Artificial Intelligence and Heritage Management

Definition

Artificial Intelligence (AI) in heritage management refers to the use of computational systems for tasks conventionally associated with human intelligence. These range from pattern recognition to prediction and language processing, among other dimensions such as image analysis or decision support, across the cultural and creative industries. AI applications are commonly applied in daily settings, namely for site management and visitor experience enhancement. Beyond its multiple technical functions, AI acts as a mediating infrastructure that influences how heritage knowledge is produced. Consequently, AI raises epistemological, ontological, ethical, and political questions for the future of heritage management. This entry takes into account the interconnected concepts of authenticity, authority, representation, and participation.

1. Introduction

Artificial Intelligence is rapidly affecting the process through which cultural assets are documented and shared with the public, which has become a topic of intense scholarly interest [1,2,3]. The idea of managing cultural resources evolved from an originally conservation-centric notion to a broad interdisciplinary field [4] that brings together intangible expressions, community outreach, economic representation, and many other social dimensions [5]. This progression is parallel to an academic movement that takes heritage as constructed and everchanging, rather than a passively inherited element of the past [6].

The term “community” used in this entry refers to the entirety of stakeholder groups involved in the production and consumption of cultural heritage. Within the wide collective of local residents, one may look specifically at source communities (e.g., indigenous peoples, descendant communities, ethnic minorities, etc. ) in a given functional relation with museum or site audiences, practitioners, and heritage professionals of different types. Such groups express diverse interests and values, forcing community participation into a continuous negotiation on concepts such as meaning and ownership, hence the interpretation of heritage resources.

A dynamic perception of heritage allows for integrated possibilities in terms of AI, namely through complex data acquisition and processing, from sites to cultural performances and archives [7,8,9]. Yet its qualitative nature also inhibits mere technological deployments and requires human investment in the production, validation, and consumption of heritage. Communities, both local and visiting, provide a cultural meaning to components of everyday life, and this is not spontaneously captured by AI solutions alone [10].

The underlying paradigms and their ongoing shifts are conceptually analyzed below. A growing use of AI in heritage management raises important questions about how cultural knowledge is mediated. In short, how is cultural knowledge to be processed, and what are the ethical and socio-political implications of AI instruments in heritage management? Progressively integrating AI into cultural heritage domains also means that, apart from operational advantages, the very conceptualization of collective memory is transformed [11]. Contemporary scholarship acknowledges that mediation through technology is moving beyond a function of interpretation and transforms heritage [12,13]. In other words, historical consciousness becomes partly shaped by algorithms that manipulate cultural data in ways that remain opaque to end users.

As a result, societal memory is processed mediated through predictive digital systems. Most daily life environments are unproblematically data-driven, but tensions do arise when communities are forced to negotiate between their very specific emotional memory and representations optimized by AI [14]. The challenge lies not in categorizing and optimizing heritage preservation with computational support [15,16], but in determining the sort of cultural translation that ensures convenient heritage representations. These originate in experiential, non-digital meanings and their dataset encoding results in abstractions contrary to perceived authenticity.

With these premises in mind, AI and heritage management are examined below through a practical and theoretical lens. The discussion takes heritage as a socially constructed process before exploring the several implications of AI on cultural authority, which in turn affects knowledge production and interpretation. This part is followed by an overview of key AI applications in heritage management: computer vision, machine learning, natural language processing, and digital reconstruction. The entry then considers the ramifications for responsible governance.

2. Theorizing Heritage: A Constructed and Contested Concept

2.1. Heritage as a Process

Common ground for several variations in constructivist theory [17,18] is the absence of a single objective reality and the acceptance that social constructions can be manifold. It justifies a processual approach to heritage, hence its understanding as a dynamic process. Conversely, this academic trend deconstructs materialist views on monuments and objects as vehicles of inherent value [19]. Present scholarship, especially gravitating towards Grounded Theory [20,21,22], argues for a performative meaning that is conditioned by power structures and the expression of social values.

