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

Histories & Theories of AI Ali, Syed Mustafa, Stephanie Dick, Sarah Dillon, Matthew L. Jones, Jonnie Penn, and Richard Staley. 2023. “Histories of Artificial Intelligence: A Genealogy of Power.” BJHS Themes 8: 1–18. https://doi.org/10.1017/bjt.2023.15. The work of these authors seeks to move beyond “conventional origin myths” of AI by contextualizing the academic historical approach within their interdisciplinary backgrounds. As a result, Ali et al. argue that histories of broader processes (like industrialization, colonialism, and social science) can represent the “genealogy” of AI, which is not identified as a specific object but as a grouping of diverse technologies under a loose banner, requiring an equally diverse approach. Although AI is the “flagship of the Information Age,” it was clearly conditioned by the “Management Age,” as these “genealogies” feature four common thematic threads of “hidden labour,” “encoded behaviour,” “disingenuous rhetoric,” and “cognitive injustice.” AI is therefore not only situated within the history of computing, but the history of control, the product of the interacting formalist and empiricist views of “intelligence” within Cold War epistemology. In many ways, AI’s refinement of formal abstraction reinforces and impresses Western systems and structures, so these authors do not critique AI with the goal of improvement, but to clarify what AI “is,” “is not,” and perhaps “should not” be. Anderson, Marc M. 2024. “AI as Philosophical Ideology: A Critical Look Back at John McCarthy’s Program.” Philosophy & Technology 37 (2): 44. https://doi.org/10.1007/s13347-024-00731-1. This paper critically analyzes the synthesis of McCarthy’s AI development program, identifying its goals’ nature, purpose, relationship to society, and implications upon morality and social control. Firstly, Anderson considers the most problematic themes of McCarthy’s early and middle writings, then arguing that McCarthy’s later works consolidated his stance by replacing “tentative assumptions” with “statements whose linkages with contemporaneous analytic philosophy are made unequivocally” and thereby present AI as “the outcome of a certain philosophical way of looking at the world.” Here, Anderson emphasizes McCarthy’s biases toward objectivism, scientism, and analytic behaviourism; tendency to enable progress in AI by waiting to identify risks; positioning of philosophy as subordinate “handmaiden” to science; and uneasy presupposition that humans have innate knowledge of the world’s object character (which AI will need added). To Anderson, the sum of the “ethical” aspects, “philosophical” aspects, and “competitive” nature of McCarthy’s program give it all the fundamental “characteristics of an ideological program.” This ideology has had deep direct influence, for example by joining AI ethics to consequentialist dilemmas like the “Trolley problem,” and deep indirect influence, for example by influencing a modular and component technical approach. Anderson argues this “ideology” was caused by McCarthy’s motive of making a “servile” and “stripped down model of human mental engagement without emotional qualities.” Overall, Anderson argues McCarthy’s conception of AI as a “perfectly rational abstraction of human thinking” involves severe misconceptions around rationality, and has “come to serve as a blind for increasing techno-corporate control of society” with its propagation of the belief that “AI systems should be controlled as servants.” Bender, Emily M., Timnit Gebru, Angelina McMillan-Major, and Shmargaret Shmitchell. 2021. “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜.” In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, 610–23. https://doi.org/10.1145/3442188.3445922. Bender et al. critically analyze the environmental and social risks of ever larger language models trained with huge uncurated datasets from the web. Their first concern is the environmental costs of training language models, which require increasing energy and compute requirements that further environmental damages that are more likely to fall on marginalized populations. Their social concerns are that language models have the potential to reproduce hegemonic views, reinforce stereotypes, and further reify inequality due to the unrepresentativeness of their training data. Despite the increasing amounts of data ingested in these models, the authors argue that size does not guarantee diversity because there are people who are not on the web (and therefore not included in training data), the content of marginalized people online is less likely to be included in these large datasets and, if it is, it might be filtered out during the data preparation process. Therefore, the authors urge developers to stop using larger training datasets if they cannot be documented. As an alternative, they suggest curating and documenting smaller datasets created for specific purposes, evaluating models by the amount of resources they consume, and developing research that centres the people who are more likely to be adversely affected by the resulting technology. Berry, David M. 2025. “Synthetic Media and Computational Capitalism: Towards a Critical Theory of Artificial Intelligence.” AI & Society 40: 5257–5269. https://doi.org/10.1007/s00146-025-02265-2. Berry develops a critical theory of artificial intelligence centred on what he terms the "algorithmic condition"—a historical moment when computational systems generate cultural content indistinguishable from human production, thereby destabilizing traditional markers of authenticity and authorship. Engaging with Frankfurt School critical theory, particularly concepts of false consciousness and reification, Berry argues that contemporary