and conceptualizing governance costs as critical mediating mechanisms that connect organizational structures with AI innovation performance.\u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1. Costs and outcomes in OSAIOs\u003c/h2\u003e \u003cp\u003eThe concept of governance cost originates from transaction cost economics (TCE), which posits that governance structures evolve to mitigate inefficiencies caused by bounded rationality, opportunism, and asset specificity (Coase, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e1937\u003c/span\u003e; Williamson, \u003cspan citationid=\"CR75\" class=\"CitationRef\"\u003e1996\u003c/span\u003e). In the context of AI innovation governance, these costs refer to the total resource expenditures required to establish, coordinate, monitor, and transform collective AI activities, including ethical audits of algorithms (Boumgarden et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Benkler, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2013\u003c/span\u003e). Unlike traditional firms, OSAIOs operate at the intersection of community norms, market incentives, and institutional frameworks, facing hybrid governance challenges amplified by AI's unique risks like bias and opacity (Powell, 1996; O\u0026rsquo;Mahony \u0026amp; Ferraro, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2007\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eBuilding on this theoretical foundation, we define governance cost in OSAIOs as: \u003cem\u003e\u0026ldquo;the aggregate economic, cognitive, and institutional efforts required to align diverse participants, enforce shared rules including ethical guidelines, and convert collective outputs into value under conditions of openness and uncertainty in AI development\u003c/em\u003e\u0026rdquo;. (Stead, \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e1976\u003c/span\u003e; Cavallo et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Secundo et al, \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e summarizes the defining characteristics of governance costs in open-source AI organizations.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eCharacteristics of governance costs in open-source AI organizations\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCategory\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDescription\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAI-Specific Examples\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIncentive Coordination\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eResources to align motivations and mitigate free-riding\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAligning contributors' incentives for ethical AI data labeling\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRule Enforcement\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eExpenditures for monitoring and adapting rules\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eEnforcing bias mitigation protocols in AI models\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eValue Transformation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eInvestments to translate knowledge into usable innovations\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eConverting open AI datasets into ethically deployable applications\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"3\"\u003eTherefore, governance costs in OSAIOs can be conceptualized along three interrelated channels, each addressing a fundamental coordination problem in open AI ecosystems:\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThe first governance cost is incentive coordination cost, which is the resources devoted to aligning diverse motivations, attracting and retaining contributors, and mitigating free-riding or attrition in AI projects (Simon, \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e1978\u003c/span\u003e; Lerner \u0026amp; Tirole, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2002\u003c/span\u003e, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2005\u003c/span\u003e; von Krogh et al., \u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Ke \u0026amp; Zhang, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2010\u003c/span\u003e);\u003c/p\u003e \u003cp\u003eThe second governance cost is rule enforcement cost, which is the expenditures related to establishing, monitoring, and adapting formal or informal rules in rapidly changing AI technological and policy environments, including ethical standards (March, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2010\u003c/span\u003e; Weber, \u003cspan citationid=\"CR71\" class=\"CitationRef\"\u003e2004\u003c/span\u003e; Joblin et al., 2023);\u003c/p\u003e \u003cp\u003eThe last but not least is value transformation cost, which is the investments required to translate collective AI knowledge into usable innovations, manage intellectual property, and ensure equitable benefit distribution aligned with cultural values (Williamson, \u003cspan citationid=\"CR75\" class=\"CitationRef\"\u003e1996\u003c/span\u003e; Chesbrough, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Hall et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2005\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThese three channels function as mediating mechanisms that link OSAIO types to AI innovation outcomes (Schmidt \u0026amp; Schnitzer, \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2002\u003c/span\u003e). Each governance structure produces a distinctive cost profile, shaping performance across three dimensions (shown by Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e): knowledge convergence in AI, market diffusion of AI, and industrial collaboration for AI.\u003c/p\u003e \u003cp\u003eFor instance, autonomous communities benefit from low enforcement costs due to shared norms but face high transformation costs because of informal value appropriation in AI ethics (Ostrom, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e1990\u003c/span\u003e; von Krogh et al., \u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2012\u003c/span\u003e). In contrast, firm-led OSAIOs reduce transformation costs through centralized IP management yet incur higher incentive coordination costs when community trust declines, especially in ethically sensitive AI areas (West \u0026amp; O\u0026rsquo;Mahony, 2008). The effectiveness of a governance mode thus depends on its ability to balance these cost trade-offs and foster both efficiency and ethical collaboration in AI (Foss \u0026amp; Klein, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2012\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIn conclusion, governance costs serve as a comprehensive analytical lens for understanding how governance structures shape AI innovation outcomes in open-source ecosystems. By conceptualizing governance costs as dynamic, multidimensional mediators, this study advances theoretical understanding of open-source AI governance and provides actionable insights for optimizing AI innovation governance in emerging economies such as China, where policies must integrate values like collective harmony to address ethical concerns.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eCharacteristics of innovation outcomes in open-source AI organizations\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOutcome\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDescription\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAI-Specific Metrics\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eKnowledge Convergence\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIntegration of distributed AI expertise\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eConvergence of AI model architectures\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMarket Diffusion\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCommercial adoption of open-source AI solutions\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDeployment of open AI in industries\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIndustrial Collaboration\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDepth of partnerships for AI ecosystems\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCross-sector ethical AI initiatives\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2. OSAIO types and open-source AI innovation\u003c/h2\u003e \u003cp\u003eOpen-source AI innovation emerged from decentralized, self-organizing communities bound by shared technical values rather than formal hierarchies, as seen in early AI projects like TensorFlow (Hippel \u0026amp; Krogh, \u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e2003\u003c/span\u003e; Gabriel \u0026amp; Goldman, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2002\u003c/span\u003e). Early collaborations relied on voluntary participation and peer recognition rooted in ethical AI principles (Torvalds, \u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). These informal structures effectively fostered creativity and rapid AI knowledge exchange but struggled to maintain long-term stability as AI projects scaled, particularly with ethical challenges like algorithmic fairness.\u003c/p\u003e \u003cp\u003eAs open-source AI initiatives expanded, they adopted formalized governance mechanisms to provide stability, reduce transaction costs, and align incentives without suppressing openness, especially in China where state policies emphasize AI for societal benefit (Benkler, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; Bogers et al., \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Drawing on governance and organizational theory (Williamson, \u003cspan citationid=\"CR75\" class=\"CitationRef\"\u003e1996\u003c/span\u003e; Gulati et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2012\u003c/span\u003e), as shown by Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e, four ideal OSAIO types can be identified, each balancing openness and control through distinct institutional logics influenced by Asian values.