As such, heritage exists only as a deliberate selection of references to the past, however recent, subject to institutional embodied mediation [23]. Preservation becomes decided by dominant, consensual narratives, whereas alternative angles are not displayed, or only inconveniently so. This curatorial dilemma has been debated for a long time and produced polarized views on issues such as representativeness and contested history [24] but the introduction of AI heightened concerns on the validation of certain heritage components at the expense of others, established not by communities but by algorithmic systems [25].

Central to this reconsideration is that heritage can only stem from social negotiation and institutional validation [26]. It is not fundamentally embedded within objects or practices, which need cultural and ideological layers to acquire a special meaning. They are intersected by a temporal dimension because the past is continuously reinterpreted in light of present concerns, which undergo rapid change [27]. Such a dynamic assessment of heritage is problematic when AI processes a set of layers for future data training. The subjective aspects of heritage are classified into categories and models that may incorrectly come over as objective [28] and that are therefore detrimental to interpretative plurality.

2.2. Cultural Authority and AI

This apprehension connects with the notion of cultural authority [29] and with subsequent questions about the legitimacy of heritage mediators [30]. Traditionally, responsibilities have been ensured by an administrative entity, for instance, a museum or a destination management organization. By operating AI systems for translating cultural realities into datasets, the aim is to create abstraction for classification, followed by prediction, in all forms of heritage [31,32]. Such a process reconfigures not only the intrinsic complexity but especially the interpretation procedures leading to heritage commodification. A mainstream angle sees AI as one among many actors in a wider conceptual landscape, wherein agency is a shared responsibility [33]. From this viewpoint, algorithms are not neutral tools executing human instructions but, to an extent, shape and prioritize certain patterns, which requires a validation of authority.

It needs to be stated that cultural authority is not independently acquired through AI. Instead, it becomes redistributed among institutional agents and ultimately embedded into algorithmic systems. AI does prioritize information, yet the process remains dependent on a broader socio-technical network. As a result, authority emanates from an interaction between human expertise and computational systems. In this sense, data-processing systems effectively become legitimate heritage mediators [34] and introduce new human actors, such as data scientists or software developers, thus creating new levels of epistemic power. By challenging customary practice, AI becomes an entity capable of governing knowledge itself, reminiscent of Foucault’s power/knowledge dialectics [35], within a digital environment.

2.3. Epistemological Challenges for Intelligent Systems

Due to the intense permeation of AI into heritage management and the mediating challenge mentioned above, the subject of knowledge warrants a reconsideration. Fundamental differences with traditional forms of human-structured interpretation and context analysis are methodological, as AI relies on statistical inference and probabilistic reasoning. Current research is exploring the boundaries of human–algorithm relations [36,37]. Explicitly applied to the heritage industry, this technical optimization begs the question of what counts as knowledge in the first place. Heritage, as a social construct, cannot be defined truthfully through computational modeling alone [38]. Human oversight remains indispensable within AI-supported heritage systems, wherein professionals and communities provide contextual interpretation. A symbiotic interaction is essential for evaluating cultural significance in ways that statistical inference alone cannot achieve. Human-in-the-loop governance therefore secures the interpretive plurality that heritage builds on. That is to say, any application of AI to heritage management is subject to a critical epistemology that assumes not only the benefits but also the limits of algorithmic approaches and their subordination to the human condition [39].

The epistemological ramifications are manifold, as the interpretation of heritage has, until recently, been based on cultural-historical empathy within a context of group dialog [40]. The absence of consciousness in current AI systems cannot yet be compensated through probabilistic correlations and lacks true semantic cohesion, notwithstanding the increasingly efficient improvements in language processing. Trust and reliance on non-human intelligence carry the risk of sterile empiricism [41], overlooking the fabrication of heritage.