AI represents a qualitative transformation he calls "the Inversion," where machine-generated works not only replicate but actively reshape the grounds upon which experience and meaning are constituted. This moves beyond mechanical reproduction (Benjamin) or cybernetic feedback (Wiener) toward what Berry theorizes as "post-consciousness," where boundaries between individual and synthetic consciousness become porous. His concept of "diffusionisation" describes how AI systems dissolve cultural forms into probabilistic vector spaces and reconstitute them through latent space manipulation, producing "AI slop"—low-quality synthetic content that infiltrates information ecosystems with garbled meaning. Berry positions this transformation within "computational capitalism," arguing that algorithmic mediation now structures forms of life in ways that demand new critical methods. His "constellational analysis" proposes mapping the interdependencies among technical systems, cultural production, and political-economic structures to resist subsumption into algorithmic logics. Where earlier automation debates focused on task substitution, Berry foregrounds how synthetic media transforms the infrastructure of meaning-making itself, rendering questions of interpretability and embedded bias not as technical problems but as symptoms of deeper epistemic and political reconfigurations under computational conditions. Chun, Jon, and Katherine Elkins. (2023). “The Crisis of Artificial Intelligence: A New Digital Humanities Curriculum for Human-Centred AI.” International Journal of Humanities and Arts Computing 17 (2). https://doi.org/10.3366/ijhac.2023.0310. Chun and Elkins position AI's rapid advancement as precipitating multiple interlocking crises—in higher education, in diversity and inclusion within technology fields, and in the broader socioeconomic fabric—that demand a specifically humanities-oriented pedagogical response. Their intervention challenges the prevalent assumption that AI literacy should emerge from STEM-dominated curricula, arguing instead that Digital Humanities offers distinctive pathways for cultivating critical engagement with computational systems. Where conventional computer science education treats AI as primarily a technical phenomenon requiring engineering skills, Chun and Elkins foreground reflective and collaborative meaning-making, insisting that the most effective AI tools amplify rather than replace human intellectual engagement. Their framework explicitly connects pedagogical design to civic preparation, positioning students not as mere consumers or operators of AI systems but as critically engaged citizens capable of interrogating the social and economic implications of algorithmic mediation. Drawing on theories of human-centred computing, they articulate an AI DH curriculum structured around two core commitments: opening meaningful research avenues for humanities scholars working with computational methods and addressing DEI shortcomings by engaging students traditionally alienated by conventional STEM pathways. Their approach implicitly challenges the epistemic authority claims embedded in AI discourse—the fantasy that technical expertise alone can adjudicate questions about algorithmic deployment—by insisting that humanistic inquiry offers essential resources for navigating AI's transformative effects on knowledge production, labour organization, and cultural representation. Chung, Hiu-Fung. 2025. “Betting on (Un)Certain Futures: Sociotechnical Imaginaries of AI and Varieties of Techno-Developmentalism in Asia.” Information, Communication & Society, 1–18. https://doi.org/10.1080/1369118X.2025.2535427. Chung redirects scholarly attention from dominant AI powers—China, the United States—to economically advanced but geographically non-dominant Asian societies: Singapore, Hong Kong, and Taiwan. Mobilizing the concept of "techno-developmentalism" to analyze how developmental states harness technological innovation for political-economic projects, Chung identifies three distinct imaginaries that emerge from specific historical, institutional, and geopolitical conditions. Singapore's "cybernetic pragmatism" deploys AI to legitimize neoliberal authoritarianism, embedding computational governance as a continuation of technocratic rule. Hong Kong's "techno-entrepreneurship" imaginary seeks to refashion financial capitalism through AI-driven innovation, positioning the territory as a global fintech hub amid shifting relationships with mainland China. Taiwan's "defensive survival modality" frames AI development as simultaneously addressing internal socioeconomic instability and external threats from superpower rivalry, particularly cross-strait tensions. Chung's analysis challenges the Global North/South binary that structures much AI governance literature, revealing how small advanced economies navigate strategic coupling with global tech industries while managing profound uncertainties about AI-centric reforms. His discourse analysis of policy documents from the early 2010s through 2024 demonstrates how national imaginaries encode assumptions about automation's effects on labour markets, the appropriate balance between state coordination and market mechanisms, and the role of computational infrastructure in securing geopolitical position. This comparative framework illuminates how different state formations mobilize AI to address divergent crises of legitimacy, capital accumulation, and sovereign security. Van Es, Karin, and Dennis Nguyen. 