\u003c/p\u003e \u003cp\u003eEach governance mode reflects a distinctive configuration of authority, incentives, and coordination for AI. Importantly, the institutionalization of governance introduces governance costs, which are the economic, cognitive, and organizational resources required to design, operate, and adapt governance structures that sustain AI innovation while addressing ethical concerns (Hauge, Ayala \u0026amp; Conradi, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2010\u003c/span\u003e). These costs can be viewed as both transactional relating to AI coordination and control and transformational relating to AI value creation and appropriation. While they impose operational burdens, they also function as strategic investments that enable scalability, stability, and ethical accountability in complex AI systems (McGahan \u0026amp; Silverman, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; Autio et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2013\u003c/span\u003e). The magnitude and composition of these costs vary with institutional context, organizational form, and temporal dynamics (Aghion \u0026amp; Howitt, 1992). Understanding their mechanisms is therefore essential for explaining how OSAIOs balance the efficiency of markets, the accountability of hierarchies, and the adaptability of networks in AI, particularly in Asia where cultural values shape policy responses (Powell, \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e1990\u003c/span\u003e; Benkler, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2016\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eFour ideal OSAIO types in China\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOSAIO Types\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDescription\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSupporting Literatures\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAutonomies\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eself-organized, meritocratic communities where technical contribution defines authority, emphasizing collective AI advancement\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRaymond, \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e1999\u003c/span\u003e; von Krogh et al., \u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2012\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBazaars\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFoundation- or alliance-based systems that coordinate diverse actors through modular AI design and participatory consensus, exemplified by Chinese AI consortia\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWest \u0026amp; O\u0026rsquo;Mahony, 2008; West \u0026amp; Gallagher, \u003cspan citationid=\"CR72\" class=\"CitationRef\"\u003e2006\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFirms\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCorporate-led entities integrating open and proprietary AI under strategic management, such as Baidu's PaddlePaddle\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDahlander \u0026amp; Wallin, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Radziwon \u0026amp; Bogers, \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2019\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePublic institutions\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGovernment- or university-led organizations advancing public AI missions through hybrid governance, aligning with Chinese values of harmony\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMergel, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Liu et al., \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"3. The development of hypotheses","content":"\u003cp\u003eBuilding on the governance cost framework, this section develops testable hypotheses linking OSAIO types to AI innovation outcomes in OSAIOs. It proceeds in two stages. First, we delineate how distinct governance structures shape AI innovation performance through variations in coordination, incentive alignment, and institutional embedding, with a focus on ethical AI policy needs. Second, we theorize three core governance cost mechanisms - incentive coordination, rule enforcement, and value transformation - that mediate these effects, thereby explaining the differential AI innovation trajectories across OSAIOs in China\u0026rsquo;s evolving open-source AI ecosystem.\u003c/p\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e3.1. Innovation outcomes in OSAIOs\u003c/h2\u003e \u003cp\u003eOpen-source AI innovation has undergone significant structural evolution, transitioning from informal community-driven initiatives to formally organized entities encompassing non-profit foundations and corporate-dominated structures for AI model development (Raymond, \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e1999\u003c/span\u003e; Benkler, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2013\u003c/span\u003e). This evolution reflects a broader transformation in collaborative AI, where OSAIO types critically shape the efficiency and sustainability of open-source AI ecosystems, especially in addressing ethical concerns like fairness in AI decision-making. In the Chinese context, open-source AI organizations (OSAIOs) exhibit a unique four-tier architecture integrating technology communities, social organizations, market entities, and government systems (Wang et al., \u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). These tiers manifest in four distinct OSAIO types: autonomies (self-organized technical communities for AI experimentation), bazaars (social organizations as collaborative platforms for AI ethics discussions), firms (enterprise-driven market entities for commercial AI), and public institutions (administratively governed hierarchical organizations for national AI strategies) (Gambardella \u0026amp; Panico, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2014\u003c/span\u003e; Wang, \u003cspan citationid=\"CR68\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThis diversity in governance structures underscores the adaptive nature of open-source AI innovation, balancing decentralized collaboration with strategic oversight to meet policy needs for ethical AI. For instance, corporate involvement often introduces resource efficiency and market alignment while incorporating Chinese values of collective progress, whereas community-driven autonomies prioritize knowledge democratization and rapid AI iteration with ethical inclusivity (Chesbrough, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Schweik, \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). The performance of these modes determines whether open-source AI initiatives achieve visibility or transformative impact, particularly in digital technology and knowledge-intensive sectors where collaboration scalability and ethical AI diffusion are pivotal.\u003c/p\u003e \u003cp\u003eThe efficacy of open-source OSAIO types is reflected in three core AI innovation outcomes: knowledge convergence, market diffusion, and industrial collaboration. Knowledge convergence emphasizes collective problem-solving and the generation of novel AI insights through transparent and open participation, including ethical dataset sharing (Schweik, \u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e2012\u003c/span\u003e). Market diffusion measures the commercialization and ecosystem expansion potential of open-source AI outputs, with safeguards against misuse (Lerner \u0026amp; Tirole, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2005\u003c/span\u003e; Bessen, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2006\u003c/span\u003e). Industrial collaboration reflects the capacity to integrate diverse stakeholders into a sustainable AI network that aligns with Asian ethical values (Li et al., \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Chen et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). These outcomes are interdependent: robust knowledge convergence underpins market diffusion, while industrial collaboration amplifies both, especially in policy-driven AI ecosystems (Gambardella \u0026amp; Panico, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2014\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe interplay between OSAIO types and AI innovation outcomes reveals inherent trade-offs. Autonomies excel in knowledge convergence due to their open, meritocratic culture but struggle with resource scalability for ethical AI scaling (Wang, Xue \u0026amp; Zhao, \u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Bazaars foster industrial collaboration through neutral platforms but face coordination challenges in AI ethics. Firm-driven governance accelerates market diffusion yet risks undermining openness in AI transparency. Public institutions provide stability and funding but lack agility in responding to ethical AI dilemmas. Hence, no single governance mode universally optimizes all AI innovation outcomes; rather, each exhibits comparative advantages across different dimensions, informing policies that integrate Chinese values like harmony.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eHypothesis\u003c/strong\u003e \u003cp\u003e \u003cb\u003ea (H1a)\u003c/b\u003e: \u003cem\u003eDecentralized OSAIO types are more effective than formalized governance models in knowledge convergence for AI innovation.\u003c/em\u003e\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eHypothesis\u003c/strong\u003e \u003cp\u003e \u003cb\u003eb (H1b)\u003c/b\u003e: \u003cem\u003eFormalized OSAIO types are more effective than decentralized models in market diffusion and industrial collaboration for AI innovation.\u003c/em\u003e\u003c/p\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e3.2. Governance costs in OSAIOs\u003c/h2\u003e \u003cp\u003eGovernance costs in OSAIOs arise from tensions between collective AI goals and individual motivations under bounded rationality, exacerbated by ethical uncertainties in AI (Williamson, \u003cspan citationid=\"CR74\" class=\"CitationRef\"\u003e1993\u003c/span\u003e; March, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2010\u003c/span\u003e). In early stages, ambiguous rules and weak institutions heighten these costs, manifesting as incentive coordination costs, which are resources expended to reconcile diverse motivations and sustain participation in AI projects (Lerner \u0026amp; Tirole, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2002\u003c/span\u003e; Ostrom, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e1990\u003c/span\u003e). Drawing on institutional and behavioral economics, we identify three motivational mechanisms - intrinsic, extrinsic, and outcome-based - that impose distinct alignment costs in AI contexts (Deci \u0026amp; Ryan, \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2000\u003c/span\u003e; von Krogh et al., \u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Wang, Li \u0026amp; Yu, \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Autonomies incur high costs maintaining intrinsic motivation for ethical AI, bazaars balance stakeholder incentives through legitimacy, firms invest in outcome-based alignment at the expense of AI openness, and public institutions rely on policy-backed but rigid incentives aligned with national AI ethics (Bagozzi \u0026amp; Dholakia, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2006\u003c/span\u003e). These variations influence AI innovation efficiency and ethical knowledge diffusion.\u003c/p\u003e \u003cp\u003eAs OSAIOs mature, tensions emerge between structured governance and decentralized AI collaboration, generating rule enforcement costs - expenses for monitoring compliance, resolving conflicts, and maintaining coordination in ethical AI (Williamson, \u003cspan citationid=\"CR75\" class=\"CitationRef\"\u003e1996\u003c/span\u003e; Weber, \u003cspan citationid=\"CR71\" class=\"CitationRef\"\u003e2004\u003c/span\u003e). Such costs arise from hierarchy-innovation misalignment, temporal trade-offs, and stability-flexibility dilemmas in AI policy. Autonomies minimize formal enforcement but suffer scalability limits for AI ethics; bazaars balance coordination via consensus mechanisms; firms reduce short-term costs through control but lose adaptability in ethical AI; public institutions incur high bureaucratic costs yet maintain stability for national AI goals. These patterns shape AI innovation efficiency and collaboration outcomes.