Generative artificial intelligence systems are increasingly deployed as knowledge infrastructures rather than discrete tools. They summarize archives, translate across languages, retrieve historical information, generate scholarly prose, and mediate access to cultural memory with an authority that often exceeds that of traditional institutional actors. In policy and governance contexts, however, AI is still primarily treated as a technical artifact: something to be regulated through risk management, bias mitigation, transparency requirements, or compliance frameworks. This paper argues that such approaches are insufficient because they fail to address a more fundamental issue: generative AI systems function as epistemic institutions that shape what is remembered, what is retrievable, and what is rendered authoritative. Governing AI therefore requires not only technical oversight, but governance of knowledge itself.
This paper advances the claim that Digital Humanities (DH) constitutes an underrecognized but critically necessary AI governance field. DH brings a mature set of methods, theoretical frameworks, and institutional practices for evaluating mediated knowledge, curating memory, and contesting epistemic authority, capacities that are largely absent from contemporary AI governance regimes. Rather than positioning humanists as peripheral “ethical advisors” or downstream critics of technological development, the paper reframes DH as a site of governance expertise capable of shaping how AI systems are designed, evaluated, and institutionally deployed.
The paper builds from a growing body of feminist, decolonial, and critical race scholarship that demonstrates how contemporary AI systems inherit rather than transcend historical epistemologies structured by race, gender, colonialism, and exclusion. This work has shown that algorithmic harm is not a technical aberration but a predictable outcome of epistemic inheritance: the compression of historical hierarchies into computational infrastructure and their reappearance as neutral, objective, or data-driven outputs. While this scholarship has been effective in diagnosing the epistemological foundations of AI, less attention has been paid to the governance implications of treating AI as a system that produces and authorizes knowledge. This paper takes that next step, shifting the analytical focus from epistemic critique to epistemic governance.
Central to this argument is the concept of generative AI as a “memory machine.” Large language models and related systems do not merely store information; they operationalize memory through retrieval, ranking, synthesis, and narration. In so doing, they collapse distinctions that have long structured humanities practice: between archive and interpretation, source and summary, provenance and paraphrase, memory and meaning. When AI-generated outputs are treated as recall rather than interpretation, epistemic authority is laundered through fluency. The result is a new form of mediated knowledge in which historical specificity, uncertainty, and contestation are smoothed away, while claims appear increasingly self-evident.
The paper argues that these dynamics cannot be adequately governed through prevailing AI policy frameworks, which tend to focus on fairness metrics, model performance, or post hoc explanations. Such frameworks often assume that knowledge exists prior to AI and that governance concerns arise only when outputs deviate from expected norms. By contrast, DH scholarship has long treated knowledge as produced through mediation: shaped by archival selection, classification systems, metadata standards, interface design, and institutional power. This perspective enables a different governance question to be asked: not only whether AI outputs are accurate or unbiased, but how authority is constructed, whose knowledge is centered or erased, and what forms of interpretation are foreclosed by design.
Methodologically, the paper is conceptual and synthetic. It draws on digital humanities scholarship in archival studies, critical classification, metadata theory, and interface critique, alongside work in AI governance, feminist epistemology, design justice, and Indigenous data sovereignty. Rather than presenting new empirical findings, the paper develops a governance framework grounded in humanities methods of interpretation and accountability. It also employs a small set of illustrative deployment scenarios, not as evaluative case studies, but as analytic prompts, to show how governance decisions become embedded in AI systems through seemingly mundane design choices. These scenarios include the use of generative AI to summarize or describe archival collections; AI-mediated translation in scholarly and cultural contexts; and institutional deployment of large language models as research or teaching assistants.
In each case, the paper identifies where epistemic authority shifts from human communities and institutions to computational systems, often without explicit deliberation or consent.
The paper’s primary contribution is a DH-derived AI governance framework organized around five core dimensions of epistemic accountability. First, provenance visibility: AI systems must make legible what sources are being drawn upon, at what level of abstraction, and with what degree of representativeness. Second, context preservation: systems should be evaluated on their capacity to maintain historical, cultural, and linguistic context rather than collapsing difference into generalized summaries. Third, contestability and redress: communities and institutions must have mechanisms to challenge, annotate, correct, or refuse AI-mediated representations of their histories and knowledges. Fourth, uncertainty communication: rather than presenting fluent outputs as settled knowledge, AI systems should surface ambiguity, disagreement, and limits of inference as first-order features. Fifth, community authority and benefit: governance must ask not only who is protected from harm, but who has decision-making power over how AI systems engage with cultural memory as well as who materially benefits from their deployment.
By articulating these dimensions, the paper reframes AI governance as a question of institutional design rather than technical optimization. It argues that DH provides governance primitives—conceptual tools, evaluative criteria, and institutional practices—that can travel across policy, infrastructure, and design contexts. These primitives do not replace legal or technical regulation, but they address a layer of governance that is currently underdeveloped: the governance of meaning, memory, and authority.