2024. “‘Your Friendly AI Assistant’: The Anthropomorphic Self-Representations of ChatGPT and Its Implications for Imagining AI.” AI & Society 40: 3591–3603. https://doi.org/10.1007/s00146-024-02108-6. Van Es and Nguyen analyze how “social-technical imaginaries” invoked by “self-representations” of GenAI could influence public perception of the technology. These “socio-technical imaginaries” are a category of trending concept, which can exist in contrast, as both utopian and dystopian conceptions of AI are present in public discourse. These are dynamically changeable, often strategically promoted by tech companies to influence regulation, and influenced by a variety of media technologies and ecologies with equally diverse social, cultural, and political implications. The authors identify a range of current “socio-technical imaginaries” of AI. This includes: misconceived “magical thinking,” largely produced by poor terminology; speculative, exaggerative debates on AI capabilities, which distract from actualized and present risks; visual motifs correlating intelligence, efficiency, logic, duty, and trust, which are not inherent to Generative AI; and anthropomorphism, which brings a slew of problems surrounding social biases. To study AI “self-representations,” van Es and Nguyen have ChatGPT generate fifty images and fifty-eight sections of text, responding to several variations of the prompt “create an image of yourself.” The authors then perform an empirical, qualitative-exploratory analysis of this content to examine which “socio-technical imaginaries” dominate the material. These images most significantly emphasized ChatGPT’s alleged “social intelligence,” depicted as a friendly research assistant possessing “real” intellect—ironically often surrounded by books. Anthropomorphism was usually present, with eighty-two percent of images portraying AI as humanoid, and a further six percent as a human brain. Futuristic motifs like holograms, metallic shades, and the cosmos portrayed AI as near-magical. Textual responses expressed these same themes, demonstrating consistent messaging. Taken together, these generated responses encourage overlooking the ethical and legal challenges posed by trending GenAI technologies, “overestimate their capabilities”, and “potentially lead to mistakenly perceive them as trustworthy companions.” This raises questions as to what extent specific guidelines to these responses were programmed by OpenAI, or indicative of popular opinion. Jones, Matthew L. 2023. “AI in History.” American Historical Review 128 (3): 1360–67. https://doi.org/10.1093/ahr/rhad361. Jones explores how evolving approaches to AI research reflect upon the importance of studying artificial intelligence’s history. He believes traditional attempts to produce “symbolic” AI have been proven as misguided, comparing their failures to the traditionally unpopular “empiricist” approach. This approach is now dominant and involves the use of large-scale algorithms to capitalize upon the “unreasonable effectiveness of data” via “machine learning.” However, this shift in trends has also forced the meaning and capabilities of “artificial intelligence” to be reconsidered, now threatening “professions centred on particularity” like history. In response, Jones argues modern AI cannot be considered a “neutral substratum” but as a reflection of human creations—including their best and worst traits. Historians are experts at detecting bias, and this understanding of AI’s historical context is clearly necessary for its appropriate application, so historians should be playing a fundamental role in ensuring AI’s critical use. Jones believes historians can best do this by highlighting the “traces of labour” that can illustrate the “granular complex reality” of AI. Klein, Lauren, Meredith Martin, André Brock, Maria Antoniak, Melanie Walsh, Jessica Marie Johnson, Lauren Tilton, and David Mimno. 2025. “Provocations from the Humanities for Generative AI Research.” Preprint, arXiv, February 26. https://arxiv.org/abs/2502.19190. Klein et al. explore the stakes of engaging with AI by juxtaposing the humanistic scholar’s investigation of the "human" with the "anti-human" approach of AI, a dichotomy they contextualize within late-stage capitalism. Klein et al. argue that AI’s generalist, binary approach "distorts 'culture’ to content." AI programs thereby conduct a "statistical enactment of… ideology," imposing an objectivist European modernist framework of understanding upon the data. This binary conception of data is highlighted within ideas of "pure" versus "toxic" training data, which dismiss that the material inherently "reflects cultures and consists of expressions of those cultures." Even content not being included "biases" the rest of the data, so the best practices of open scholarship cannot fix the root issue—the only way to resolve the issues caused by "bias" is to target the sources of the underlying structural problems. Klein et al. argue this flawed situation was fuelled by "corporate spokespeople parroting AI hype" to research—and that this relationship’s politicization should be a particular source of concern. Klein et al. argue AI can still serve a purpose, however. They propose AI with smaller datasets will produce more accurate information when tailored for specific topics and with input from experts of the humanities. Humanists cannot solely effect change, however, as "technical researchers must recognize the institutional asymmetries" of administrative support and funding available to ensure development of "humanistic" AI is a truly equal and interdisciplinary effort. McIntosh, Timothy R., Teo Susnjak, Tong Liu, Paul Watters, and Malka N. Halgamuge. 