\u003c/p\u003e \u003cp\u003eFinally, value transformation costs emerge when OSAIOs institutionalize and appropriate AI outputs, reflecting tensions between value creation and capture in ethical AI (Benkler, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; Chesbrough, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). These costs stem from power imbalances, path dependencies, and resource constraints that hinder equitable value realization in AI. Autonomies preserve openness but underperform in AI commercialization with ethics; bazaars ensure fair distribution but incur negotiation costs; firms commercialize efficiently but risk eroding community trust in AI transparency; public institutions sustain legitimacy but lack responsiveness to ethical AI needs. Balancing collaboration integrity with commercialization agility minimizes these costs, informing AI policies.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eHypothesis 2\u003c/strong\u003e \u003cp\u003e \u003cb\u003e(H2)\u003c/b\u003e: \u003cem\u003eIncentive coordination and value transformation costs have greater impact on achieving knowledge convergence and industrial collaboration outcomes in ethical AI governance, while rule enforcement cost plays a dominant role in market diffusion of ethical AI governance.\u003c/em\u003e\u003c/p\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"4. Research design","content":"\u003cp\u003eTo empirically examine how OSAIO types influence AI innovation outcomes through governance costs, this study adopts a rigorously structured survey-based research design. The following section outlines the data collection procedures, measurement development, and analytical strategy, ensuring methodological transparency and empirical robustness in linking theoretical constructs to observable AI innovation dynamics among Chinese open-source AI organizations.\u003c/p\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e4.1. Methodology and data collection\u003c/h2\u003e \u003cp\u003eThe study employs a structured survey design to investigate the relationship between OSAIO types, governance costs, and AI innovation outcomes in Chinese OSAIOs. A survey approach is particularly suitable given the limited availability of standardized data in China\u0026rsquo;s emerging open-source AI ecosystem and the need to capture organizational-level governance practices directly, including ethical AI considerations. Institutional collaboration with the China Computer Federation (CCF) and the OpenAtom Foundation ensured survey credibility and broad dissemination among qualified respondents focused on AI.\u003c/p\u003e \u003cp\u003ePrimary data were collected during the October 2023\u0026ndash;2024 China Open-Source Conference, the most comprehensive national forum for OSAIO leaders, maintainers, and community decision-makers in AI. The event provided a concentrated sampling frame of knowledgeable informants, enhancing both response quality and representativeness in AI contexts.\u003c/p\u003e \u003cp\u003eSurvey development followed a six-month, iterative process involving semi-structured interviews with 36 experts representing all four OSAIO types in AI. Through multiple pretests and pilot studies, the final instrument was refined to maximize content validity and contextual relevance, incorporating questions on ethical AI governance. It captured detailed measures of organizational characteristics, governance mechanisms, AI innovation outputs, and governance cost perceptions.\u003c/p\u003e \u003cp\u003eThe final digital survey yielded 601 valid responses out of 891 distributed questionnaires (valid response rate: 95.1% among completed attempts). Quality control procedures included completion time thresholds, IP and regional validation, and logical consistency checks. The respondent composition, which are technical contributors (42.3%), maintainers (26.8%), community managers (18.2%), and organizational leaders (12.7%), ensures diverse perspectives on AI governance and innovation processes.\u003c/p\u003e \u003cp\u003eTo mitigate potential biases, we conducted several diagnostic tests. Non-response bias was assessed by comparing early and late respondents, with no significant differences detected (p\u0026thinsp;\u0026gt;\u0026thinsp;0.05). Sample representativeness was validated against the 2023 China Open-Source Ecosystem Report, showing strong alignment across governance mode distributions in AI. Role-based filters ensured respondent authority, while ex-post harmonization with publicly available AI project data further enhanced the reliability of self-reported AI innovation indicators. These methodological safeguards strengthen the validity, generalizability, and analytical rigor of the dataset for AI policy insights.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e4.2. Measurements and variables\u003c/h2\u003e \u003cp\u003eThe constructs of the survey were operationalized through theoretically grounded, empirically validated multi-item scales in \u003cspan refid=\"Sec21\" class=\"InternalRef\"\u003eAppendix\u003c/span\u003e Figure A1, A2 and Table \u003cspan refid=\"Tab9\" class=\"InternalRef\"\u003eA1\u003c/span\u003e, \u003cspan refid=\"Tab10\" class=\"InternalRef\"\u003eA2\u003c/span\u003e. The descriptive statistics and correlation matrix are shown by Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cb\u003eOSAIO Types.\u003c/b\u003e Drawing on open-source AI governance literature (Herzog \u0026amp; Leker, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2010\u003c/span\u003e; Schweik, \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2014\u003c/span\u003e), OSAIOs are classified into four types: (1) autonomies, (2) bazaars, (3) firms, and (4) public institutions. Each category is represented by a dummy variable, with commercial firms as the reference group. In our sample, decentralized governance dominates: bazaars (59.73%) and autonomies (10.48%) together constitute 70.21%, followed by public institutions (12.65%) and firms (7.14%), reflecting China's AI landscape.\u003c/p\u003e \u003cp\u003e \u003cb\u003eInnovation Outcomes.\u003c/b\u003e AI innovation performance is conceptualized along three dimensions: (1) knowledge convergence is the integration of distributed AI expertise, including ethical model training; (2) market diffusion is the commercial adoption and spread of open-source AI solutions with ethical safeguards; (3) industrial collaboration is the depth of partnerships with industry stakeholders for ethical AI. Each construct was measured using five-point Likert scales, refined through expert review and pretesting to ensure validity and reliability (Cronbach\u0026rsquo;s \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\alpha\\:\\)\u003c/span\u003e\u003c/span\u003e \u0026gt; 0.70; AVE\u0026thinsp;\u0026gt;\u0026thinsp;0.50).\u003c/p\u003e \u003cp\u003e \u003cb\u003eGovernance Costs.\u003c/b\u003e Governance costs refer to the resources expended to coordinate AI activities, enforce rules including ethical guidelines, and align incentives within OSAIOs. We operationalize three complementary dimensions: (1) incentive coordination costs are disparities between contributor motivations and actual participation in AI ethics; (2) rule enforcement costs are difficulties in implementing and maintaining governance mechanisms amid AI technical and institutional complexity; (3) value transformation costs are inefficiencies arising from information asymmetry and multi-party reward distribution in AI. Each dimension is measured using multi-item Likert scales normalized to a 0\u0026ndash;1 range for analytical comparability.\u003c/p\u003e \u003cp\u003e \u003cb\u003eControl variables\u003c/b\u003e include respondent demographics (gender, age, education, etc.), time devoted to open-source AI work, and provincial fixed effects to account for regional institutional heterogeneity in AI policy. Confirmatory factor analysis confirmed strong convergent and discriminant validity (all factor loadings\u0026thinsp;\u0026gt;\u0026thinsp;0.70; CR\u0026thinsp;\u0026gt;\u0026thinsp;0.80).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eDescriptive statistics and correlation matrix\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariable\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMean\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSD\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1. Knowledge convergence\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e4.30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.87\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2. Market diffusion\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e4.20\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e3. Industrial collaboration\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e4.25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.72\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e.068\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e4.Incentive coordination costs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.17\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.31*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-0.28**\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-0.35**\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e5. Rule enforcement costs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.25**\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-0.32**\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-0.29**\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.42\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e6. Value transformation costs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.38**\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-0.35**\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-0.41**\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.37\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.39\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"8\"\u003eNote: \u003csup\u003e*\u003c/sup\u003e \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.1, \u003csup\u003e**\u003c/sup\u003e \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05, \u003csup\u003e***\u003c/sup\u003e \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.01\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eGiven the ordered categorical nature of AI innovation outcomes, we employed ordered logistic regression (ordinal logit) with maximum likelihood estimation. Brant tests confirmed that the proportional odds assumption was satisfied, validating model specification. Meanwhile, to test the mediating role of governance costs, we implemented bootstrapped indirect effects (5,000 resamples; bias-corrected CIs per Hayes, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). This approach allows simultaneous estimation of direct and indirect pathways and provides distribution-free, robust inference for AI contexts.