di Mika (Jaeyun) Noh

Il contributo di Mika (Jaeyun) Noh apre con forza il numero verso una dimensione internazionale, collocando il tema dell’intelligenza artificiale all’interno di una prospettiva che considero oggi decisiva: quella della governance culturale.

Non si tratta soltanto di regolare tecnologie, ma di interrogarsi su chi detiene l’autorità culturale, su come si costruisce la legittimità delle istituzioni e su quali forme di partecipazione siano necessarie quando sistemi algoritmici iniziano a mediare l’esperienza culturale.

L’analisi proposta è di grande rilievo perché evidenzia un limite spesso trascurato: i modelli di governance dell’AI, pur avanzati sul piano tecnico e normativo, risultano ancora insufficienti quando si confrontano con la complessità dei contesti culturali. È qui che emerge la necessità di un approccio specifico, capace di integrare responsabilità epistemica, partecipazione e sovranità culturale.

In questo quadro, assume particolare valore il riferimento alle esperienze italiane e, in modo specifico, al lavoro di DiCultHer, riconosciuto come esempio di pratica partecipativa nella costruzione del patrimonio culturale digitale. Un riconoscimento che non è solo motivo di soddisfazione, ma anche responsabilità: continuare a sviluppare modelli che mettano al centro le comunità, l’educazione e la cittadinanza.

Questo contributo ci invita a compiere un passaggio fondamentale: considerare l’intelligenza artificiale non soltanto come tecnologia da utilizzare, ma come infrastruttura che ridefinisce le condizioni della produzione culturale, della memoria e della democrazia.

In questa prospettiva, la governance culturale dell’AI non è un ambito specialistico, ma una delle questioni centrali del nostro tempo.

Carmine Marinucci

Abstract

As artificial intelligence systems increasingly mediate cultural experiences—from museum exhibitions to digital heritage platforms—cultural institutions face a profound governance challenge: how to maintain curatorial authority, institutional legitimacy, and public trust when algorithmic systems make aesthetic, pedagogical, and representational decisions. This article examines the emergence of AI-driven cultural curation across museums, digital heritage institutions, and creative platforms, arguing that existing AI governance frameworks—focused predominantly on technical safety, data privacy, and regulatory compliance—systematically overlook cultural dimensions essential to institutional accountability. Drawing on policy analysis of South Korea’s AI Basic Act (2024), institutional ethnography of Korean cultural institutions deploying AI systems, and comparative examination of EU and Italian digital heritage frameworks, I propose a Cultural AI Governance Framework integrating three core dimensions: epistemic accountability (algorithmic transparency adapted to cultural contexts), multi-stakeholder participation (including cultural workers, communities, and audiences), and cultural sovereignty (protecting heritage data and creative labor from extractive AI training practices). The article demonstrates this framework through case studies including Korea’s National Museum AI curator controversy, Italy’s digital cultural heritage initiatives, and participatory design methodologies developed by DiCultHer. I argue that cultural AI governance is not supplementary to technical governance but constitutes a distinct domain requiring specialized frameworks, competencies, and institutional arrangements—and that pioneering institutions developing robust cultural governance will define standards as AI transforms cultural production globally.

Riassunto

Mentre i sistemi di intelligenza artificiale mediano sempre più le esperienze culturali—dalle mostre museali alle piattaforme del patrimonio digitale—le istituzioni culturali affrontano una sfida profonda di governance: come mantenere l’autorità curatoriale, la legittimità istituzionale e la fiducia pubblica quando i sistemi algoritmici prendono decisioni estetiche, pedagogiche e rappresentative. Questo articolo esamina l’emergere della curatela culturale guidata dall’IA attraverso musei, istituzioni del patrimonio digitale e piattaforme creative, sostenendo che i quadri di governance dell’IA esistenti—focalizzati prevalentemente sulla sicurezza tecnica, la privacy dei dati e la conformità normativa—trascurano sistematicamente le dimensioni culturali essenziali per la responsabilità istituzionale. Attingendo dall’analisi politica della Legge di Base sull’IA della Corea del Sud (2024), dall’etnografia istituzionale delle istituzioni culturali coreane che implementano sistemi di IA, e dall’esame comparativo dei quadri del patrimonio digitale dell’UE e italiani, propongo un Quadro di Governance dell’IA Culturale che integra tre dimensioni fondamentali: responsabilità epistemica (trasparenza algoritmica adattata ai contesti culturali), partecipazione multi-stakeholder (inclusi lavoratori culturali, comunità e pubblico), e sovranità culturale (protezione dei dati del patrimonio e del lavoro creativo da pratiche estrattive di addestramento dell’IA). L’articolo dimostra questo quadro attraverso casi di studio, inclusa la controversia del curatore IA del Museo Nazionale coreano, le iniziative italiane del patrimonio culturale digitale e le metodologie di progettazione partecipativa sviluppate da DiCultHer. Sostengo che la governance dell’IA culturale non è supplementare alla governance tecnica ma costituisce un dominio distinto che richiede quadri, competenze e accordi istituzionali specializzati—e che le istituzioni pionieristiche che sviluppano una governance culturale robusta definiranno gli standard mentre l’IA trasforma la produzione culturale globalmente.