2023. “From Google Gemini to OpenAI Q* (Q-Star): A Survey of Reshaping the Generative Artificial Intelligence (AI) Research Landscape.” Technologies. https://doi.org/10.3390/technologies13020051. McIntosh et al. offer a survey of AI development’s frontiers targeted at fellow experts of the field, focusing on evolving “agentic” AI and its resulting research applications. Covering developments, they primarily employ case studies to demonstrate how multimodal AI could eclipse undynamic LLMs, but various “advanced learning techniques” and developments in model architecture are also outlined. On this topic, a table is presented to quantify the relevance of research fields and subfields of AI development. Agentic AI’s possible applications include aiding research by bolstering academic integrity, its market relevance, and its “creative” purposes. McIntosh et al. even argue that development of AGI with “symbolic reasoning” and “probabilistic inference” could handle issues like climate change, considered as a long-term but significant possibility. Therefore, this work is primarily theoretical, with McIntosh et al. centralizing AI’s idealistic applications. However, they do acknowledge AI must be developed with ethical and social principles, for which interdisciplinary cooperation is necessary. Mao, Yishu, Vanessa Richter, and Christian Katzenbach. (2025). "Strategising imaginaries: How Corporate Actors in China, Germany and the US Shape AI Governance." Big Data & Society 12 (1). https://doi.org/10.1177/20539517251400727. Mao, Richter, and Katzenbach systematize the concept of "sociotechnical imaginaries" as an analytical framework for understanding how collective visions of AI futures shape governance, investment, and public discourse. Engaging explicitly with science and technology studies scholarship—particularly Jasanoff and Kim's formulation of imaginaries as "collectively held, institutionally stabilized, and publicly performed visions"—they demonstrate how AI's meaning emerges through contested negotiations among stakeholders from industry, government, academia, media, and civil society. Their comparative analysis across the United States, China, and Germany challenges simplistic national characterizations, revealing heterogeneous and often contradictory imaginaries even within single regulatory regimes. Where US discourse fragments across multiple geographic AI hubs, German imaginaries centre on EU policy compliance and regulatory frameworks, while Chinese articulations align tightly with party-state directives that minimize local variation. The source argues that these imaginaries function performatively, not merely representing AI's future but actively mobilizing resources, legitimating interventions, and establishing trajectories for development. Their framework illuminates how benchmark cultures, optimization narratives, and investor expectations become embedded in technical choices through discursive processes that precede and exceed engineering decisions. By foregrounding stakeholder co-dependencies and cross-national dynamics, the authors position AI imaginaries as sites where epistemological assumptions about intelligence, automation, and progress become institutionalized through political-economic mechanisms—a contribution that connects discourse analysis to questions of power, accountability, and the material reorganization of labour and knowledge production. Whiteley, Paul. 2023. “Why Artificial Intelligence Is a Misnomer.” London School of Economics and Political Science Politics and Policy Blog, October 19. https://blogs.lse.ac.uk/politicsandpolicy/why-artificial-intelligence-is-a-misnomer/. Whiteley criticizes the use of the term artificial intelligence, noting that while such technologies can be valuable for automating specific tasks, the term John McCarthy used in the 1950s was a “disservice.” He argues that the term anthropomorphizes what is essentially a form of computer-assisted statistical analysis. Much of what we call AI functions as a sophisticated prediction system that processes vast amounts of data to produce answers, without any real understanding of the underlying concepts or theories. Whiteley further distinguishes between algorithms, which follow set procedures to identify patterns and make predictions, and inference, which involves interpreting and explaining why those patterns occur. He concludes that the ultimate goal in the field of AI is to create systems capable of predicting behaviour across diverse problems. However, he emphasizes that current AI algorithms remain highly specialized, excelling in narrow tasks but lacking the flexibility and understanding required for true general intelligence. Zeffiro, Andrea. 2024. “Automating (In)securities: Cybersecurity’s AI Imaginaries.” Paper presented at EASST/4S, 2024. https://nomadit.co.uk/conference/easst-4s2024/p/14156. Zeffiro interrogates cybersecurity's AI imaginaries through case studies of IBM Watson for Cybersecurity, CrowdStrike's Charlotte AI, and Google's Sec-PaLM, revealing how corporate narratives of AI as "game changer" and "democratizing force" systematically obscure differential vulnerabilities and normative biases. Her analysis challenges the prevalent framing of AI-driven threat detection as technically neutral automation, arguing instead that these systems encode and amplify assumptions about what constitutes risk, who merits protection, and whose insecurities remain invisible. Drawing on critical security studies and science and technology studies, Zeffiro demonstrates how generative AI applications in cybersecurity automate not merely data collection and pattern recognition but also the normative judgments about threat hierarchies embedded in training regimes and performance metrics. Her concept of "automating insecurities" captures this dual process: while AI tools promise enhanced security through real-time threat response with minimal human intervention, they simultaneously institutionalize particular understandings of vulnerability that reflect corporate priorities and market positioning in the "AI arms race." Zeffiro's analysis connects these imaginaries to epistemic claims about AI's inevitability, showing how IBM, CrowdStrike, and Google construct future visions that naturalize their technological approaches while marginalizing alternative security paradigms. By foregrounding what these imaginaries omit—the differential distribution of cyber vulnerabilities across social positions, the political economy of security infrastructure, the discretionary authority embedded in algorithmic triage—Zeffiro positions cybersecurity AI as participating in broader patterns of automation that redistribute power and accountability rather than merely enhancing technical capabilities. Past Relation to OSS-Aligned Communities Carnegie Endowment for International Peace. 