\u003c/p\u003e \u003cp\u003eRobustness checks include: re-estimating models via OLS using composite AI innovation scores; testing alternative control variable sets; verifying absence of multicollinearity (VIF\u0026thinsp;\u0026lt;\u0026thinsp;3.5); conducting subgroup analyses by governance mode and likelihood-ratio tests for model comparison. This multi-method validation strategy ensures that results are statistically robust, theoretically coherent, and empirically reliable, which are presented in \u003cspan refid=\"Sec21\" class=\"InternalRef\"\u003eAppendix\u003c/span\u003e Table A4 to A6.\u003c/p\u003e \u003c/div\u003e"},{"header":"5. Empirical findings","content":"\u003cp\u003eThis section presents the empirical findings from ordered logistic regressions and structural equation modeling to examine how OSAIO types influence AI innovation outcomes directly and indirectly through governance costs. By testing the two hypotheses, the analysis provides systematic evidence on the distinct effects of governance structures and the mediating role of governance cost mechanisms in shaping open-source AI innovation performance in China.\u003c/p\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e5.1. Main effects\u003c/h2\u003e \u003cp\u003eThe ordered logistic regression analysis provides robust evidence supporting H1a and H1b, confirming that OSAIO types exert distinct and statistically significant effects on the three dimensions of AI innovation outcomes: knowledge convergence, market diffusion, and industrial collaboration. The models demonstrate satisfactory explanatory power (Pseudo R\u0026sup2;=0.032\u0026thinsp;~\u0026thinsp;0.042), consistent with established benchmarks in organizational and AI innovation studies.\u003c/p\u003e \u003cp\u003eFor knowledge convergence in AI, both autonomous communities (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\beta\\:\\:\\)\u003c/span\u003e\u003c/span\u003e= 1.186, p\u0026thinsp;\u0026lt;\u0026thinsp;0.05) and bazaar-style governance (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\beta\\:\\:\\)\u003c/span\u003e\u003c/span\u003e= 0.854, p\u0026thinsp;\u0026lt;\u0026thinsp;0.1) exhibit positive and significant coefficients. This suggests that decentralized, meritocratic structures foster AI knowledge sharing and collaborative problem-solving, including ethical algorithm refinement, consistent with prior findings on open coordination in AI (von Krogh et al., \u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; O\u0026rsquo;Mahony \u0026amp; Ferraro, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2007\u003c/span\u003e). In contrast, public institutions do not show significant influence in this domain, reflecting their weaker alignment with bottom-up AI knowledge generation.\u003c/p\u003e \u003cp\u003eFor market diffusion of AI, public institutions display the strongest positive effect (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\beta\\:\\:\\)\u003c/span\u003e\u003c/span\u003e= 0.987, p\u0026thinsp;\u0026lt;\u0026thinsp;0.05), underscoring their ability to leverage policy networks, legitimacy, and institutional infrastructure to scale ethical AI diffusion. The firm-led reference category also performs well, while autonomous communities show an insignificant and negative effect, suggesting that decentralized forms may excel at AI ideation but face challenges in commercialization and ethical diffusion due to weak market linkages.\u003c/p\u003e \u003cp\u003eFor industrial collaboration in AI, public institutions again demonstrate the most substantial effect (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\beta\\:\\:\\)\u003c/span\u003e\u003c/span\u003e= 1.245, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01), followed by bazaars (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\beta\\:\\:\\)\u003c/span\u003e\u003c/span\u003e= 0.432, p\u0026thinsp;\u0026lt;\u0026thinsp;0.1). This pattern highlights the unique role of institutional coordination in mobilizing multi-party collaboration across AI industry boundaries, while semi-formal bazaar structures facilitate inclusive participation and ethical AI exchange. The overall findings confirm that decentralized modes (autonomies and bazaars) are superior for AI knowledge integration, whereas formalized modes (firms and public institutions) are more effective for market and industrial outcomes in ethical AI.\u003c/p\u003e \u003cp\u003eCollectively, Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e presents the estimated main effects of OSAIO types on three AI innovation outcomes using ordered logistic regression. These results substantiate the theoretical argument that OSAIO types serve as distinct coordination architectures, each optimizing specific AI goals. They provide strong empirical support for H1, while establishing a foundation for subsequent mediation analyses.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eOrdered logistic regression results: main effects of OSAIO types\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariable\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eKnowledge convergence\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMarket diffusion\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eIndustrial collaboration\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAutonomies\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.186**\u003c/p\u003e \u003cp\u003e(0.215)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-0.324\u003c/p\u003e \u003cp\u003e(0.287)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.215\u003c/p\u003e \u003cp\u003e(0.302)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBazaars\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.854*\u003c/p\u003e \u003cp\u003e(0.342)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-0.187\u003c/p\u003e \u003cp\u003e(0.401)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.432*\u003c/p\u003e \u003cp\u003e(0.258)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePublic institutions\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.430\u003c/p\u003e \u003cp\u003e(0.376)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.987**\u003c/p\u003e \u003cp\u003e(0.321)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.245***\u003c/p\u003e \u003cp\u003e(0.287)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFirms (ref)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eControls\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePseudoR\u0026sup2;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.032\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.038\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.042\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eObservations\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e601\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e601\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e601\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"4\"\u003eNote: Robust standard errors in parentheses. \u003csup\u003e*\u003c/sup\u003e \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.1, \u003csup\u003e**\u003c/sup\u003e \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05, \u003csup\u003e***\u003c/sup\u003e \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.01\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e5.2. Mechanisms in governance costs\u003c/h2\u003e \u003cp\u003eTo further examine whether governance costs function as the key mechanisms through which OSAIO types influence AI innovation outcomes (H2), we applied bootstrap estimation with 5,000 replications. The indirect effects and confidence intervals confirm the statistical significance of all three governance cost dimensions - incentive coordination costs, rule enforcement costs, and value transformation costs - as mediators. This evidence highlights that governance costs constitute the micro-level transmission channels linking structural governance arrangements to AI innovation performance, with implications for ethical AI policy.\u003c/p\u003e \u003cp\u003eFor knowledge convergence outcome (in Table\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e), the results show that value transformation costs exert the strongest mediating effect (indirect effect = -0.287, 95% CI [-0.382, -0.198]), indicating that barriers in information exchange and unequal value distribution substantially inhibit the integration of dispersed AI knowledge resources across OSAIO types. Rule enforcement costs also demonstrate a meaningful mediation effect (indirect effect = -0.134, 95% CI [-0.217, -0.068]), suggesting that difficulties in coordinating and enforcing rules among multiple AI contributors weaken collective problem-solving and ethical knowledge co-creation. Although comparatively smaller in magnitude, the mediation effect of incentive coordination costs (indirect effect = -0.087, 95% CI [-0.152, -0.031]) remains significant, reflecting that motivational disparities between contributors and organizational objectives still create friction within collaborative AI innovation. Together, these results confirm that effective knowledge convergence in open-source AI organizations depends not only on the structural openness of OSAIO types but also on the efficiency with which specific governance costs are managed, especially those related to value transformation and enforcement consistency in ethical AI.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab6\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eMediation analysis for innovation outcome of OSAIOs: knowledge convergence\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariables\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIndirect effect\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eBootSE\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eBootLLCI\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eBootULCI\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIncentive coordination costs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-0.087\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.031\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.152\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-0.031\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRule enforcement costs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-0.134\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.038\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.217\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-0.068\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eValue transformation costs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-0.287\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.047\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.382\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-0.198\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eTurning to market diffusion outcome (in Table\u0026nbsp;\u003cspan refid=\"Tab7\" class=\"InternalRef\"\u003e7\u003c/span\u003e), rule enforcement costs emerge as the most critical mediating channel (indirect effect = -0.203, 95% CI [-0.301, -0.125]), underscoring that difficulties in implementing governance rules and adapting institutional arrangements under conditions of AI technological and regulatory uncertainty can significantly impede ethical AI diffusion. Value transformation costs also play a considerable role (indirect effect = -0.165, 95% CI [-0.241, -0.097]), indicating that asymmetric information and complex reward allocation mechanisms reduce incentives for broader market adoption of ethical AI. Incentive coordination costs, while statistically significant (indirect effect = -0.076, 90% CI [-0.138, -0.021]), contribute less prominently, suggesting that motivational issues matter less for AI diffusion outcomes than do structural coordination and enforcement challenges. These findings suggest that successful market diffusion in open-source AI ecosystems depends critically on the ability of governance actors to minimize coordination and enforcement costs, translating institutional stability into market legitimacy for ethical AI development.