2.4. Posthumanism and Human Decentralization

Posthumanism refers to a set of theoretical perspectives questioning that humans are the exclusive source of knowledge. Given that AI systems commonly participate in the production of knowledge, heritage itself can be interpreted through the angle of posthumanist theory [42]. Following the principle that the human subject is not necessarily central to knowledge, or even action and experience, heritage management systems might perform adequately even by marginalizing human authority. It is true that, even today, AI provides non-human cultural creation. The immediate question then relates to the root of community-based heritage, namely a balance between commodification, on the one hand, and the concept of ownership, on the other, within the frame of synthetic authenticity. This concept [43] refers to the perceived authenticity of a digitally mediated experience. Heritage consumers may encounter a convincing virtual representation that is fully created and mediated by technology, potentially compromising historical truth.

In posthuman analysis, human decentralization is not inherently problematic because agency is shared with technology, but it requires proficient digital literacy [44]. This perspective validates a cognitive hybridity between man and machine [45] and, as such, cultural relevance does not need to emerge exclusively from a human origin. However, if AI-fabricated and validated heritage becomes unproblematic, ethical questions will emerge regarding structural connections with the history of communities.

2.5. From Digital to AI Heritage

Concerns about digital mediation or authority are often associated with AI but, in fact, they stem from earlier discussions related to a complexification of digitized heritage. The CIDOC Conceptual Reference Model (CRM), Categories for the Description of Works of Art (CDWA), and the Europeana Data Model (EDM) are examples of late twentieth century frameworks created to facilitate interoperability for heritage information [46]. AI currently magnifies these earlier developments due to computational power and resulting predictive capacity, rather than introducing new challenges. The awareness of a historical continuity is important for situating AI within a broader digital transformation in the heritage sector.

Theoretical questions on the evolution of digital authority and authenticity are important, but practice shows that AI is already used intensively by the heritage industry. Mainstream applications are found across museum operations, consumer experience, and strategic planning. Computer vision techniques, and machine learning models in particular, are commonly used in conservation and restoration. As detailed below, archives and oral history initiatives also benefit from semantic analysis leading to automatic transcription and translation. Visitor-centered integration aims at the public interaction with search engines and chatbots, on the one hand, and experiential interfaces, often personalized, on the other. Such systems have been complementing traditional curatorial mediation.

3. An Epistemic Infrastructure

3.1. The Datafication of Cultural Heritage

Heritage datafication refers to a process through which complex sociocultural events are transformed into a machine-readable format, so that they can be computationally processed. The resulting advantages are straightforward in terms of large-scale analysis, yet cultural meanings suffer from simplification and, in addition, are disconnected from their specific historical and social contexts.

In heritage management, AI is best understood not only as a tool but also as an infrastructure. This means that it underpins sociocultural realities [47] and shapes them in the form of transport, communication, or interactive digital systems. Accordingly, AI becomes an epistemic network that directly conditions data collection and processing, while simultaneously affecting public accessibility to digital mediation [48,49]. The infrastructural take on AI-heritage relations needs to acknowledge the dynamics of asymmetric power over systems and platforms. Not all heritage institutions count on the means to implement AI solutions and, when they do, are forced to delegate major tasks to technological actors external to the institution. The process of global corporations determining AI models for application to unique local heritage has been classified by some as digital colonialism [50].

Still, quantifying heritage represents a primary task of AI, and both tangibles and intangibles are today embodied in the form of datasets to be used for computation. Heritage datafication does allow for improved analysis but simultaneously demands abstraction of complex cultural meanings [51]. As these need to fit prelabelled categories, they are bound to lose some degree of nuance and context. This challenge is not entirely new, as mentioned above. Heritage institutions have long relied on classification and interoperability systems for structuring a fair representation of heritage without compromising on expanding scale and automation. By becoming variables useful to AI, data-driven heritage representations might not be fully accurate, though. In addition, the individual consumer aspect increasingly expects some personalization of the heritage experience. Customized versions of heritage are already available in the form of Augmented and Virtual Reality (AR and VR) and interactive storytelling [52].