2024. Beyond Open vs. Closed: Emerging Consensus and Key Questions for Foundation AI Model Governance. https://carnegieendowment.org/research/2024/07/beyond-open-vs-closed-emerging-consensus-and-key-questions-for-foundation-ai-model-governance?lang=en. The report dismantles the binary framing—open vs. closed—that has organized foundation model governance debates, replacing it with a multidimensional spectrum where weight release is only one variable among many (architecture, training data, documentation, licensing terms, structured access). Seven consensus points establish that "openness" serves multiple, sometimes conflicting values (transparency, competition, safety, inclusion) and that weight release amplifies both beneficial and harmful potential without symmetry—its irreversibility and resistance to post-release monitoring create a qualitatively different governance problem than closed deployment, even though closed models' theoretical safety advantages are unevenly realized in practice. The report introduces "precautionary friction" (staged/structured release calibrated to marginal risk over a defined baseline) as the operative governance concept, explicitly rejecting both blanket openness and blanket restriction. Seventeen open questions then expose the infrastructure gaps: evaluation science remains embryonic, post-release monitoring is under-theorized, risk thresholds published by labs lack enforcement specificity, and Global South labour and data contributions are structurally undervalued. This source clarifies the claim that software-style "open source" maps poorly onto foundation models and that hybrid, graduated-release frameworks are where policy consensus is actually forming. Mitra, Bhaskar, Henriette Cramer, and Olya Gurevich. 2024. “Sociotechnical Implications of Generative Artificial Intelligence for Information Access.” Preprint, arXiv, May 19. https://doi.org/10.48550/arXiv.2405.11612. Bhaskar et al. diagnose “information access” under generative AI as a sociotechnical settlement rather than a neutral upgrade to search: models do not just retrieve, they intermediate credibility, reshape attention, and re-allocate epistemic authority through interfaces that compress provenance and uncertainty. Their framing is especially useful for open, social scholarship because it links familiar problems of access (indexing, ranking, evaluation) to newer constraints created by platform dependence, opaque model behaviour, and the enclosure of research pathways behind proprietary tooling and data. Read through the lenses of industry capture and AI imaginaries, the chapter clarifies how promises of universal assistants can normalize monopoly infrastructures while shifting audit burdens onto users and public institutions. It also implicitly strengthens open communities’ insistence on reproducibility: without open benchmarks, inspectable logs, and preservable corpora, “trustworthy information” becomes a brand claim rather than an evaluable property. The ideas bridge directly to libraries, repositories, and information policy by treating curation, metadata, and stewardship as counter-monopoly infrastructure for accountable AI-mediated access. Open Source Initiative. The Open Source AI Definition 1.0 (2024–2025). https://opensource.org/ai OSI’s Open Source AI Definition 1.0 operationalizes “open” as a bundle of enforceable freedoms rather than a branding claim, and it does so by porting the open source software tradition into the AI stack where “source” is more than just code. The definition makes the four freedoms—use, study, modify, share—the evaluative core, then adds a crucial precondition: exercising any of these rights requires access to the preferred form for making modifications and the means to use the system. That move is the governance lever: it blocks “open weights” releases from standing in for openness when the artifacts needed to understand or change system behaviour are withheld. On OSI’s framing, openness is not satisfied by permissive inference access or a model file alone; it is satisfied when the system is provisioned so that auditability and alteration are practically possible, including transparency sufficient to trace how results were created and where components (notably data sources) come from. The definition is therefore designed as an anti-openwashing instrument: it supplies a community standard that can be applied to legal/technical packaging to distinguish genuinely open AI systems from hybrid offerings that preserve vendor control through restrictions or missing components. OSI explicitly positions this as policy-relevant infrastructure—an interpretive anchor for regulators and OSS communities trying to resist enclosure pressures while preserving permissionless collaboration. This is a community standard, not a journal article; it is nonetheless important for governance, policy, and OSS alignment. Vake, Domen, Bogdan Šinik, Jernej Vičič, and Aleksandar Tošić. 