The paper also intervenes in debates about “engagement” in digital humanities and AI policy. Engagement is frequently framed as access: making archives searchable, translation frictionless, or knowledge more widely available through technological mediation. While such efforts are often presented as inherently democratizing, this paper argues that engagement without governance risks becoming extractive. When generative AI systems engage with cultural memory without accountability to the communities, institutions, and historical contexts they mediate, they reproduce long-standing patterns of epistemic domination under the guise of inclusion and participation. Meaningful engagement, I contend, requires governance structures that treat interpretation as a political act, memory as a contested site of authority, and access as inseparable from responsibility. From this perspective, engagement is not merely a question of interface design or scale, but of who is empowered to shape how knowledge is remembered, narrated, and legitimized.
In positioning DH as an AI governance field, the paper challenges a common division of labor in which humanists are invited to critique technologies after they are built, while governance decisions are made elsewhere. Instead, it argues for integrating DH expertise into the full lifecycle of AI systems that mediate knowledge: from design and evaluation to institutional deployment and policy formation. The paper concludes by outlining how DH governance frameworks could be operationalized within universities, libraries, cultural institutions, and public-sector AI initiatives, offering a pathway for humanities scholarship to move from critique to structural intervention.
Ultimately, the paper contends that if generative AI systems are becoming part of how societies remember, interpret, and authorize knowledge, then governance must be grounded in fields that have spent decades interrogating precisely these processes. Digital humanities, with its attention to mediation, history, and power, is not adjacent to AI governance; it is indispensable to it.
References
Assmann, Jan (2011): Cultural Memory and Early Civilization: Writing, Remembrance, and Political Imagination. Cambridge: Cambridge University Press.
Benjamin, Ruha (2019): Race After Technology: Abolitionist Tools for the New Jim Code. Cambridge: Polity Press, 2019.
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, 610-623.
Bowker, Geoffrey C., and Susan Leigh Star (2000): Sorting Things Out: Classification and Its Consequences. Cambridge, MA: MIT Press, 2000.
Buolamwini, Joy, and Timnit Gebru (2018): "Gender shades: Intersectional Accuracy Disparities in Commercial Gender Classification", in: Conference on Fairness, Accountability and Transparency, PMLR: 77-91.
Caswell, Michelle (2014): Archiving the Unspeakable: Silence, Memory, and the Photographic Record in Cambodia. Madison: University of Wisconsin Press.
Costanza-Chock, Sasha (2020): Design justice: Community-led Practices to Build the Worlds We Need. Cambridge, MA: MIT Press.
Derrida, Jacques (1996): Archive Fever: A Freudian Impression. Chicago: University of Chicago Press.
D'ignazio, Catherine, and Lauren F. Klein (2023): Data Feminism. Cambridge, MA: MIT Press, 2023.
Foucault, Michel (1972): The Archaeology of Knowledge and the Discourse on Language. A.M. Sheridan Smith (trans). New York: Pantheon Books.
Haraway, Donna (1988): “Situated Knowledges: The Science Question in Feminism and the Privilege of Partial Perspective”, in: Feminist Studies 14, no. 3: 575–99.
Harris, Verne (2002): “The Archival Sliver: A Perspective on the Construction of Social Memory in Archives and the Transition from Apartheid to Democracy”, in: Carolyn Hamilton / Verne Harris / Jane Taylor / Michele Pickover / Graeme Reid / Razia Saleh (eds.) Refiguring the Archive. Dordrecht: Springer, 2002.
Kember, Sarah, and Joanna Zylinska (2012): Life After New Media: Mediation As a Vital Process. Cambridge, MA: MIT Press, 2012.
Milan, Stefania, and Lonneke van der Velden (2016): "The Alternative Epistemologies of Data Activism", in: Digital Culture & Society 2, no. 2: 57–74.
Nakamura, Lisa (2007): Digitizing Race: Visual Cultures of the Internet. Minneapolis: University of Minnesota.
Noble, Safiya Umoja (2018): Algorithms of Oppression: How Search Engines Reinforce Racism. New York: New York University Press.
Pasquale, Frank (2015): The Black Box Society. Cambridge, MA: Harvard University Press.
Ricaurte, Paola (2019): "Data Epistemologies, the Coloniality of Power, and Resistance", in: Television & New Media 20, no. 4: 350–65.
Risam, Roopika (2018): New Digital Worlds: Postcolonial Digital Humanities in Theory, Praxis, and Pedagogy. Evanston, IL: Northwestern University Press.
Star, Susan Leigh and Karen Ruhleder (1996): “Steps Toward an Ecology of Infrastructure: Design and Access for Large Information Spaces”, in: Information Systems Research 7(1): 111–34.
United States. National Institute of Standards and Technology (2023): Artificial Intelligence Risk Management Framework (AI RMF 1.0). Gaithersburg, MD: National Institute of Standards and Technology
Wynter, Sylvia. (2003): " Unsettling the Coloniality of Being/Power/Truth/Freedom: Toward the Human, After Man, its Overrepresentation—An Argument", in: CR: The New Centennial Review 3, no. 3: 257–337.