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab7\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 7\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eMediation analysis for innovation outcome of OSAIOs: market diffusion\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariable\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIndirect effect\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eBootSE\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eBootLLCI\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eBootULCI\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIncentive coordination costs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-0.076\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.030\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.138\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-0.021\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRule enforcement costs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-0.203\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.045\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.301\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-0.125\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eValue transformation costs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-0.165\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.037\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.241\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-0.097\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eFinally, in industrial collaboration outcome (in Table\u0026nbsp;\u003cspan refid=\"Tab8\" class=\"InternalRef\"\u003e8\u003c/span\u003e), the mediation analysis again identifies value transformation costs as the most substantial channel (indirect effect = -0.312, 95% CI [-0.408, -0.224]), highlighting that information opacity, attribution uncertainty, and inequitable distribution of jointly created value hinder the formation and sustainability of cross-sector partnerships for ethical AI. Rule enforcement costs (indirect effect = -0.156, 95% CI [-0.237, -0.089]) and incentive coordination costs (indirect effect = -0.123, 95% CI [-0.198, -0.062]) also significantly mediate the relationship, reflecting the coordination difficulties that arise in multi-stakeholder AI collaborations where divergent interests and governance expectations coexist. These results demonstrate that effective industrial collaboration in AI is highly sensitive to the management of governance costs, particularly those associated with the fair transformation and distribution of value in ethical contexts.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab8\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 8\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eMediation analysis for innovation outcome of OSAIOs: industrial collaboration\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariable\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIndirect effect\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eBootSE\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eBootLLCI\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eBootULCI\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIncentive coordination costs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-0.123\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.035\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.198\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-0.062\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRule enforcement costs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-0.156\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.038\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.237\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-0.089\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eValue transformation costs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-0.312\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.047\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.408\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-0.224\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e5.3. Discussion: balancing coordination and creativity in Asian AI innovation\u003c/h2\u003e \u003cp\u003eSynthesizing these empirical findings reveals a coherent mechanism through which OSAIO types shape AI innovation outcomes via differentiated governance costs. Distinct governance structures engender unique constellations of transaction, coordination, and motivational costs that mediate their AI innovation performance. Decentralized modes, such as autonomies and bazaars, demonstrate superior knowledge convergence due to their openness, peer recognition, and inclusive contribution norms, echoing findings in studies of community-anchored open-source AI (von Krogh, Spaeth, \u0026amp; Lakhani, \u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e2003\u003c/span\u003e; Lakhani \u0026amp; von Hippel, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2003\u003c/span\u003e). However, they also face elevated value transformation and coordination costs stemming from fragmented accountability, information asymmetry, and weaker institutional enforcement- similar to observations in \u0026ldquo;private-collective\u0026rdquo; models of AI innovation where public goods logic clashes with excludability pressures in ethical AI (von Hippel \u0026amp; von Krogh, \u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e2003\u003c/span\u003e; Verdegem, \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIn contrast, more formalized structures- firms and public institutions - excel in market diffusion and industrial collaboration by leveraging concentrated resources, institutional legitimacy, and standardized procedures for ethical AI, as observed in studies of firm-community relationships and institutional sponsorship in open-source AI (Dahlander \u0026amp; Magnusson, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2008\u003c/span\u003e). Yet, these formal modes encounter higher rule enforcement costs that may constrain adaptability and inclusiveness - parallel to findings in research on governance rigidity and its impact on AI innovation performance in hierarchical or firm-led open-source engagements (Fitzgerald, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; Crowston \u0026amp; Howison, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2005\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThese patterns suggest that governance costs are not simply inefficiencies to be minimized but structural mechanisms that translate organizational design into AI innovation outcomes, with policy implications for ethical AI in Asian countries. Among the three cost types, value transformation costs emerge as the most pervasive constraint, reflecting the intrinsic tension between openness and equitable value distribution in AI, where cultural values like Chinese harmony can guide fair allocation (Bessen, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; Baldwin \u0026amp; von Hippel, 2011). Rule enforcement costs prove decisive for market-oriented outcomes, where procedural clarity and compliance are essential for scaling ethical AI across diverse stakeholders - a theme echoed in research on institutional legitimacy and boundary work in OSAIOs. Incentive coordination costs, while weaker in effect, remain a persistent source of friction in sustaining voluntary engagement for ethical AI, consistent with work on motivation shifts in open-source AI contributors (Gerosa et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eTogether, these mediating dynamics reveal a fundamental trade-off between openness and control that defines the governance of open-source AI ecosystems: decentralized structures foster creativity but risk inefficiency and value leakage in ethical AI, whereas formalized systems secure coordination and diffusion but may suppress diversity and slow boundary spanning - an insight that aligns with debates in open AI innovation and institutional variety (Chesbrough, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2003\u003c/span\u003e; Hall \u0026amp; Soskice, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2001\u003c/span\u003e). Overall, the findings provide compelling evidence that governance costs serve as the underlying mechanisms linking OSAIO types to AI innovation outcomes. Specifically, value transformation costs most strongly mediate knowledge and collaboration processes in AI, whereas rule enforcement costs play a dominant role in market-oriented AI innovation. Incentive coordination costs remain relevant across contexts but exert relatively smaller effects. This pattern reveals a nuanced mechanism: decentralized OSAIO types enhance openness and participation but incur higher coordination and value transformation costs, while more formalized modes achieve efficiency in diffusion and collaboration but are constrained by enforcement rigidity. By delineating these mechanisms, the analysis not only validates the hypotheses but also advances our understanding of how governance costs operationalize the broader relationship between structural design and AI innovation performance in open-source ecosystems, offering a basis for Asian AI policies rooted in cultural values.\u003c/p\u003e \u003c/div\u003e"},{"header":"6. Contribution and implications","content":"\u003cp\u003eThis study examines how governance costs shape AI innovation outcomes in open-source AI organizations (OSAIOs) in China. Rather than emphasizing organizational typologies, the analysis identifies governance costs as the key explanatory mechanism linking governance design with AI innovation outcomes, particularly in addressing ethical and social concerns (Chesbrough, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2003\u003c/span\u003e). Evidence from Chinese OSAIOs shows that the configuration and management of governance costs largely determine differences in knowledge convergence for AI models, market diffusion of ethical AI applications, and industrial collaboration for sustainable AI ecosystems.