Résumé

Alors que les systèmes d’intelligence artificielle médiatisent de plus en plus les expériences culturelles—des expositions muséales aux plateformes du patrimoine numérique—les institutions culturelles font face à un défi de gouvernance profond : comment maintenir l’autorité curatoriale, la légitimité institutionnelle et la confiance publique lorsque les systèmes algorithmiques prennent des décisions esthétiques, pédagogiques et représentatives. Cet article examine l’émergence de la curation culturelle pilotée par l’IA à travers les musées, les institutions du patrimoine numérique et les plateformes créatives, soutenant que les cadres de gouvernance de l’IA existants—axés principalement sur la sécurité technique, la confidentialité des données et la conformité réglementaire—négligent systématiquement les dimensions culturelles essentielles à la responsabilité institutionnelle. M’appuyant sur l’analyse politique de la Loi fondamentale sur l’IA de la Corée du Sud (2024), l’ethnographie institutionnelle des institutions culturelles coréennes déployant des systèmes d’IA, et l’examen comparatif des cadres du patrimoine numérique de l’UE et italiens, je propose un Cadre de Gouvernance de l’IA Culturelle intégrant trois dimensions fondamentales : responsabilité épistémique (transparence algorithmique adaptée aux contextes culturels), participation multi-acteurs (incluant les travailleurs culturels, les communautés et les publics), et souveraineté culturelle (protection des données patrimoniales et du travail créatif contre les pratiques extractives d’entraînement de l’IA). L’article démontre ce cadre à travers des études de cas, y compris la controverse du curateur IA du Musée National coréen, les initiatives italiennes du patrimoine culturel numérique et les méthodologies de conception participative développées par DiCultHer. Je soutiens que la gouvernance de l’IA culturelle n’est pas supplémentaire à la gouvernance technique mais constitue un domaine distinct nécessitant des cadres, des compétences et des arrangements institutionnels spécialisés—et que les institutions pionnières développant une gouvernance culturelle robuste définiront les normes alors que l’IA transforme la production culturelle globalement.

Keywords: Cultural AI governance; Digital cultural heritage; Algorithmic curation; Museum technology; Participatory design; Cultural data sovereignty; Institutional legitimacy

1. Introduction: The Curatorial Crisis of Algorithmic Intelligence

In February 2024, Seoul’s National Museum of Korea launched an AI-powered “intelligent curator” designed to personalize visitor experiences through algorithmic recommendation of artworks, optimal gallery navigation, and contextual interpretation.¹ Within three months, the museum faced its first major institutional crisis—not from technical malfunction, but from cultural backlash. Korean artists discovered their works had been used to train the AI system without consultation or consent. Museum visitors reported feeling “manipulated” by invisible algorithmic curation. Cultural critics questioned whether institutions could maintain curatorial authority when algorithms made aesthetic decisions. The technical success became a governance failure.

This case exemplifies a pattern emerging globally as cultural institutions deploy artificial intelligence across curation, heritage preservation, audience engagement, and digital experience design. While technical AI governance frameworks address data privacy, algorithmic fairness, and safety—issues codified in regulations like the EU AI Act² and South Korea’s AI Basic Act³—they systematically overlook cultural dimensions: Who holds authority when algorithms curate? How do institutions maintain legitimacy amid algorithmic mediation? What frameworks protect cultural heritage data and creative labor from extractive AI training practices? How can communities participate in governing AI systems shaping cultural representation?

This article proposes that cultural AI governance constitutes a distinct domain requiring specialized frameworks beyond technical governance—analogous to how environmental governance extends beyond engineering to encompass ecological, social, and intergenerational dimensions.⁴ Cultural institutions pioneering robust governance frameworks will not only mitigate risks but define emerging standards as AI fundamentally transforms cultural production, preservation, and access globally.

My analysis draws on three complementary methodologies: (1) policy analysis of AI governance frameworks in South Korea, the EU, and Italy, examining how (or whether) they address cultural dimensions; (2) institutional ethnography based on direct observation and interviews with Korean museums, cultural institutions, and creative organizations deploying AI systems (2023-2026); and (3) comparative examination of participatory design methodologies, particularly Italy’s DiCultHer network, demonstrating alternative governance models.⁵

The article proceeds in five sections. Section 2 analyzes governance gaps in current AI frameworks. Section 3 develops a Cultural AI Governance Framework integrating epistemic accountability, multi-stakeholder participation, and cultural sovereignty. Section 4 demonstrates this framework through comparative case studies. Section 5 provides implementation guidance for cultural institutions. I conclude by arguing that cultural AI governance represents not merely risk mitigation but an opportunity to reimagine institutional accountability for the algorithmic age.

2. The Governance Gap: What Technical Frameworks Miss

2.1 Current State: AI Governance in Cultural Heritage

Existing AI governance frameworks—from the European Commission’s Ethics Guidelines for Trustworthy AI⁶ to UNESCO’s Recommendation on the Ethics of AI⁷—emphasize principles including transparency, accountability, fairness, and human oversight. These frameworks inform regulatory instruments like the EU AI Act, which classifies certain AI systems affecting “cultural and societal values” as potentially high-risk.⁸

However, implementation in cultural contexts reveals systematic gaps. My analysis of South Korea’s AI Basic Act—the world’s second comprehensive AI legislation after the EU—demonstrates this pattern. Despite extensive provisions on “high-impact AI systems,” the Act barely addresses cultural AI applications. There are no special protections for AI systems generating cultural content, no requirements for cultural heritage consultation in training data sourcing, no frameworks governing AI’s impact on creative labor, and no standards for algorithmic transparency adapted to cultural contexts.⁹

This regulatory gap is not accidental but structural. Policymakers treat AI governance as fundamentally a technical/industrial issue rather than a cultural infrastructure question. When I interviewed officials at Korea’s Ministry of Science and ICT (MSIT) who drafted the AI Basic Act, they acknowledged that cultural dimensions “fell through the cracks” between technical safety regulation (MSIT’s mandate) and cultural policy (Ministry of Culture, Sports and Tourism’s mandate), with neither ministry claiming primary responsibility for cultural AI governance.¹⁰

2.2 Three Dimensional Gaps

The Authority Gap: Technical governance asks “Is the AI system safe and compliant?” Cultural governance must ask “Who holds institutional authority when algorithms make curatorial decisions?” Traditional cultural institutions derive legitimacy from expertise—curators, conservators, educators whose professional judgment is accountable to publics, communities, and professional standards. Algorithmic curation distributes authority across human-AI hybrid systems in ways existing accountability frameworks cannot capture.