2025. “Is Open Source the Future of AI? A Data‑Driven Approach.” Preprint, arXiv, January 27. https://arxiv.org/abs/2501.16403. Vake et al. analyze the data of open-source large language models shared on HuggingFace to explore if the open-source community influences the development of LLMs. The authors found that the AI open-source community is expanding rapidly, and it has enhanced the performance of a handful of popular models. However, open-source AI depends on businesses to develop base models and release them openly, but there are few incentives to do so because it risks their intellectual property and competitive edge. Furthermore, AI is different from the development of open-source software because the general public cannot privately run these models. Therefore, the authors conclude that the future of AI development could be similar to the software-as-a-service model, in which the open-source community contributes to model development, while companies generate revenue from model usage. White, Matt, Ibrahim Haddad, Cailean Osborne, Xiao-Yang Liu, Ahmed Abdelmonsef, Sachin Varghese, and Arnaud Le Hors. 2024. “The Model Openness Framework: Promoting Completeness and Openness for Reproducibility, Transparency, and Usability in AI.” Preprint, arXiv, March 20. https://doi.org/10.48550/arXiv.2403.13784. White et al.'s Model Openness Framework (MOF) represents an effort to translate open-source software principles into AI research practices, addressing the growing gap between claims of "openness" and actual transparency in AI model development. Developed through the Linux Foundation's AI & Data Foundation, the framework establishes a three-tiered classification system that specifies which components (code, data, documentation, trained weights, evaluation procedures) must be released under open licenses for models to qualify as "Class III: Open Model," "Class II: Open Tooling," or "Class I: Open Science." The authors' intervention addresses what they term "openwashing"—the strategic use of "open source" rhetoric by companies releasing models with restrictive licenses or incomplete artifacts. By codifying 17 critical components for complete model releases, they make visible the specific practices required for reproducibility and scrutiny, challenging the binary conception of openness inherited from software. Their framework reveals that AI "openness" exists on a spectrum of completeness, from minimal weight release to full disclosure of training data, intermediate checkpoints, and development documentation. Positioned within debates about responsible AI development, the MOF represents a community-driven effort to establish norms before they become ossified by corporate practice or regulatory fiat. The authors draw explicitly on open science principles (FAIR data, reproducibility standards) while adapting them to AI's unique characteristics—particularly the centrality of training data and the distinction between model architecture (code) and trained parameters (data). Bias and Technological Determinism Akter, Shahriar, Grace McCarthy, Shahriar Sajib, Katina Michael, Yogesh K. Dwivedi, John D’Ambra, and K. N. Shen. 2021. “Algorithmic Bias in Data-Driven Innovation in the Age of AI.” International Journal of Information Management 60: 102387. https://doi.org/10.1016/j.ijinfomgt.2021.102387. Akter and colleagues theorize algorithmic bias as emerging from three distinct but interacting sources—data bias, method bias, and societal bias— positioning bias as structural phenomenon inscribed across the entire data-driven innovation (DDI) lifecycle rather than localized technical artifact amenable to isolated correction. Their case study of Australia's Robo-Debt scheme demonstrates how algorithmic systems inherit and amplify existing inequities and foregrounds "dynamic managerial capability" as essential for addressing bias. The authors challenge purely technical approaches, arguing instead that organizational capacity to recognize, interrogate, and respond to bias across data provenance, algorithmic design, and deployment contexts determines whether DDI produces equitable or discriminatory outcomes. This positions bias mitigation as ongoing institutional work requiring cross-functional expertise rather than one-time technical intervention. Their framework implicitly contests technological determinism by revealing how managerial choices about data collection priorities, acceptable error rates, and stakeholder consultation shape algorithmic outcomes in ways that exceed engineering decisions. The emphasis on societal bias as distinct category acknowledges that algorithms operate within and reproduce broader patterns of structural inequality—assumptions about creditworthiness, employability, or welfare eligibility that reflect historical discrimination. Akter et al.'s intervention thus connects bias scholarship to organizational studies and innovation management, positioning algorithmic fairness not as mathematical property but as emergent from institutional practices, power relations, and the political-economic contexts within which DDI unfolds. Jarrahi, Mohammad Hossein, Gemma Newlands, Min Kyung Lee, Christine T. Wolf, Eliscia Kinder, and Will Sutherland. 2021. “Algorithmic Management in a Work Context.” Big Data & Society 8 (2). https://doi.org/10.1177/20539517211020332. Analyzes algorithmic management as a sociotechnical phenomenon reshaping power, discretion, and information flows in organizations; details opacity at technical and organizational levels. Exposes determinist narratives in “smart” automation of management decisions; shows how design choices redistribute authority and labour—central to a nondeterministic, governance-forward stance. Sartori, L., and A. Theodorou. 