\u003c/p\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e6.1. Theoretical contributions\u003c/h2\u003e \u003cp\u003eThe study makes three theoretical contributions to the literature on open-source AI innovation and governance (von Krogh, Spaeth, and Lakhani, \u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e2003\u003c/span\u003e). First, it develops and tests a governance cost framework that specifies three functional cost categories -incentive coordination, rule enforcement, and value transformation -and demonstrates their distinct effects on AI innovation outcomes. Value transformation costs are positively associated with knowledge convergence and collaboration in ethical AI, whereas rule enforcement costs primarily facilitate market diffusion of transparent AI. These results clarify how the efficiency and allocation of governance costs explain heterogeneity in AI innovation outcomes across organizations, especially in Asian contexts.\u003c/p\u003e \u003cp\u003eSecond, the results highlight that governance costs vary systematically across organizational arrangements (Schweik \u0026amp; English, \u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Remneland-Wikhamn \u0026amp; Knights, \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Greiner \u0026amp; Goodhue, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2005\u003c/span\u003e). Autonomous and bazaar forms tend to lower coordination barriers and enhance AI knowledge integration, while firm-led and public institution-based forms strengthen enforcement and resource mobilization for ethical AI. The comparative evidence suggests that governance effectiveness depends on the alignment between cost structures and AI innovation objectives rather than on any particular governance mode, informing policies that balance openness with cultural values.\u003c/p\u003e \u003cp\u003eThird, the study conceptualizes governance costs as purposeful, not merely frictional (Chakraborti et al., 2023; Yin et al., \u003cspan citationid=\"CR78\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Governance costs represent strategic investments in incentive coordination, rule enforcement, and value transformation under open and uncertain AI conditions, ensuring ethical alignment. This perspective integrates transaction cost economics with AI innovation governance, extending cost-based reasoning beyond firm boundaries to distributed AI ecosystems in Asia.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e6.2. Policy implications for Asian AI governance and more\u003c/h2\u003e \u003cp\u003eBased on the cost-outcome framework developed in this study, which highlights the mediating role of incentive coordination, rule enforcement, and value transformation costs in shaping open-source AI (OSAI) innovation outcomes, we expand the policy suggestions to provide more comprehensive, regionally tailored guidance. These recommendations emphasize reducing governance costs while integrating Asian value systems, such as Confucian harmony (he, \u0026ldquo;和\u0026rdquo;), collectivism, human-centeredness (ren or benevolence, \u0026ldquo;仁\u0026rdquo;), environmental stewardship rooted in Buddhist and Daoist philosophies, and pragmatic innovation aligned with diverse cultural contexts. By drawing on these values, Asian policymakers can counterbalance Western-dominated AI frameworks (e.g., the EU\u0026rsquo;s precautionary, rights-based approach or the U.S.\u0026rsquo;s market-driven model) and foster \u0026ldquo;sovereign AI\u0026rdquo; that prioritizes societal well-being, equity, and regional cooperation. This expansion incorporates insights from recent developments, including the ASEAN Guide on AI Governance and Ethics (2024), national strategies across Asia, and ethical concerns like bias mitigation, ecological sustainability, and cultural preservation.\u003c/p\u003e \u003cp\u003eFor policymakers, fostering an enabling institutional environment that reduces systemic transaction costs in AI is critical. Rather than prescribing a single model (Hertel et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2003\u003c/span\u003e; von Hippel, \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e2005\u003c/span\u003e), policies should support diverse organizational forms and shared infrastructure that facilitate ethical collaboration and diffusion across sectors, countering the dominance of EU, US, and even Chinese frameworks. In China, policies could incentivize public institution-led OSAIOs to integrate collectivist values for ethical AI diffusion, while in other Asian countries like South Korea or Singapore, frameworks could emphasize Confucian harmony or pragmatic innovation to develop national AI capacities. This includes subsidies for value transformation in ethical AI, regulatory sandboxes for rule enforcement testing, and incentives for cross-border Asian AI collaborations to promote shared values over Western individualism.\u003c/p\u003e \u003cp\u003ePolicies should not prescribe a uniform model but support diverse OSAI organizational types (autonomies, bazaars, firms, and public institutions) through flexible, adaptive frameworks. This aligns with the study\u0026rsquo;s findings that decentralized structures excel in knowledge convergence (e.g., collaborative AI model development) but incur higher coordination and value transformation costs, while formalized ones enhance market diffusion and industrial collaboration but face enforcement rigidities. Recommendations are structured at national, regional, and cross-border levels, with specific examples for key Asian countries and groupings.\u003c/p\u003e \u003cdiv id=\"Sec18\" class=\"Section3\"\u003e \u003ch2\u003e6.2.1. National level policy implications\u003c/h2\u003e \u003cp\u003eAt the national level, policies should focus on building AI ecosystems that minimize governance costs by investing in human capital, infrastructure, and ethical tools, while embedding cultural values to ensure AI serves collective harmony and well-being rather than exacerbating inequalities. To reduce incentive coordination costs (e.g., aligning diverse motivations in OSAI communities), governments should subsidize upskilling programs that incorporate Asian ethical traditions. For instance, in China, expand existing initiatives like the National AI Talent Development Plan to include Confucian virtues training, emphasizing xin (trustworthiness, \u0026ldquo;信\u0026rdquo;) in AI collaboration to sustain contributor engagement. In South Korea, leverage the country\u0026rsquo;s high-stakes education system by integrating AI ethics curricula based on Confucian harmony, addressing concerns like AI misuse in competitive exams (e.g., through programs similar to the proposed AI Responsibility Act of 2023). In India, promote Gandhian principles of equitable access by funding frugal AI training for underserved populations, mitigating biases against marginalized groups like Dalits and ensuring diverse datasets reflect India\u0026rsquo;s multicultural fabric. Subsidies could cover certification programs to build a workforce skilled in ethical AI, reducing attrition in OSAI projects.\u003c/p\u003e \u003cp\u003eTo supporting innovation ecosystems and reduce value transformation costs, policy-makers shall provide grants and shared infrastructure to lower the costs of translating collective AI outputs into equitable value, aligning with human-centered values. In Japan, build on \u0026ldquo;Society 5.0\u0026rdquo; principles - emphasizing human-AI coexistence and Daoist-inspired environmental balance - by funding open-source repositories for sustainable AI (e.g., low-energy models to address ecological destruction from data centers). India\u0026rsquo;s National Strategy for AI could offer tax incentives for OSAI start-ups using local, bias-free data, promoting \u0026ldquo;sovereign AI\u0026rdquo; to counter colonial dependencies on Western tech. In Singapore, expand regulatory sandboxes to test value transformation mechanisms, such as attribution systems for fair benefit distribution in AI, fostering pragmatic innovation while ensuring transparency.\u003c/p\u003e \u003cp\u003eTo invest in ethical R\u0026amp;D and rule enforcement and minimize rule enforcement costs, policy-makers should amid rapid AI evolution, allocate R\u0026amp;D funding for tools that embed cultural safeguards. In Thailand and Vietnam, under ASEAN frameworks, investments in cybersecurity-aligned AI research, drawing on Buddhist teleology for nature-respecting systems shall be encouraged. South Korea could enact hard regulations like the pending AI laws, incorporating Confucian relational ethics to enforce accountability in elder care AI, preventing dehumanization in low-birth-rate societies. Policies should include subsidies for tools like explainable AI and privacy-by-design, reducing enforcement burdens in formalized OSAI structures.\u003c/p\u003e \u003cp\u003eTo raise public awareness and address social concerns, promote media literacy campaigns rooted in Asian values to build trust and reduce coordination frictions is actionable. Many Asian countries are transitioning from soft (guidelines) to hard (laws) regulations. This shift should be gradual, prioritizing ethical audits in high-risk areas like biometrics and employment to balance innovation with societal harmony.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section3\"\u003e \u003ch2\u003e6.2.2. Cross-border and policy suggestions\u003c/h2\u003e \u003cp\u003eTo promote Asian values over Western individualism, policies should encourage collaborations that assert cultural sovereignty in global AI discourse. First and foremost, incentives for Asian AI cross-border OSAI projects shall be well-discussed, such as Japan-China-South Korea consortia on elder care AI, integrating Confucian familial obligations with technological augmentation. This reduces coordination costs by pooling resources and counters U.S.-EU dominance, as seen in China\u0026rsquo;s Global AI Governance Initiative engaging the Global South. Meanwhile, policies should prioritize the investments in emphasizing local data and philosophical anthropologies (e.g., excluding AI from personhood in Buddhist or Islamic views). International advocacy, such as through UNESCO\u0026rsquo;s 2021 Ethics Recommendation, could push for global standards incorporating Asian relational ethics.\u003c/p\u003e \u003cp\u003eBy implementing these expanded suggestions, Asian policymakers can optimize OSAI outcomes - enhancing knowledge convergence, market diffusion, and collaboration - while embedding values like harmony, equity, and sustainability. This not only addresses bias, privacy, ecological harm concerns, but also positions Asia as a leader in human-centered AI governance, fostering inclusive innovation in a multipolar world. Future research should evaluate these policies longitudinally to refine their impact on governance costs.