Consider a concrete example: When Seoul’s National Museum deployed its AI curator, the system recommended artworks based on visitor behavioral data (time spent viewing, movement patterns, interaction history). From a technical governance perspective, this raised data privacy questions (appropriately addressed through consent mechanisms and anonymization). From a cultural governance perspective, however, the deeper issue involved epistemic authority: the AI optimized for engagement metrics (time on site, return visits) that may conflict with educational goals (challenging perspectives, contemplative viewing) or curatorial values (historically significant but less immediately engaging works). No framework existed for making these trade-offs transparent, debatable, or accountable to museum professionals and publics.¹¹

The Participation Gap: While technical governance emphasizes “human oversight,” it rarely specifies which humans. Cultural governance requires multi-stakeholder participation including cultural workers (whose expertise AI potentially displaces), source communities (whose heritage AI systems represent), and audiences (whose experiences AI mediates).

Italy’s DiCultHer network exemplifies alternative models. Their “Crowddreaming” methodology engages diverse stakeholders—students, researchers, cultural professionals, communities—in co-designing digital cultural heritage projects. Rather than treating AI deployment as technical implementation requiring only IT expertise, DiCultHer positions it as participatory cultural practice requiring collaborative governance.¹² This approach produced initiatives like “Quintana 4D,” where schools in Foligno co-created digital reconstructions of historical festivals, establishing community ownership over digital heritage representation rather than outsourcing to technical specialists.¹³

The Cultural Sovereignty Gap: Technical governance addresses data privacy but overlooks cultural data sovereignty—the right of communities to control representations of their heritage, knowledge systems, and creative expressions. When AI systems train on museum collections, indigenous cultural knowledge, or artists’ portfolios without consultation or compensation, they enact what scholars call “data colonialism”¹⁴—extracting cultural resources to generate value elsewhere.

The Getty Images lawsuit against Stability AI illustrates this gap. While framed legally as copyright violation, the underlying issue involves cultural sovereignty: should commercial AI companies extract value from creative labor (photographers’ work) without consent or compensation? Technical governance’s privacy frameworks don’t address this because data was legally accessible. Cultural governance must establish consent, compensation, and control frameworks for cultural data use in AI training.¹⁵

2.3 Why Technical Governance Alone Fails Cultural Contexts

Technical AI governance operates through standardization—common metrics (accuracy, fairness), auditable processes (testing protocols), and universal principles (transparency, accountability). This approach succeeds in domains where optimization targets align across contexts: fraud detection should minimize false positives universally; medical diagnosis should maximize accuracy consistently.

Cultural domains, however, involve fundamentally contested values. What constitutes “good” curation? It depends on institutional mission (educational vs. entertainment), audience (specialists vs. general public), cultural context (Western canon vs. decolonial critique), and temporal considerations (preservation vs. innovation). AI systems trained to optimize engagement may produce algorithmically successful but culturally inappropriate outcomes—popular but historically inaccurate, accessible but intellectually shallow, inclusive in representation but extractive in data practices.

Nello Cristianini’s concept of “the shortcut” illuminates this challenge: AI systems optimize for measurable proxies (clicks, time on site, sentiment scores) that shortcut deeper cultural goals (critical thinking, aesthetic appreciation, community ownership).¹⁶ Technical governance can ensure AI systems achieve their optimization targets accurately and fairly. Cultural governance must ensure those targets align with institutional missions and cultural values—a fundamentally different question requiring different expertise, frameworks, and accountability mechanisms.

3. Cultural AI Governance Framework: Three Pillars

Based on comparative analysis of governance failures (National Museum of Korea), partial successes (Korea’s Ministry of Culture cultural AI committees), and alternative models (DiCultHer’s participatory methodologies), I propose a Cultural AI Governance Framework integrating three dimensions: Epistemic Accountability, Multi-Stakeholder Participation, and Cultural Sovereignty.

3.1 Epistemic Accountability: Making Algorithmic Curation Legible

Epistemic accountability extends technical transparency (explaining how algorithms work) to cultural contexts (explaining why algorithmic decisions align with institutional values). This requires four mechanisms:

1. Cultural Impact Assessment Prior to AI deployment, institutions must conduct systematic assessment examining:

  • What curatorial/pedagogical/representational decisions will AI influence?
  • How do algorithmic optimization targets align (or conflict) with institutional mission?
  • Which stakeholders hold expertise relevant to these decisions?
  • What values trade-offs does AI deployment entail?

Korea’s Ministry of Culture, Sports and Tourism (MCST) developed Cultural AI Impact Assessment templates for museums deploying recommendation systems. These assess not just technical functionality but cultural implications: Does personalization enhance or undermine collective cultural experience? Does algorithmic efficiency displace professional curatorial judgment? Does data collection respect cultural sensitivities around sacred objects or sensitive heritage?¹⁷

2. Algorithmic Explanation Adapted to Cultural Literacy Standard algorithmic transparency (model cards, technical documentation) assumes technical expertise. Cultural institutions serve diverse audiences requiring culturally literate explanations.

