Now more than ever: the role of museum and archival objects in an age of generative artificial intelligence Open Access

Traditionally, museums contain collections of tangible items such as geological and biological specimens and items in material culture. These objects have a tangible, ontic, as well as an intangible, ontological dimension, with the former being observable by the visitor, whereas the latter is contextual and subject to interpretation by curators in an exhibition context. The development of digital technologies has allowed curators to not only digitally document collections items and develop searchable asset databases but also to augment and enrich exhibits with visual and auditory elements, extending the ontic dimensions with interpretive sensory experiences.

The pandemic accelerated existing trends to a digitalisation of society which were driven by the digital native Generation Y (“Millenials”), and the social-media native Generation Z. Since the pandemic, the use and consumption of digital media has become pervasive in many sectors of society, from working from home approaches among white collar workers and online teaching at tertiary education institutions to on-demand online streaming of movies, TV shows and music. The lockdowns, movement restrictions and social distancing rules enacted and enforced during the SarS-Cov-2/COVID-19 pandemic of 2020–2021 forced many museums to embrace a digital pivot to serve their audiences and to maintain relevance (Giannini and Bowen, 2022). It is against this backdrop that museums have to engage in digital multi-media endeavours. At the same time, the immediacy of social media and the ready visual access to almost all of the World’s iconic natural and cultural heritage sites, including many iconic exhibition objects as “attractions”, has fuelled a demand for the superlative – which many museums feed into by hosting a succession of “blockbuster” exhibitions.

An emerging trend are immersive realities, facilitated by advances in 3D image capture as well as 3D visualisation are digital twins.

Digital twins of heritage sites and objects are highly detailed digital replicas created using technologies like 3D scanning, photogrammetry and sensors. These twins not only capture the physical appearance but also integrate data on condition, environment and usage to monitor and manage preservation.

There are a number of benefits for the creation of digital twins of heritage buildings and sites: they allow the wider public to view and study a building or site that is not easily accessible in person, because it is in a remote area or because the visitor has no financial or physical means to access the site (“digital tourism”) (Florido-Benítez, 2024; de Almeida and Boavida-Portugal, 2025); they allow the public to “access” and view aspects of the building or site that are closed to physical visitation for reasons of visitor safety, visitor impact or preservation conditions; they allow to study decaying structures and design approaches for repair (Kong and Hucks, 2023); they allow to repair, restore or even reconstruct a building or part of a site that has been damaged during natural disasters or impact of war or civil unrest (Neglia et al., 2024; Cinquepalmi and Cumo, 2022).

Digital twins of artefacts and objects of varied quality have been used in museum settings to allow visitors to examine fragile or complex artefacts from all angles without the need to physically handle these (Parsinejad et al., 2021), to develop conservation approaches (Wang et al., 2023; Marra et al., 2021) and for academic collaboration (Spennemann and Hurford, 2024). Future applications could be a repatriation of the original to its decolonised country of origin, whereas the digital twin is accessible for viewing in the former coloniser’s museum(s). The generation of digital twins of sites and the associated objects permits curators to generate virtual worlds where visitors can explore an existing, or reconstructed building (or site) and can view, handle and otherwise interact with objects with genuine provenance to that place or with historically accurate contemporary pieces, for example, the furnishings of a room (gamification of heritage) (Marques et al., 2023). A major shortcoming of digital twins of objects and sites, however, is the lack on ontic characteristics that can be experienced, such as haptic (both texture and weight) and olfactory, as well as the material composition.

While widely espoused as a “game changer”, in interpretive opportunity and enhanced visitor experience, there is also a “dark side” to generative artificial intelligence (AI). This paper will, by way of deliberation, examine the interface between diminishment in museum authority, a rise in alternative “truths” and the emergence of deep fakes. Given that this paper is a deliberation, it does not follow the standard IMRAD (Introduction, methodology, results and discussion) format of papers.

The public discourse over the past decade has been characterised by an increasing distrust in civic and professional authorities, as well as science (post-truthism’) (Kienhues et al., 2020; Lewandowsky et al., 2017) and a fragmentation and tribalisation of politics (Pildes, 2021). In part this has been facilitated by post-modernist thinking, valuing the individual construction of knowledge in academia and among the teaching profession and associated curriculum standards (Davis and Sumara, 2000).