3.2. Ontological Consequences

These developments reflect a broader societal trend in which complex phenomena are translated into machine-readable formats suitable for digital platforms. As many community expressions are difficult to encode, and digital legibility is key to the experience, the preservation of heritage is not automatically ensured through more integrated synthetic intellect, although contextually useful in situations such as site interpretation [53].

Regardless of their specific expression, digital representations of heritage originate new forms of existence. This consequently means that heritage can be reconceptualized and generate interactions not feasible with traditional, more labor-intensive instruments. Such a shift blurs the boundaries of perceived authenticity and often creates a blended “phygital” experience [54]. When the singularity of heritage gives way to replicability and disconnected ubiquity, that is, to heritage becoming accessible synchronously as digital renderings, public perception is shaped by non-original factors.

4. AI Technology and Theory

4.1. Predictive Epistemology

The introduction of machine learning sets the framework for a predictive heritage dimension [55]. Heritage management no longer depends on linear data of past performance for key dimensions such as conservation or visitor flows [56] and can increasingly rely on robust future scenario building. The AI processing of unstructured heritage variables [57] adds prospective knowledge to the traditional retrospective methods.

This is closely connected with the application of computer vision systems and robotics to heritage in order to counter traditional resource scarcity [58]. However, the identification of features and patterns is not a neutral process, as the training datasets are not, either. Embedded biases may privilege certain heritage features, such as momentarily popular esthetics, while excluding others whose absence, however, would impact the integral representation of a given heritage attraction. Many institutions have invested in large databases but lack quality curation measures to minimize AI-induced bias [59]. Progressively favoring predictive automation without guardrails is conducive to digital capitalism, in the sense that it aims at managerial optimization at the expense of components considered marginal. In such a situation, predictive epistemology can induce a self-reinforcing loop, effectively increasing heritage data inequality while appearing technically neutral.

Predictive AI also benefits the visual analysis of heritage, as computer vision systems can be trained on art historical and architectural datasets. The detection of otherwise unnoticeable details on experimental projects such as paintings [60] and buildings [61] enable comparisons between human and machine interpretations, with immediate applications in conservation or stylistic research. Cultural biases still embedded in the original training datasets remain significant, though.

4.2. Natural Language Processing (NLP) and Automation

Large-scale textual data acquisition, including optical character recognition (OCR) and handwritten text recognition (HTR) accelerated by AI [62,63], is facilitated by NLP. A complex series of documents can be processed and lead to both semantic information and discourse analysis. Further downstream, conversational systems [64] can currently function as cultural interfaces in the heritage industry, for instance, as chatbots [65] and more complex automatic speech recognition (ASR) applications [66]. A caveat is that NLP remains trained on existing corpora and therefore may amplify existing biases on heritage. Catalogs and archives, and other forms of historical record, make sense only within their very precise cultural ecosystems. They embody a distinctiveness constructed by dialects and geography, where ambiguity affects the standardization expected from predictive flattening.

The human expertise needed for fine-tuning datasets for NLP tends to be secondarized by automation. While some may trust that specialized knowledge always commands AI, the reality is that algorithmic efficiency has already taken over part of the knowledge production. Heritage discourse is altered for improved efficiency, yet the curatorial identity suffers change as well [67], by adapting to perceived consumer preferences.

5. Ethics and Politics

As stated, AI systems result from the data they are trained on, and this does not come without socio-political cost factors [68]. Because heritage reflects the values and experiences of specific communities, incomplete or unbalanced datasets may marginalize underrepresented power structures. The critical management of AI-driven solutions addresses this by adding variables on inclusive and participatory design, by enhancing rather than compromising social value [69]. Ethical concerns surrounding AI in heritage management are of different sorts and may refer to algorithmic bias and data sovereignty to social issues such as inequality and community consent. Heritage institutions can systematically frame these challenges by fostering technological innovation together with responsible governance.