Best practice example: Amsterdam’s Rijksmuseum deployed AI-generated artwork descriptions with clear labeling (“This text was generated by AI and reviewed by curators”) plus explanations framed for museum visitors: “The AI analyzed X artworks in our collection to identify common themes. Curators then verified historical accuracy and added contextual nuance the AI missed.”¹⁸ This maintains institutional authority (curators review AI output) while making algorithmic mediation transparent in culturally accessible terms.

3. Value Alignment Documentation Institutions must explicitly document how AI optimization targets align with cultural values. When Seoul’s National Museum calibrated its AI curator, technical teams optimized for “visitor satisfaction” measured through dwell time and return visits. Only after backlash did the museum realize this proxy conflicted with educational goals (challenging perspectives often reduce immediate satisfaction but produce deeper learning).¹⁹

Robust governance requires documenting: What are we optimizing for? Why? What values does this prioritize? What does it de-prioritize? Who decided? This documentation becomes accountable to stakeholders—staff, communities, audiences—creating space for contestation and revision.

4. Ongoing Epistemic Auditing Unlike technical systems where correct operation means consistent performance, cultural systems require adaptation as contexts evolve. What constituted appropriate representation in 2020 may not in 2026; decolonial curation standards continue developing; community preferences shift.

Cultural epistemic auditing establishes regular review cycles where institutional stakeholders reassess whether AI systems remain aligned with evolving values. Italy’s digital cultural heritage initiatives provide models: periodic stakeholder convenings (annually or bi-annually) review AI system performance not just technically (accuracy, uptime) but culturally (Does it serve intended communities? Reinforce or challenge problematic narratives? Support institutional mission?).²⁰

3.2 Multi-Stakeholder Participation: Beyond “Human Oversight”

Technical governance’s “human oversight” typically means engineers and managers reviewing AI outputs. Cultural governance requires substantive participation from:

Cultural Professionals: Curators, conservators, educators, archivists whose expertise AI potentially augments or displaces. These professionals must participate in:

  • Defining what AI should optimize (not defaulting to technical teams’ proxy metrics)
  • Reviewing training data for cultural appropriateness (not just legal compliance)
  • Establishing override protocols (when human judgment supersedes algorithmic recommendations)
  • Evaluating cultural impacts post-deployment

Source Communities: When AI systems represent cultural heritage—indigenous knowledge, diaspora history, marginalized voices—affected communities must participate in governance. This extends beyond consultation to co-decision-making authority over:

  • Whether their heritage is used in AI training
  • How it’s represented algorithmically
  • Who benefits from AI-generated value
  • What constitutes appropriate vs. extractive use

DiCultHer’s participatory methodologies demonstrate feasibility. Their projects establish community advisory boards with decision-making power over digital heritage representation, creating accountability mechanisms where communities can challenge or veto institutional decisions.²¹

Audiences/Publics: Museum visitors, digital heritage users, and creative platform participants experience AI-mediated culture. Participation mechanisms include:

  • Transparent communication about algorithmic mediation (not concealing AI involvement)
  • Feedback channels allowing contestation (reporting inappropriate recommendations)
  • Participatory evaluation informing AI system refinement

Creative Workers: Artists, photographers, writers, musicians whose labor trains AI systems. Current practice treats publicly accessible creative work as free training data. Cultural governance must establish:

  • Consent frameworks (opt-in rather than opt-out for AI training use)
  • Compensation mechanisms (analogous to copyright licensing)
  • Attribution requirements (crediting source material)
  • Professional development support (helping creative workers adapt to AI-augmented practice)

3.3 Cultural Sovereignty: Protecting Heritage and Creative Labor

Cultural sovereignty extends data sovereignty principles—already recognized for personal data (GDPR) and indigenous knowledge—to cultural heritage and creative expressions.

Heritage Data Governance: Cultural institutions hold stewardship responsibility over collections representing communities, historical periods, and cultural traditions. Using these collections to train commercial AI systems without community consultation constitutes governance failure.

Best practice framework, developed through Korea-EU comparative analysis:²²

  1. Heritage Data Audit: Catalog what cultural data exists, who it represents, what sensitivities attach
  2. Community Consultation: Engage represented communities about appropriate AI use
  3. Consent Protocols: Establish whether/how heritage data may be used in AI training
  4. Benefit Sharing: When AI systems generate value from cultural data, mechanisms return benefits to source communities (funding cultural programs, supporting heritage preservation, capacity building)
  5. Ongoing Stewardship: Regular review of whether AI uses remain aligned with community values

Creative Labor Protection: The Getty Images v. Stability AI case revealed extractive AI training practices: commercial systems trained on millions of creative works without consent or compensation.²³ Cultural governance must address this through:

  • Transparent Training Data Sourcing: AI developers must document and disclose what cultural content trains their systems
  • Licensing Frameworks: Cultural institutions and creative workers should control commercial AI training use through licensing analogous to existing copyright mechanisms
  • Fair Compensation: When AI systems derive value from creative labor, compensation frameworks must reward source creators
  • Professional Adaptation Support: As AI transforms creative work, institutions should invest in helping creative professionals develop AI-augmented practices rather than simply facing displacement

Cultural Commons Management: Some argue all publicly accessible cultural content should be available for AI training, analogous to reading/viewing for human learning. Cultural governance must distinguish legitimate cultural commons use from extractive commercialization.