2022. “A sociotechnical perspective for the future of AI: narratives, inequalities, and human control.” Ethics and Information Technology 24 (1), 4. https://doi.org/10.1007/s10676-022-09624-3. Sartori and Theodorou position AI as a "magnifying glass" that automates and amplifies existing social inequalities rather than introducing bias as a novel technical problem, directly challenging narratives that frame fairness as achievable through post hoc algorithmic adjustment. Their sociotechnical framing insists that intelligent machines operate within specific institutional contexts where historical patterns of discrimination become encoded in data collection, annotation practices, and deployment decisions—a structural account that resists reduction to technical fixes. The authors engage critically with the AI technical community's emphasis on transparency, explainability, accountability, and contestability, acknowledging these as necessary but insufficient responses that risk becoming "panaceas" if divorced from attention to power asymmetries and organizing visions. Their analysis of technological narratives reveals how AI discourse reflects and reproduces traditional lines of social, economic, and political inequality: who gets to articulate AI futures, whose concerns register as legitimate risks, and which imaginaries gain institutional traction. By foregrounding narratives as both reflecting organizing visions and constituting "tangible signs" of inequality, Sartori and Theodorou demonstrate how deterministic framing—treating AI development as inevitable—forecloses deliberation about alternative design pathways. Their call for "diverse approaches" and "richer knowledge about narratives" positions non-determinism as methodological commitment: attending to contingent choices, plural stakeholder perspectives, and varied criteria of success that resist singular optimization logics. This intervention connects bias scholarship to governance debates, arguing that human control requires not just technical interpretability but institutional mechanisms for contestation that acknowledge AI practice as fundamentally interdisciplinary and politically consequential. Knowledge Foundations Diversity Arbuckle, Alyssa. 2020. “How Can We Broaden and Diversify Humanities Knowledge Translation?” Pop! Public. Open. Participatory, no. 2. https://doi.org/10.48404/pop.2020.12. Articulates diversity as a condition for valid public-facing scholarship; emphasizes inclusive modes of knowledge translation and participatory approaches resonant with OSS. Arthur, Paul L., Lyida Hearn, Lucy Montgomery, Hugh Craig, Alyssa Arbuckle, and Ray Siemens. (2021). “Open Scholarship in Australia: A Review of Needs, Barriers, and Opportunities.” Digital Scholarship in the Humanities 36 (4), 795–812. https://doi.org/10.1093/llc/fqaa063. Australian context scan identifies structural barriers and opportunities for diverse participation in open scholarship. Crompton, Constance, Lori Antranikan, Ruth Truong, and Paige Maskell. 2020. “Familiar Wikidata: The Case for Building a Data Source We Can Trust.” Pop! Public. Open. Participatory, no. 2. https://doi.org/10.48404/pop.2020.02. Makes the case for community governance and trust in open data infrastructures to safeguard plural representation and mitigate erasure in knowledge graphs. Fanning, Katie, Claire Kim, and Jon Saklofske. 2023. “Interactive Inspirations: The Case for Incorporating Joy and Play in Open Social Scholarship.” Pop! Public. Open. Participatory, no. 5. https://doi.org/10.54590/pop.2023.012. Argues that Open Social Scholarship must move beyond simple open access by cultivating "joy and play" as critical, anti-capitalist methodologies for community engagement. The authors demonstrate this through a prototype parody app and use its practical limitations to expose the structural friction between sustaining non-extractive digital commons and the realities of privatized app ecosystems and precarious academic funding. Rockwell, Geoffrey, Kaylin Land, and Andrew MacDonald. 2021. “Social Analytics Through Spyral.” Pop! Public. Open. Participatory, no. 3. https://doi.org/10.54590/pop.2021.004. Demonstrates community-aware analytics for scholarly communities; touches on representational choices and the risks of flattening diverse practices through metrics. Siemens, Lynne, and the INKE Research Group (2023). “I Stayed for the Community: Collaboration and Community in an Open Social Scholarship Research Project.” Pop! Public. Open. Participatory, no. 5. https://doi.org/10.54590/pop.2023.013. Views the success of open social scholarship as depending heavily on "heritage relationships" and the invisible labour of interpersonal maintenance, treating cross-sector collaboration as a deliberate methodological challenge rather than a natural byproduct of funding. Highlights a structural tension between community-driven qualitative goals and rigid academic metrics, urging institutions to formally recognize and resource the relational infrastructure that sustains Digital Humanities projects. Mobilisation Arbuckle, Alyssa, Ray Siemens, Jon Bath, Constance Crompton, Laura Estill, Tanja Niemann, Jon Saklofske, and Lynne Siemens. 2022. “An Open Social Scholarship Path for the Humanities.” The Journal of Electronic Publishing 25 (2). https://doi.org/10.3998/jep.1973. Defines OSS practices and infrastructures to support translation, engagement, and reciprocal exchange between researchers and publics. Jensen, Graham. 2023. “Introduction: Digital Knowledge Commons, Scholarly Connection, and the Evolution of Open Scholarship.” Open Scholarship Press Curated Volumes: Connection. https://doi.org/10.21428/47bc126e.0ca461a4. Connects social scholarly tools and commons-based infrastructures to mobilization pathways; highlights interoperability and community practices. Nelson, Brent, Miguel Dela Pena, the Prototyping the Digital Archive Team, and the INKE Research Group. 