Taking Bearings: Artificial Intelligence in Knowledge Platforms and Open, Social Scholarship/Essential Ideas

Executive Summary: A Snapshot of This Work Taking Bearings: Artificial Intelligence in Knowledge Platforms and Open, Social Scholarship is an environmental scan of how AI is reshaping the infrastructures, practices, and institutions through which knowledge is produced, circulated, evaluated, and preserved. Drawing on more than one hundred and fifty sources and informed by earlier work on the theoretical foundations of open, social scholarship, it examines AI across three linked domains—the open, the social, and scholarship—and identifies four recurring tensions around authority, enclosure, capacity, and process. Together, these tensions show that AI is already affecting research, publishing, pedagogy, governance, and public knowledge exchange in ways that demand careful institutional response. The recommendations that follow highlight six areas where organizations can strengthen governance, protect scholarly integrity, and support more accountable and equitable engagement with AI. Summative Overview: A More Detailed Snapshot For organizations and communities engaged in scholarly publishing, research infrastructure, library services, and public knowledge exchange, the rapid integration of artificial intelligence presents both immediate practical challenges and deep structural questions. This volume, Taking Bearings: Artificial Intelligence in Knowledge Platforms and Open, Social Scholarship, synthesizes over one hundred and fifty sources to map this shifting terrain. Grounded in the principles of open, social scholarship, it is intended to clarify what is at stake in the current moment and to support informed, principled institutional responses. The scan proceeds from a foundational observation: AI emerges from a broad, heterogeneous data ecology where scholarly knowledge holds a distinctive, perhaps disproportionately authoritative place. Generative AI and large language models models have been built from vast mixtures of web text, books, code, media, platform content, cultural materials, and other data sources; within that mixture, open scholarship, research data, metadata, citation networks, software, and institutional knowledge infrastructures help supply forms of credibility, structure, and epistemic authority that exceed their share by volume. The relationship between AI and scholarship is therefore fundamentally recursive, not because scholarship alone produced AI, but because AI systems increasingly draw upon, reorganize, and feed back into the same knowledge systems through which scholarship is discovered, evaluated, circulated, and trusted. To map the practical implications of this shift, the scan unpacks the framework of open, social scholarship into its three constituent elements: openness, social engagement, and scholarly practice. Within each area, it examines how AI is changing existing workflows and raising new questions. Looking first at the open, the scan addresses open access, open data, open source, and open science—tracing how AI both depends on and threatens the principles of openness that have shaped scholarly communication over the past two decades. Turning to the social dimensions of this work, it examines platform governance, algorithmic moderation, critical literacies, and questions of coloniality, diversity, and justice—areas where AI's social consequences are most visible and most contested. Finally, regarding scholarship as a formal practice, it addresses research infrastructure, peer review, pedagogy, metadata, and the governance of knowledge institutions—where AI is most rapidly being integrated into professional life. Across all these three domains, four tensions recur with enough consistency to function as the scan's central analytic frame: Operational assistance vs. epistemic authority: While AI can reduce friction in many workflows, it raises fundamental concerns about where judgment, accountability, and interpretive responsibility should reside. Openness vs. enclosure: The open infrastructure that has supported equitable access to knowledge is being absorbed into commercial AI systems in ways that may foreclose openness rather than extend it. Technical capability vs. organizational capacity: The pace of AI development consistently outstrips the ability of institutions to evaluate, govern, and respond thoughtfully. Efficiency vs. process: Although AI accelerates tasks, scholarly knowledge earns its credibility through deliberate, iterative processes of review and community validation that efficiency-oriented tools may erode. These tensions resist neat resolution. What they mark, instead, are sites of genuine institutional choice—places where the decisions organizations make about adoption, governance, disclosure, and investment will have lasting consequences for how knowledge is produced, who benefits from it, and whether its production remains accountable to the communities it serves. The following recommendations translate these findings into six areas where institutional attention is most needed. They are offered as points of orientation rather than prescriptions, in recognition that organizations differ in mission, scale, and resources. The analytical introduction that follows the recommendations provides the conceptual frameworks and intersectional analysis