Proposed framework: Non-commercial, educational, research AI applications may access cultural data through cultural commons principles (with attribution). Commercial AI systems must obtain licenses, pay compensation, and demonstrate community benefit. This parallels Creative Commons licensing adapted to AI era.²⁴

4. Comparative Case Studies: Governance Failures and Successes

4.1 Failure Case: National Museum of Korea AI Curator

Seoul’s National Museum of Korea AI curator controversy (2024) demonstrates governance failure consequences. The museum partnered with a Korean AI company to develop a recommendation system personalizing visitor experiences. Technical governance was robust: data privacy protections, algorithmic fairness testing, accessibility features. Cultural governance was absent.

What went wrong:

Training Data Extraction: The AI company scraped museum’s digital collection plus Korean artists’ online portfolios without specific consent for AI training. While legally defensible (publicly accessible data), culturally this violated professional norms and artist expectations.

Authority Displacement Without Consultation: Museum curators learned about AI deployment through internal memo, not participatory design process. The system made aesthetic recommendations (which artworks visitors should see) traditionally within curatorial expertise, without involving curators in defining optimization criteria.

Optimization Misalignment: The AI optimized for visitor engagement (measured through app interactions, dwell time, return visits). This created incentives toward popular, accessible works over historically significant but challenging pieces—inverting museum’s educational mission.

Concealed Algorithmic Mediation: Visitors received “personalized recommendations” without clear disclosure these came from AI, not human curators. When revealed, many felt manipulated.

Institutional Consequences:

  • Artist boycott threatening to withdraw works from exhibitions
  • Media backlash questioning museum’s cultural authority
  • Internal staff conflict between technical/administrative teams (supporting AI) and curatorial staff (feeling displaced)
  • Public trust erosion requiring years to rebuild
  • Government review of cultural institution AI policies

Governance Lessons:

Had the museum implemented cultural governance frameworks—cultural impact assessment, curatorial participation, transparent communication, artist consent protocols—the technical capabilities could have enhanced rather than undermined institutional legitimacy. The failure was governance, not technology.²⁵

4.2 Partial Success: MCST Cultural AI Committees

Following the National Museum controversy, Korea’s Ministry of Culture, Sports and Tourism (MCST) established Cultural AI Ethics Committees for institutions deploying AI systems. These committees differ from standard AI ethics boards by including cultural professionals and community representatives, not just technologists and ethicists.

Structure:

  • Museum curators and educators
  • Artists and creative workers
  • Cultural heritage community representatives
  • AI technical experts
  • Legal/ethics specialists
  • Public members

Mandate:

Committees review AI deployments before launch, assessing:

  • Cultural appropriateness of training data sourcing
  • Alignment between AI optimization targets and institutional mission
  • Impact on cultural workers’ professional roles
  • Community consultation adequacy
  • Transparency of algorithmic mediation to publics
  • Long-term cultural implications

Outcomes:

Cultural institutions using this governance model report higher staff acceptance, fewer controversies, and more effective AI adoption. For example, National Museum of Modern and Contemporary Art (MMCA) deployed an AI-enhanced archive search system that:

  • Involved archivists in design (they defined what “relevant” search results mean)
  • Obtained specific consent from living artists for AI training use
  • Maintained professional archivist review of AI-generated metadata
  • Transparently labeled AI vs. human-generated content
  • Established community feedback mechanisms allowing contestation

The system achieved technical objectives (faster archive access, improved discoverability) while maintaining cultural integrity and institutional legitimacy.²⁶

Limitations:

MCST committees are advisory, not regulatory. Institutions can ignore recommendations. Funding/capacity constraints limit committee reach (only major institutions participate). The framework needs scaling and enforcement mechanisms.

4.3 Alternative Model: DiCultHer’s Participatory Governance

Italy’s DiCultHer network demonstrates participatory cultural governance through their “Crowddreaming” methodology applied to digital heritage projects.²⁷

Core Principles:

  • Co-creation over consultation: Communities don’t merely advise but co-design digital cultural heritage
  • Distributed expertise: Technical knowledge doesn’t supersede cultural/local knowledge
  • Transparent process: Design decisions made visibly, accountably, with clear rationale
  • Iterative governance: Ongoing community input, not one-time approval

Case Study: Quintana 4D

Schools in Foligno collaborated with DiCultHer to digitally reconstruct historical Quintana festival. The project integrated AI technologies (3D reconstruction, automated archival analysis) within participatory governance:

Community Cultural Authority: Local historians and festival participants defined what needed preservation, what constituted accuracy, how heritage should be represented—not external technical teams.

Student Participation: Students weren’t passive learners but active co-creators, developing digital competencies while exercising cultural authority over their community’s heritage representation.

Transparent AI Use: When AI assisted archival analysis or 3D reconstruction, this was documented and explained. Community members could challenge AI interpretations.

Ongoing Stewardship: The digital heritage isn’t “finished” but continuously maintained by community, with governance structures for future updates/revisions.

Scaling Insights:

DiCultHer’s model demonstrates participatory cultural AI governance is feasible beyond elite institutions. Schools, small museums, community organizations can implement these frameworks. Key success factors:

  • Institutional commitment to genuine participation (not tokenistic consultation)
  • Capacity building (helping communities develop digital/AI literacy for meaningful participation)
  • Patient timelines (participatory processes require more time than top-down implementation)
  • Flexible tools (technology must adapt to community needs, not communities to technology)²⁸

4.4 Comparative Synthesis: Governance Design Principles

Analyzing these cases reveals core principles distinguishing effective cultural AI governance:

1. Cultural Expertise as Co-Equal to Technical Expertise
Successful governance treats cultural professionals, communities, and audiences as legitimate authorities whose expertise shapes AI design—not afterthought reviewers approving technical teams’ decisions.