2023. “No Journal is an Island: The John Donne Journal and the Possibilities of Open Access.” Pop! Public. Open. Participatory, no. 5. https://doi.org/10.54590/pop.2023.007. Case study of OA pathways as mobilization; addresses circulation, visibility, and engagement with broader publics. Winter, Caroline. 2023. “Introduction: Open Scholarship Policy in Focus.” Open Scholarship Press Curated Volumes: Policy. https://doi.org/10.21428/47bc126e.5abba88b. Frames policy as a vehicle for knowledge mobilization across sectors; emphasizes provenance, accountability, and public value alignment. Winter, Caroline, Tyler Fontenot, Luis Meneses, Alyssa Arbuckle, Ray Siemens, and the ETCL and INKE Research Groups. 2020. “Foundations for the Canadian HSS Commons: Exploring the Possibilities of Digital Research Communities.” Pop! Public. Open. Participatory, no. 2. https://doi.org/10.48404/pop.2020.05. Maps HSS Commons as mobilization infrastructure; addresses governance mechanisms, provenance, and dialogic engagement in a Canadian context. Platforms Bullard, Julia. 2023. “Describing the HSS Commons: The View from Metadata.” Pop! Public. Open. Participatory, no. 5. https://doi.org/10.54590/pop.2023.004. Metadata as platform backbone; shows how descriptive schemas shape visibility/legitimacy; connects to interoperability and inclusive description. Goddard, Lisa. (2021). “Persistent Identifiers as Open Research Infrastructure to Reduce Administrative Burden.” Pop! Public. Open. Participatory, no. 3. https://doi.org/10.54590/pop.2021.006. PIDs as platform connective tissue; crucial for provenance, traceability, and platform interoperability across repositories/journals. Jensen, Graham, Alyssa Arbuckle, Caroline Winter, Talya Jesperson, Tyler Fontenot, Ray Siemens, and the ETCL and INKE Research Groups. 2022. “Fostering Digital Communities of Care: Safety, Security, and Trust in the Canadian HSS Commons.” IDEAH 3 (2). https://ideah.pubpub.org/pub/h7927ugt. Governance and moderation as platform design commitments; aligns with OSS values around safety and reciprocity. Meneses, Luis. 2020. “Integrating the Social Media Engine with Large-scale Open Access Repositories: A Discussion.” Pop! Public. Open. Participatory, no. 2. https://doi.org/10.48404/pop.2020.04. Explores platform coupling between OA repositories and social layers; anticipates AI-mediated discovery/recommendation impacts on visibility. Turin, Mark. 2021. “From Orality to Open: Innovations in Multimedia Monograph Publishing in the Humanities.” Pop! Public. Open. Participatory, no. 3. https://doi.org/10.54590/pop.2021.005. Platform affordances for multimodal scholarship; how platform standards and workflows enable plural forms and publics. Open Social Scholarship Arbuckle, Alyssa, and John Maxwell. (2019). “Modelling Open Social Scholarship Within the INKE Community.” KULA: Knowledge Creation, Dissemination, and Preservation Studies 3 (1), 1–8. https://doi.org/10.5334/kula.15. Moves “beyond open access” toward OSS practice by reworking scholarly communication workflows around collaboration, transparency, and community needs. Emphasizes infrastructure choices and governance as sites where values are enacted. Gives concrete criteria for participatory infrastructures and power‑sharing in scholarly communication; helpful to evaluate platform design choices. Arbuckle, Alyssa, Ray Siemens, Jon Bath, Constance Crompton, Laura Estill, Tanja Niemann, Jon Saklofske, and Lynne Siemens. 2022. “An Open Social Scholarship Path for the Humanities.” Journal of Electronic Publishing 25 (2). https://doi.org/10.3998/jep.1973. A concise, peer‑reviewed articulation of open social scholarship (OSS) from INKE/ETCL leaders, translating principles into programs of action: public engagement, community training, policy, and infrastructure. Provides a normative baseline to assess whether platforms and practices foster dialogic exchange and mutuality rather than extraction. Arthur, Paul L., Lyida Hearn, Lucy Montgomery, Hugh Craig, Alyssa Arbuckle, and Ray Siemens. 2021. “Open Scholarship in Australia: A Review of Needs, Barriers, and Opportunities.” Digital Scholarship in the Humanities, 36 (4): 795–812. https://doi.org/10.1093/llc/fqaa063. Peer‑reviewed assessment from CA‑Aus collaborators surfaces structural barriers (policy, incentives, infrastructure) and opportunities for publicly engaged open scholarship. Frames how national policy, incentives, and infrastructure shape whether openness reduces or reproduces inequities. El Khatib, Randa, Lindsey Seatter, Tracey El Hajj, Conrad Leibel, Alyssa Arbuckle, Ray Siemens, Caroline Winter, and the ETCL and INKE Research Groups. 2019. “Open Social Scholarship Annotated Bibliography.” KULA 3 (1): 1–141. https://doi.org/10.5334/kula.58. Field‑defining synthesis of OSS foundations, surveying open access, participatory publishing, crowdsourcing, social knowledge creation, and policy. Establishes terminology and exemplars that connect openness with social responsibility. Defines a master index to map subdomains of openness and identify non‑AI precedents for care, accessibility, and accountability in scholarly infrastructures. Maxwell, John W. 2015. “Beyond Open Access to Open Publication and Open Scholarship.” Scholarly and Research Communication 6 (3): 1–10. https://doi.org/10.22230/src.2015v6n3a202. Argues that simply “access” is insufficient; emphasizes usability, participatory review, and community‑responsive publishing workflows. Repositions openness as an ecosystem of practices with embedded care and accountability. Yields criteria for evaluating whether “openness” translates to meaningful, non‑extractive participation and shared authority in knowledge production. Introduction AI and Open