For decades, heritage authenticity has remained a complicated concept and has triggered multiple discussions on standard data analytics [70,71]. While traditional questions had to do with the reduction in cultural heritage to survey results or other fallible metrics, AI now adds complexity, leading to an exploration of the limits of authenticity in the digital age. One such limit is, precisely, the technical sophistication required to deploy an AI system, which can obscure any real accountability when decision-makers within governance structures delegate responsibility to the optimizing model. Political choices are data-driven but should not solely rely on AI outputs, for the epistemological reasons mentioned above. They produce optimized market suggestions based on market profitability that can overlook heritage significance if the latter is not heavily factored into the algorithm. Management priorities may tend to support content with greater potential for exposure, thus favoring commercially attractive expressions of heritage over less visible traditions that are still culturally significant.

This matters when discussing AI data sovereignty, as control over public culture cannot be clearly defined throughout the lifecycle of an AI system. Multiple national laws and intergovernmental agreements refer to collective cultural rights and responsibilities, but datasets can be legally generated and optimized for commercial purposes. Adapting the existing protection frameworks for indigenous culture [72] is part of a wider AI investment in heritage-compliant governance structures. The affirmation of Indigenous Knowledge Systems (IKS) in relation to AI [73] has been a normalizing stance for the former. This process involves algorithmic transparency, accountability, and public participation in the validation of heritage, all fundamental to context-sensitive ethical frameworks [74].

The opacity inherent in many machine-learning systems represents one of the most persistent challenges in AI governance. The so-called “black box” problem is a metaphor for the difficulty in explaining an internal process, despite observing clear inputs and results. This is cause for concern in heritage management, as accountability becomes unclear when AI systems prioritize certain narratives over others. End users, including heritage professionals, may accept AI-generated outputs impacting culture while remaining unable to grasp the computational pathways that produced them. The resulting risk of unintended bias and hallucinations can be perpetuated by recursive training, whereby AI systems enter a feedback loop as they learn from truncated content produced by earlier models. This continuous reproduction and optimization of patterns leads toward an unnatural homogenization of heritage representation, in detriment of the original sources, a process sometimes described as an “Ouroboros effect”.

6. The Future

6.1. An Agenda for AI-Supported Heritage Management

A practical implementation of AI in heritage management [75,76] can be based on common principles. First, AI projects need to remain fully integrated with community engagement to ensure the validity of the heritage component. Second, decision-making on invasive procedures altering the expression of cultural heritage, such as conservation, interpretation, and governance require human oversight. Third, accountability is not to be delegated to algorithmic reasoning without clear instruments for scrutinizing explainability and transparency. Fourth, bias and representational imbalance need to be monitored and countered through periodical audits on datasets. Fifth, AI system design requires an interdisciplinary collaboration where scientists work with heritage professionals, policymakers, and communities. Sixth, heritage practitioners must invest in skills allowing them to critically evaluate automated systems in heritage contexts.

Historical context is to be coupled to social and technical development, which is not feasible without widespread AI literacy. It is human agency within a critical capacity building that allows for questioning AI systems, notably by identifying data bias and algorithmic hallucinations. Given the societal nature of heritage, a broad participatory alignment is a prime concern for converting AI-driven results into governance robustness [77], aimed not only at culture production but also at the consumer experience [78].

6.2. Concluding Outlook

Due to its transformative capabilities, AI has confronted both formal institutions and social groups with rapidly evolving circumstances. While applied solutions are welcome for heritage management optimization, theoretical implications need to be considered. Many existing paradigms on the conceptualization and interpretation of heritage are being challenged by continuous, data-driven knowledge production. However, this may come at the expense of higher ethical and political risk levels. It therefore matters to advance critical heritage theory specifically aimed at AI, ensuring that agentic systems remain culturally sustainable.

As AI progressively contributes not only to the mediation but also to the very production of heritage, it can no longer be regarded as a mere technical tool. Its sustainable implementation depends on transparent governance and human oversight, coupled with accountability towards the communities it serves. For this reason, it is imperative to adhere to heritage standards that, through participatory frameworks, induce an enhancement of cultural integrity.