2. Proactive Rather Than Reactive
Governance must inform design from inception, not review completed systems. National Museum’s failure stemmed partly from governance as post-deployment oversight rather than co-design.

3. Transparency as Cultural Practice, Not Technical Disclosure
Explaining how algorithms work (technical transparency) differs from explaining why algorithmic decisions align with values (cultural transparency). Latter requires culturally literate communication.

4. Authority Preservation Through Hybrid Arrangements
Rather than AI replacing human authority or humans merely overseeing AI, effective governance creates hybrid arrangements where professional expertise shapes AI design/oversight while benefiting from AI capabilities.

5. Iterative Rather Than Static
Cultural values evolve; AI systems must too. Governance requires ongoing review, not one-time approval.

5. Implementation Guidance for Cultural Institutions

Based on comparative analysis, I offer practical guidance for museums, archives, libraries, and digital heritage institutions implementing cultural AI governance.

5.1 Phase 1: Cultural Governance Infrastructure (Months 1-3)

Establish Cultural AI Review Committee:

Composition:

  • Cultural professionals from relevant departments (curatorial, education, conservation, archives)
  • Community representatives (from populations the institution serves and represents)
  • Technical staff (IT, digital, data specialists)
  • Ethics/legal expertise
  • External advisors (academics, other institutions)

Charter: Document committee mandate, decision-making authority, review processes, accountability mechanisms.

Capacity Building: Committee members need training—cultural professionals in AI basics, technical staff in cultural governance principles, all members in participatory methodologies.

Conduct Institutional AI Audit:

Map current and planned AI systems:

  • What AI applications exist or are planned?
  • What cultural decisions do they influence?
  • What training data do they use?
  • Who designed them? Who oversees them?
  • What governance (if any) currently exists?

Identify gaps where cultural governance is absent.

5.2 Phase 2: Framework Development (Months 4-6)

Develop Cultural Impact Assessment Template:

Create standardized assessment asking:

  • What cultural content/heritage does AI use?
  • What curatorial/pedagogical/representational decisions does AI make?
  • How do optimization targets align with institutional mission?
  • Which communities does this AI affect?
  • What consultation has occurred?
  • What transparency/accountability mechanisms exist?
  • What are risks? Mitigation strategies?

Require assessment before any AI deployment.

Establish Consent and Licensing Protocols:

For training data:

  • Audit what cultural content AI uses
  • Establish consent requirements (especially for contemporary work, sensitive heritage, indigenous knowledge)
  • Develop licensing frameworks (when commercial AI companies request institutional data)
  • Create benefit-sharing mechanisms (when AI generates value from cultural data)

For AI outputs:

  • Disclosure requirements (when AI generates content, label it clearly)
  • Attribution protocols (credit both AI systems and source data)
  • Review processes (human cultural professionals review AI outputs)

5.3 Phase 3: Participatory Design Pilots (Months 7-12)

Select Pilot AI Project:

Choose a manageable AI application (e.g., archive search enhancement, digital exhibition recommendation, collection metadata generation) for participatory governance pilot.

Implement Co-Design Process:

Rather than technical teams developing AI then seeking approval:

  1. Community Consultation: Engage affected stakeholders (staff, audiences, source communities) about needs, concerns, values
  2. Collaborative Design: Co-create specifications: what should AI optimize? What constitutes success? What are red lines?
  3. Transparent Development: Regular updates, ongoing input, visible decision-making
  4. Iterative Testing: User feedback informs refinement
  5. Participatory Evaluation: Stakeholders assess whether deployment meets goals, serves communities, maintains values

Document Lessons Learned:

  • What worked? What was challenging?
  • How did participatory governance change technical design?
  • What cultural issues arose that technical frameworks missed?
  • How much time/resources did this require vs. traditional development?
  • What would we do differently next time?

5.4 Phase 4: Scaling and Institutionalization (Year 2+)

Expand Governance to All AI Systems:

Once pilot proves feasibility, require cultural governance for all institutional AI applications.

Develop Institutional Policy:

Codify cultural AI governance as institutional policy:

  • Mandate cultural impact assessment
  • Require multi-stakeholder review
  • Establish consent/licensing protocols
  • Set transparency standards
  • Define accountability mechanisms
  • Allocate resources (staff time, training, community participation compensation)

External Accountability:

Make cultural AI governance visible:

  • Publish governance frameworks on institutional website
  • Report annually on AI use, governance processes, stakeholder input
  • Invite external review (peer institutions, community audits)
  • Share lessons learned with field

Network Learning:

Join or establish networks of cultural institutions developing cultural AI governance (e.g., extend DiCultHer model, create national/international working groups). Collective learning accelerates individual institution progress.

5.5 Resource Requirements and Sustainability

Staff Time:

  • Cultural AI Review Committee: ~10-15 hours/month per member
  • Impact assessments: ~20-40 hours per AI system
  • Participatory design: 50-100% more time than traditional development (initially)

Expertise:

  • Cultural professionals need basic AI literacy (training)
  • Technical staff need cultural governance understanding (training)
  • Facilitation skills for participatory processes (potentially hire or contract)

Community Participation:

  • Budget for compensating community participants (honoraria, stipends)
  • Support community capacity building (workshops, resources)

Sustainability Strategy:

Initial investment (governance infrastructure, training, pilot projects) is substantial. However, costs decrease as:

  • Staff develop expertise (less external support needed)
  • Processes become routine (templates, protocols established)
  • Governance prevents costly failures (reputational damage, staff conflict, community backlash)

Long-term, cultural AI governance is cost-effective risk management plus institutional capacity building.