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003e6.3. Limitation and future work\u003c/h2\u003e \u003cp\u003eThis study is limited by its cross-sectional design, which constrains causal inference in dynamic AI contexts. Longitudinal analyses could capture feedback effects between governance costs and AI innovation outcomes. Comparative research across Asian institutional contexts is needed to test generalizability and refine policy suggestions. Future studies may also examine complementary mechanisms - such as trust in AI ethics, digital infrastructure for bias mitigation, or leadership rooted in Asian values - that condition the impact of governance costs. Despite these limitations, this study establishes a robust conceptual and empirical foundation for examining governance costs in open-source AI innovation. By integrating transaction cost economics, institutional theory, and AI governance perspectives, it reframes governance costs as a central theoretical construct for understanding how open, distributed organizations achieve sustainable and ethical AI innovation under varying institutional constraints in Asia.\u003c/p\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003ePlease enter the following declarations: Statement on Ethics Approval and Consent to Participate The School of Humanities and Social Sciences and the School of Public Administration at Beihang University hereby confirm that the above-referenced study does not require formal ethics committee approval or Institutional Review Board (IRB) review. This research is based exclusively on the secondary analysis of fully anonymized and aggregated data sourced from events organized by the China Computer Federation (CCF). The datasets consist solely of non-identifiable information, such as anonymized participant rankings, scores, and statistical summaries derived from public scientific records. All direct identifiers (e.g., names, national ID numbers, contact details) were permanently removed or irreversibly transformed by the data provider prior to access by the research team. No personally identifiable information (PII) was available, and individual participants cannot be identified, directly or indirectly, from the data or the aggregated results presented in the study. In accordance with the Declaration of Helsinki, the U.S. Common Rule (45 CFR 46), the EU General Data Protection Regulation (GDPR) principles for research, and prevailing ethical standards for social sciences and public administration research in the People’s Republic of China, studies involving only fully anonymized or de-identified secondary data that pose no risk to participants are exempt from formal ethics committee review. This study is a retrospective observational analysis of existing aggregated public data, involving no interaction, intervention, biological materials, or contact with human subjects.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAutio E, Dahlander L, Frederiksen L (2013) Information exposure, opportunity evaluation, and entrepreneurial action: An investigation of an online user community. Acad Manag J 56(5):1348\u0026ndash;1371\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBagozzi RP, Dholakia UM (2006) Open-source software user communities: A study of participation in Linux user groups. Manage Sci 52(7):1099\u0026ndash;1115\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBenkler Y (2016) Peer production, the commons, and the future of the firm. Strategic Organ 15(2):264\u0026ndash;274\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBenkler Y (2006) The Wealth of Networks: How Social Production Transforms Markets and Freedom. Yale University Press\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBenkler Y (2013) Practical anarchism peer mutualism, market power, and the fallible state. Politics Soc 41(2):213\u0026ndash;251\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBessen J (2006) Open source software: Free provision of complex public goods. The economics of open source software development. Elsevier, pp 57\u0026ndash;81\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBogers M, Zobel AK, Afuah A, Almirall E, Brunswicker S, Dahlander L, Ter Wal AL (2017) The open innovation research landscape: Established perspectives and emerging themes across different levels of analysis. Ind Innovat 24(1):8\u0026ndash;40\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBoumgarden P, Nickerson J, Zenger TR (2012) Sailing into the wind: Exploring the relationships among ambidexterity, vacillation, and organizational performance. Strateg Manag J 33(6):587\u0026ndash;610\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCavallo A, Burgers H, Ghezzi A, Van de Vrande V (2022) The evolving nature of open innovation governance: A study of a digital platform development in collaboration with a big science centre. Technovation 116:102370\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen J, Han L, Qu G (2020) Citizen innovation: Exploring the responsibility governance and cooperative mode of a post-schumpeter paradigm. J Open Innovation: Technol Market Complex 6(4):172\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChesbrough HW (2017) The Future of Open Innovation. Res Technol Manage 60(6):29\u0026ndash;35\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChesbrough HW (2003) Open Innovation: The New Imperative for Creating and Profiting from Technology. Harvard Business School Press\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCoase RH (1937) The nature of the firm. Macmillan Education UK, pp 37\u0026ndash;54\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCrowston K, Howison J (2005) The social structure of free and open source software development. First Monday 10(2). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.5210/fm.v10i2.1203\u003c/span\u003e\u003cspan address=\"10.5210/fm.v10i2.1203\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDahlander L, Magnusson M (2008) Relationships between Open Source Firms and Communities: Observations from Nordic Firms. Res Policy 37(4):481\u0026ndash;493\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDahlander L, Wallin MW (2020) Governing open innovation. What matters? Res Policy 49(1):109822\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFitzgerald B (2006) The Transformation of Open Source Software. MIS Q 30(3):587\u0026ndash;598\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFoss NJ, Klein PG (2012) Organizing entrepreneurial judgment: A new approach to the firm. Cambridge University Press\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGabriel RP, Goldman R (2002) Open source: Beyond the fairytales. Perspect Bus Innov, (8): 24\u0026ndash;40\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGambardella A, Panico C (2014) On the management of open innovation. Res Policy 43(5):903\u0026ndash;913\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGerosa M, Wiese I, Trinkenreich B, Link G, Robles G, Treude C, Steinmacher I, Sarma A (2021) The shifting sands of motivation: Revisiting what drives contributors in open source. arXiv preprint. arXiv:2101.10291\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGreiner M, Goodhue D (2005) Make-or-Buy in the Age of Open Source: A Transaction Cost Analysis. Proceedings of the International Conference on Information Systems (ICIS)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGulati R, Wohlgezogen F, Zhelyazkov P (2012) The two facets of collaboration: Cooperation and coordination in strategic alliances. Acad Manag Ann 6(1):531\u0026ndash;583\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHall BH, Jaffe A, Trajtenberg M (2005) Market value and patent citations. RAND J Econ 36(1):16\u0026ndash;38\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHall PA, Soskice D (2001) Varieties of Capitalism: The Institutional Foundations of Comparative Advantage. Oxford University Press\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHauge \u0026Oslash;, Ayala C, Conradi R (2010) Adoption of open-source software in software-intensive organizations: a systematic literature review. \u0026lrm;Inf Softw Technol 52(11):1133\u0026ndash;1154\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHayes AF (2017) Introduction to mediation, moderation, and conditional process analysis: A regression-based approach. Guilford\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHertel G, Niedner S, Hermann S (2003) Motivation of Software Developers in the F/OSS Projects:an Internet-based Sur-vey of Contributors to the Linux Kernel. Res Policy 327:1159\u0026ndash;1177\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHerzog P, Leker J (2010) Open and closed innovation \u0026ndash; different innovation cultures for different strategies. Int J Technol Manage (IJTM) 52(3/4):322\u0026ndash;343\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJoblin M, Eckl-Ganser B, Bock T, Schmid A, Siegmund J, Apel S (2022) Hierarchical and Hybrid Organizational Structures in OS Software Projects: A Longitudinal Study. ACM Transactions on Software Engineering and Methodology\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKeles S (2025) Navigating in the moral landscape: analysing bias and discrimination in AI through philosophical inquiry. AI Ethics 5(1):555\u0026ndash;565\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKe W, Zhang P (2010) The effects of extrinsic motivations and satisfaction in open-source software development. Journal of the Association for Information Systems, 2010, 11(12): 784\u0026ndash;808\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLakhani KR, von Hippel E (2003) How Open Source Software Works: Free User-to-User Assistance. Res Policy 32(6):923\u0026ndash;943\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLerner J, Tirole J (2002) Some Simple Economics of F/OSS. J Industrial Econ, (2), 197\u0026ndash;234\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLerner J, Tirole J (2005) The economics of technology sharing: Open source and beyond. J Economic Perspect 19(2):99\u0026ndash;120\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi W, Wu WJ, Wang HM, Cheng XQ, Chen HJ, Zhou ZH, Ding R (2017) Crowd intelligence in AI 2.0 era. Front Inform Technol Electron Eng 18(1):15\u0026ndash;43\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu X, Yang P, Ding X (2023) Development of China\u0026rsquo;s new energy vehicle industry under the central-local government policy synergy: The perspective of innovation ecosystem. China Soft Sci, (11):38\u0026ndash;53. (In Chinese)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMarch JG (2010) The Ambiguities of Experience. Cornell University Press\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMcGahan AM, Silverman BS (2006) Profiting from technological innovation by others: The effect of competitor patenting on firm value. Res Policy 35(8):1222\u0026ndash;1242\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMergel I (2016) Agile innovation management in government: A research agenda. Government Inform Q 33(3):516\u0026ndash;523\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eO\u0026rsquo;Mahony S, Ferraro F (2007) The Emergence of Governance in Open-Source Software Communities. Acad Manag J 50(5):1079\u0026ndash;1106\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOstrom E (1990) Governing the commons: The evolution of institutions for collective action. Cambridge University Press\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePowell WW (1990) Neither market nor hierarchy: Network forms of organization. Res Organizational Behav 12:295\u0026ndash;336\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePowell WW, Koput KW, Smith-Doerr L (1996) Interorganizational collaboration and the locus of innovation: Networks of learning in biotechnology. Adm Sci Q, (2): 116\u0026ndash;145\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRadziwon A, Bogers M (2019) Open innovation in SMEs: Exploring inter-organizational relationships in an ecosystem. Technol Forecast Soc Chang 146:573\u0026ndash;587\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRaymond ES, Brock G (2003) The Art of Unix Programming. Addison-Wesley\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRaymond ES (1999) The Cathedral and the Bazaar: Musings on Linux and Open Source by an Accidental Revolutionary. O\u0026rsquo;Reilly \u0026amp; Associates, Sebastopol, CA\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRemneland-Wikhamn B, Knights D (2012) Transaction Cost Economics and Open Innovation: Implications for Theory and Practice. Creativity Innov Manage 21(3):277\u0026ndash;289. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/j.1467-8691.2012.00639.x\u003c/span\u003e\u003cspan address=\"10.1111/j.1467-8691.2012.00639.x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRyan RM, Deci EL (2000) Intrinsic and extrinsic motivations: Classic definitions and new directions. Contemp Educ Psychol 25(1):54\u0026ndash;67\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchmidt KM, Schnitzer M (2002) Public subsidies for open source-some economic policy issues of the software market. Harv, vol 16. JL \u0026amp; Tech, p 473\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchweik CM (2014) Toward the Comparison of Open-Source Commons Institutions. Governing Knowledge Commons. eds. B. M. M.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchweik CM, English RC (2012) Internet Success: A Study of Open-Source Software Commons. MIT Press\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSecundo G, Spilotro C, Gast J, Corvello V (2025) The transformative power of artificial intelligence within innovation ecosystems: a review and a conceptual framework. RMS 19(9):2697\u0026ndash;2728\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSimon HA (1978) Rationality as process and as product of thought. Am Econ Rev 68(2):1\u0026ndash;16\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShah SK (2006) Motivation, governance, and the viability of hybrid forms in open source software development. Manage Sci 52(7):1000\u0026ndash;1014\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStead H (1976) The costs of technological innovation. Res Policy 5(1):2\u0026ndash;9\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStewart KJ, Gosain S (2006) The impact of ideology on effectiveness in open source software development teams. MIS quarterly, pp 291\u0026ndash;314\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTaddeo M, Floridi L (2018) How AI can be a force for good. Science 361(6404):751\u0026ndash;752\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTeixeira J, Teixeira A, Coimbra T (2019) Governance in open source software: A systematic literature review. J Software: Evol Process, 31(5), e2160\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTorvalds L (2021) An Interview with Linus Torvalds. - Open Source And Beyond\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVerdegem P (2024) Dismantling AI capitalism: the commons as an alternative to the power concentration of Big Tech, vol 39. AI \u0026amp; society, pp 727\u0026ndash;737. 2\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evon Hippel E (2005) Democratizing Innovation. MIT Press\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evon Hippel E, von Krogh G (2003) Open source software and the private-collective innovation model: Issues for organization science. Organ Sci 14(2):209\u0026ndash;223\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVon Krogh G, Haefliger S, Spaeth S, Wallin MW (2012) Carrots and Rainbows: Motivation and Social Practice in Open Source Software Development. MIS Q 36(2):649\u0026ndash;676\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evon Krogh G, Spaeth S, Lakhani KR (2003) Community, Joining, and Specialization in Open Source Software Innovation: A Case Study. Res Policy 32(7):1217\u0026ndash;1241\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang HM, Yu Y, Wang T, Ding B (2023) Crowd Intelligence Paradigm: the new transformation of software development paradigm. SCIENTIA SINICA Informationis, (53): 1490\u0026ndash;1502. (in Chinese)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang S, Huang X, Zhang M, Bao S, Liu L, Fu X, Guan W (2025) Open science 2.0: revolutionizing spatiotemporal data sharing and collaboration. Comput Urban Sci 5(1):4\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang Z (2024) Governance Mechanisms of Open-source Organizations: Research from the Perspective of Innovation Costs. Tsinghua University, Beijing. (in Chinese)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang Z, Xue L, Zhao J (2024) The evolution of innovation costs and the dynamic governance mechanism of open-source innovation organizations. China Public Adm Rev 6(4):108\u0026ndash;138 (in Chinese)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang Z, Li Z, Yu Y (2024) Do community responses influence OSS contributor retention? A survival analysis. Sci China Inform Sci 67(8):184101\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWeber S (2004) The success of open source. Harvard University Press\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWest J, Gallagher S (2006) Challenges of Open Innovation: The Paradox of Firm Investment in Open-Source. R\u0026amp;D Manage 36(3):319\u0026ndash;331\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWest J, O\u0026lsquo;Mahony S (2008) Contrasting community building in sponsored and community founded open source projects. Proceedings of the 41st Annual Hawaii International Conference on System Sciences (HICSS)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWilliamson OE (1993) Transaction cost economics and organization theory. Ind Corp Change 2(2):107\u0026ndash;156\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWilliamson OE (1996) The mechanisms of governance. Oxford University Press\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWitt MA, Jackson G (2016) Varieties of Capitalism and Institutional Comparative Advantage: A Test and Reinterpretation. J Int Bus Stud 47(5):778\u0026ndash;806\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYin J, Jamali D (2021) Collide or collaborate: The interplay of competing logics and institutional work in cross-sector social partnerships. J Bus Ethics 169(4):673\u0026ndash;694\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYin L, Chakraborty M, Schweik C, Frey S, Filkov V (2022) Open-Source Software Sustainability: Combining Institutional Analysis and Socio-Technical Networks. arXiv preprint arXiv:2203.03144\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"Beihang University","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Open-source AI, Organizations, Governance cost, Innovation outcome, Quantitative, Asian values","lastPublishedDoi":"10.21203/rs.3.rs-8409070/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8409070/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eOpen-source AI innovation has become integral to global AI development, particularly in China, where it enables distributed knowledge creation, collaborative problem-solving, and rapid advancement in technologies like machine learning models, large language models, and ethical AI frameworks. Yet success in open-source AI systems does not arise from openness alone; it rests heavily on governance processes that align heterogeneous actors, address ethical concerns such as data privacy and algorithmic bias, and sustain orderly collaboration amid rapid technological evolution. Although previous research has explored community-based governance in open-source ecosystems, relatively little attention has been given to how formal open-source organizations manage the governance costs that arise as AI projects scale and institutionalize, especially in Asian contexts where cultural values like collectivism and harmony influence collaborative dynamics. This study proposes a governance-cost framework that captures three categories of governance costs - incentive coordination, rule enforcement and value transformation - and examines how these costs mediate the relationship between governance arrangements and AI innovation outcomes. Drawing on a survey of more than 600 key informants from Chinese open-source AI organizations, the study analyzes how organizational structures influence knowledge convergence in AI algorithms, market diffusion of AI applications, and industrial collaboration for ethical AI deployment through the mechanism of governance costs. The findings show that decentralized organizations enhance knowledge convergence by reducing coordination frictions in AI model development, while firm-led and public institution-based organizations demonstrate stronger market diffusion and collaboration performance due to more stable enforcement systems and clearer value-distribution mechanisms that align with Chinese values of collective benefit and societal harmony. Governance costs significantly mediate these relationships, indicating their central role in shaping AI innovation outcomes. By identifying governance costs as foundational mechanisms, this study advances theoretical understanding of open-source AI governance and provides actionable policy guidance for designing cost-efficient, ethically grounded governance in rapidly evolving Asian AI ecosystems, emphasizing the need for policies that integrate Asian policy tendency and cultural values to counterbalance Western-dominated AI frameworks.\u003c/p\u003e","manuscriptTitle":"The Costs and Outcomes of Organizing Open-source AI Innovation: Survey Evidence from China","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-01-06 05:58:51","doi":"10.21203/rs.3.rs-8409070/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"6aed2f7c-63df-4965-92a9-0bdd7bcddc4f","owner":[],"postedDate":"January 6th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":60131873,"name":"Management"},{"id":60131874,"name":"Software Engineering"}],"tags":[],"updatedAt":"2026-01-06T05:58:51+00:00","versionOfRecord":[],"versionCreatedAt":"2026-01-06 05:58:51","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8409070","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8409070","identity":"rs-8409070","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}