that underpin them. The annotated bibliography, organized by domain and sub-domain, provides the evidentiary foundation in the form of a structured guide to the literature for those who wish to engage with specific areas in depth. Recommendations: Summarized From the Analytical Overview Building from the Executive Summary and Summative Overview, these recommendations serve as practical points of orientation for organizations and communities working to align AI adoption, refusal, adaptation, and design with the principles of open, social scholarship: 1. Establish collaborative and iterative AI governance. AI adoption is fundamentally an organizational and ethical issue, not only a technical one. Decisions about these tools directly affect scholarly labour, institutional authority, access, and trust. Partner organizations should build governance models that draw on both technical and scholarly expertise. Because the technology shifts rapidly, policies and review structures must be designed for continuous evaluation rather than single-point approval. Governance should include explicit procedures for revisiting the boundary between operational assistance and epistemic authority as capabilities, risks, and disciplinary practices change. 2. Preserve human responsibility in scholarly judgment. A central tension identified in this volume is the distinction between operational assistance and epistemic authority. That distinction is necessary, but it is not fixed. AI can support routine tasks, and some tasks that once required substantial expert attention may become increasingly operationalized. Yet the critical judgment required to establish scholarly validity and exercise pedagogical care remains a strictly human responsibility, especially wherever decisions carry intellectual, professional, or public consequences. Institutions should clearly define the limits of appropriate AI use in contexts such as peer review, editorial assessment, pedagogical evaluation, and research interpretation, while also creating mechanisms for revising those limits as practices evolve. 3. Require transparency in AI-mediated workflows. Where AI is used in knowledge production and scholarship, its role must be made visible. The credibility of scholarly and public knowledge depends in part on clear provenance and process, and identifiable responsibility. Organizations should support disclosure standards that make AI use legible without reducing complex workflows to simplistic declarations, and should pair disclosure norms with provenance infrastructure such as persistent identifiers, content credentials, verifiable metadata, version histories, and workflow documentation. Maintaining epistemic integrity requires treating transparency as a basic requirement of scholarly communication. 4. Invest in human capacity and critical literacy. The scan repeatedly demonstrates that the primary challenge facing institutions is ensuring their communities are equipped to evaluate and use AI responsibly, regardless of what tools are available. This requires investment in staff development, pedagogical resources, and the kinds of critical literacy that help scholars and students interpret AI outputs, assess their limits, and recognize how these systems shape knowledge practices. Institutions differ widely in size, missions, and resources, and approaches to AI must be responsive to that unevenness. Capacity-building is therefore one of the preconditions for responsible engagement. 5. Protect and steward open knowledge from extractive enclosure. Contemporary AI systems rely heavily on the scholarly commons and the unpaid labour of knowledge communities, even as the benefits of these systems are often unevenly distributed. Organizations committed to open, social scholarship should actively resist the ways AI may enclose, concentrate power, or extract value from shared resources without meaningful forms of reciprocity. Governance frameworks must therefore prioritize equitable benefit-sharing alongside licensing, consent, attribution, provenance, and machine-readable terms of use. Responsible participation in AI ecosystems requires attention to the conditions under which tools operate and the interests they serve, as much as to their capabilities. 6. Build inclusive processes for ongoing evaluation and response. The technological landscape is shifting rapidly, and no single framework remains adequate for long. Institutions need mechanisms for periodic reassessment, community consultation, and iterative revision in response to the new domains and effects of changing technologies. Particular attention should be paid to whose perspectives are included in such processes, especially across differences of language, geography, discipline, institutional position, and ability. The issues identified here—including disclosure, hallucination, metadata generation, peer review, ethics, and infrastructure—point toward the need for sustained, focused follow-up work rather than one-time resolution. Promising areas for shorter, targeted interventions included AI-mediated scholarly discovery, provenance and content-credential infrastructure, community-built or purpose-specific models, integrity threats such as paper mills and synthetic citations, AI in hiring, promotion, grants, and fellowships, AI and open licensing, and AI's effects on open scholarly infrastructure. Introduction