6. Conclusion: Cultural Governance as Institutional Opportunity

This article has argued that cultural AI governance constitutes a distinct domain requiring specialized frameworks beyond technical governance. The National Museum of Korea case demonstrated failure consequences; MCST committees and DiCultHer networks showed alternative possibilities. The proposed Cultural AI Governance Framework—integrating epistemic accountability, multi-stakeholder participation, and cultural sovereignty—offers cultural institutions actionable guidance.

Three concluding reflections on broader implications:

6.1 Cultural Governance as Democratic Infrastructure

Cultural institutions serve democratic functions: preserving collective memory, enabling public deliberation, supporting informed citizenship. When AI systems mediate these functions without adequate governance, they undermine democratic infrastructure.

Robust cultural AI governance isn’t merely institutional self-interest but civic responsibility. Museums, archives, libraries are trusted stewards of cultural resources. Deploying AI systems that extract community heritage, displace professional expertise, or optimize for commercial metrics over educational mission betrays that stewardship.

Conversely, participatory cultural AI governance can strengthen democratic practice. DiCultHer’s methodology demonstrates: when communities co-govern digital heritage, they develop civic capacities—critical thinking, collaborative decision-making, technological literacy—essential to democratic participation.²⁹

6.2 The Epistemic Stakes: What Is Lost Without Cultural Governance

Technical AI governance asks “Does the system work safely and fairly?” Cultural governance asks “Should this system exist? What does it make possible? What does it foreclose?”

These questions matter profoundly for cultural institutions. AI curation optimizing engagement may systematically de-prioritize challenging, experimental, or minority cultural expressions—not through bias but through optimization toward majority preferences. Over time, this shapes what gets preserved, exhibited, remembered—the material from which future generations construct cultural understanding.

Without cultural governance ensuring algorithmic systems serve diverse cultural values (not just technical efficiency), we risk what philosopher Charles Taylor calls “subtraction stories”³⁰—narratives where cultural complexity is progressively reduced to what algorithms can optimize. Cultural institutions’ fundamental mission is resisting such reduction, maintaining space for multiple, contested, incommensurable values. This requires governance frameworks recognizing culture as irreducibly complex, not technical problem to be solved.

6.3 From Governance to Institutional Transformation

Finally, cultural AI governance offers opportunity for institutional transformation beyond AI. Participatory frameworks, transparent decision-making, community accountability—these governance principles apply to cultural institutions generally, not just AI systems.

The governance challenge AI poses may catalyze broader institutional evolution toward more democratic, participatory, community-accountable cultural practice. Institutions developing robust cultural AI governance build capacities, relationships, and frameworks applicable to all institutional functions.

This transforms AI governance from defensive risk mitigation to generative institutional development. The question is not “How do we protect existing institutional authority from AI disruption?” but “How can AI governance frameworks help us build cultural institutions better serving diverse publics, respecting source communities, and supporting cultural workers?”

Institutions embracing this opportunity will define 21st century cultural governance—not by resisting technological change but by ensuring it serves human flourishing, cultural pluralism, and democratic participation. The choice is ours to make.

Author Bio

Mika (Jaeyun) Noh is a cultural strategist, researcher, and curator working at the intersection of AI governance, cultural policy, and digital cultural infrastructure. Her work explores how algorithmic systems are reshaping museums, digital heritage, and creative institutions, and how governance frameworks can ensure that artificial intelligence serves cultural and democratic values.

She operates as a global scholar-practitioner, working across universities, policy institutions, and cultural organizations. Her research and practice focus on the transformation of creativity, authorship, cultural memory, and the political economy of cultural data in the age of AI. She develops human-centered frameworks that translate complex technological systems into cultural, legal, and institutional strategies for governance.

Her professional experience includes legislative and policy work with the National Assembly of the Republic of Korea, the Ministry of Culture, Sports and Tourism, and the Seoul Metropolitan Council, where she contributed to cultural policy and public-interest regulation. In parallel, she leads media art and digital culture initiatives through Space Ba and collaborates with global platforms such as Niio Art, connecting artists, technologists, and institutions across regions.

Her work has been presented at international forums including the PAIRS Conference, the World Design Congress, and ENCATC. Through research, exhibitions, and policy engagement, she focuses on building global frameworks for governing AI in cultural systems, positioning culture as a critical domain in shaping the future of artificial intelligence.Contact: Mika@niio.com

Footnotes

¹ Based on author’s direct observation and interviews with National Museum of Korea staff, February-May 2024. See also coverage in Korean media: “AI Curator Sparks Debate at National Museum,” Korea Herald, 15 April 2024.

² European Parliament and Council, Regulation (EU) 2024/1689 of the European Parliament and of the Council on Artificial Intelligence (AI Act), 13 June 2024.

³ Republic of Korea, Act on the Promotion of Development and Use of Artificial Intelligence (AI Basic Act), Law No. 19453, enacted 9 January 2024.

⁴ On environmental governance as extending beyond engineering, see Elinor Ostrom, Governing the Commons: The Evolution of Institutions for Collective Action (Cambridge: Cambridge University Press, 1990).

⁵ Methodological note: Institutional ethnography fieldwork was conducted 2023-2026 including participant observation at National Museum of Korea, National Museum of Modern and Contemporary Art (MMCA), Ministry of Culture Sports and Tourism (MCST) AI ethics committee meetings, and AI Art Forum events. Semi-structured interviews with 47 cultural professionals (curators, educators, archivists, artists) across Korean institutions. DiCultHer analysis based on published materials, webinar attendance, and correspondence with network coordinators.

⁶ European Commission, Ethics Guidelines for Trustworthy AI, High-Level Expert Group on Artificial Intelligence, April 2019.

⁷ UNESCO, Recommendation on the Ethics of Artificial Intelligence, adopted November 2021, https://unesdoc.unesco.org/ark:/48223/pf0000381137.

⁸ EU AI Act, Article 6 and Annex III classify AI systems in cultural sectors as potentially high-risk depending on application.

⁹ Analysis based on full Korean text of AI Basic Act and author’s interviews with MSIT officials, January-March 2024.

¹⁰ Interview with MSIT AI Policy Division official (anonymized per research ethics), 14 February 2024, Seoul.

¹¹ National Museum of Korea case details from author’s fieldwork, corroborated through museum internal documents (shared confidentially), staff interviews, and media coverage.

¹² Antonella Poce et al., “Creating Digital Culture by co-creation of Digital Cultural Heritage: the Crowddreaming living lab method,” Umanistica Digitale 8 (2020): 89-103.

¹³ Ibid., 95-97.

¹⁴ Nick Couldry and Ulises A. Mejias, The Costs of Connection: How Data Is Colonizing Human Life and Appropriating It for Capitalism (Stanford: Stanford University Press, 2019).

¹⁵ Getty Images (US), Inc. v. Stability AI, Inc., Case 1:23-cv-00135 (D. Del. filed Feb. 3, 2023).

¹⁶ Nello Cristianini, The Shortcut: Why Intelligent Machines Do Not Think Like Us (Boca Raton: CRC Press, 2023), 67-89.

¹⁷ Based on author’s analysis of MCST internal guidelines (Korean language), obtained through Freedom of Information request, August 2024.

¹⁸ Rijksmuseum Amsterdam, “AI and the Collection,” institutional documentation, accessed via https://www.rijksmuseum.nl/en/research/our-research/digitisation-research, January 2025.

¹⁹ National Museum of Korea internal evaluation report (confidential, shared with author), June 2024.

²⁰ Based on author’s analysis of Italian Ministry of Culture digital heritage guidelines and DiCultHer network practices.

²¹ Poce et al., “Creating Digital Culture,” 98-100.

²² Comparative analysis conducted through author’s participation in Korea-EU digital heritage dialogue, European Commission DG EAC and Korean MCST, November 2024.

²³ Getty Images v. Stability AI complaint details publicly available legal filings.

²⁴ Framework proposal builds on Creative Commons licensing models adapted to AI training context. See Lawrence Lessig, Free Culture: The Nature and Future of Creativity (New York: Penguin, 2004) for foundational principles.

²⁵ Comprehensive account based on author’s ethnographic fieldwork, National Museum of Korea, February-August 2024.

²⁶ MMCA case study from author interviews with MMCA staff and MCST committee members, September-November 2024.

²⁷ DiCultHer analysis synthesizes Poce et al. 2020, DiCultHer network publications, and author’s attendance at DiCultHer webinars 2024-2026.

²⁸ Scaling insights from comparative analysis across DiCultHer projects documented in network publications and author interviews with coordinators.

²⁹ On cultural participation as democratic capacity building, see Carole Pateman, Participation and Democratic Theory (Cambridge: Cambridge University Press, 1970).

³⁰ Charles Taylor, A Secular Age (Cambridge: Belknap Press, 2007), 22-26.

Bibliography

Couldry, Nick, and Ulises A. Mejias. The Costs of Connection: How Data Is Colonizing Human Life and Appropriating It for Capitalism. Stanford: Stanford University Press, 2019.

Cristianini, Nello. The Shortcut: Why Intelligent Machines Do Not Think Like Us. Boca Raton: CRC Press, 2023.

European Commission. Ethics Guidelines for Trustworthy AI. High-Level Expert Group on Artificial Intelligence, April 2019.

European Parliament and Council. Regulation (EU) 2024/1689 on Artificial Intelligence (AI Act). 13 June 2024.

Getty Images (US), Inc. v. Stability AI, Inc. Case 1:23-cv-00135 (D. Del. filed Feb. 3, 2023).

Lessig, Lawrence. Free Culture: The Nature and Future of Creativity. New York: Penguin, 2004.

Noh, Mika. “Cultural Design as Planetary Policy: Embedding AI, ESG, and Media Art into Climate-Responsive Governance.” World Design Congress London 2025 – Selected Papers, 2025.

Noh, Mika. “Explaining Inclusion: How Philanthropy Can Make AI and Digital Culture Work for Everyone.” Alliance Magazine, February 2026.

Ostrom, Elinor. Governing the Commons: The Evolution of Institutions for Collective Action. Cambridge: Cambridge University Press, 1990.

Pateman, Carole. Participation and Democratic Theory. Cambridge: Cambridge University Press, 1970.

Poce, Antonella, et al. “Creating Digital Culture by co-creation of Digital Cultural Heritage: the Crowddreaming living lab method.” Umanistica Digitale 8 (2020): 89-103.

Republic of Korea. Act on the Promotion of Development and Use of Artificial Intelligence (AI Basic Act). Law No. 19453, 9 January 2024.

Taylor, Charles. A Secular Age. Cambridge: Belknap Press, 2007.

UNESCO. Recommendation on the Ethics of Artificial Intelligence. Adopted November 2021. https://unesdoc.unesco.org/ark:/48223/pf0000381137.