Concomitant with the tribalisation of politics is a trend of increasing vilification of viewpoints that do not conform with the ideology of a given “tribe” (Roberts and Wahl-Jorgensen, 2021). Trust in mainstream media as sources of verified news has been falling as well in part due to the weakening of journalistic standards at the both the editor and the journalist levels (Ojala, 2021), as well as which is intensified by prominent politicians systematically labelling non-supplicant media outlets as peddlers of “fake news” (Rochlin, 2017). Facilitated by the pervasiveness of social media, “alternative truths” can be seeded into online communities through narrow-casting of messages, only to be amplified in discussion groups that act as echo chambers (Hannan, 2018; Sehgal et al., 2021; Levy and Razin, 2019).

Setting aside extreme examples of drastic politically and ideologically motivated high-level interventions, such as the 2025 US Presidential Executive Order “to remove improper ideology” from museums and outdoor interpretation (Trump, 2025), history and heritage have always been subject to interpretation and reinterpretation due to shifting ideological premises, diverging methodologies and analytical approaches, as well the discovery of new evidence or sources. The underlying tenet, however, is that all sources used for this process are not only authentic but are also generally recognised and accepted as such. This places an onus of the custodians of these sources, museums and archives to demonstrate authenticity if and when required.

While Photoshop has a long history of professional digital image editing (since, 1988), its increasing capacity at not only image editing, but image manipulation soon gave rise to the neologism adjective “photoshopped” (Bisaccia and Scarborough, 1990) and associated scepticism of the integrity of images (Zelle and Sutton, 1991; Lind, 1993). Recent versions have introduced AI supported placed complex digital editing tools (such as “context aware fill”) and placed them in the hands of consumers (Guida and Sra, 2020) – thereby seeding further doubt at the integrity, if not authenticity of many images (Levy and Vaickus, 2021).

Over the past couple of years, the effects of the longer-term trends on the erosion of reality and trust in the evidentiary value of images have been dramatically accelerated by developments in creative automated image manipulation technology, in particular text-to-image generation programmes such as DALL-E (Figure 1, Figure 2), Midjourney or Stable Diffusion and the associated deep fakes. At the same time, generative AI large language models, such as ChatGPT, Claude or DeepSeek, have been developed to such a state that they can generate text to justify incongruous images (Box 1). While Figure 1 and Box 1 suffice to illustrate a point, it should be stressed that both are “quick-and-dirty” (single-shot) generations that could be improved considerably with more elaborate prompting.

At present, generative AI, as well as the associated text-to-image generation programmes, have a number of political, ethnic, gender and age biases (Rutinowski et al., 2024; Vázquez and Garrido-Merchán, 2024; Weber et al., 2024; Wiegand et al., 2024; Currie et al., 2025) derived from the training data, as well as the subsequent training during the red hatting phase (Spennemann, 2025a, 2025b, 2025c, 2025d, 2025e, 2025f, 2025g). Not only do these biases result in stereotypes being perpetuated (Spennemann and Oddone, 2025; Spennemann, 2025a, 2025b, 2025c, 2025d, 2025e, 2025f, 2025g) but they may also be resistant even to direct modification or exclusion prompts (Spennemann, 2025a, 2025b, 2025c, 2025d, 2025e, 2025f, 2025g). The training data used for the development of generative AI large language models, such as ChatGPT and DeepSeek, are derived from fiction and non-fiction books, government documents, articles and Web pages to establish the parameters of language, whereas a considerable amount of factual knowledge has been scraped from sites such a Wikipedia (Spennemann, 2025a, 2025b, 2025c, 2025d, 2025e, 2025f, 2025g).

Some generative AI applications have the ability to search the internet (OpenAI, 2025) and can add that information to the data available for interpretation and analysis (pers. obs.). The dramatic uptake of generative AI by writers and contributors to traditional as well as social media, often with little editing or modification of the generated text, has resulted in an increasing number of webpages that contain or solely consist of AI-generated text (Spennemann, 2025a, 2025b, 2025c, 2025d, 2025e, 2025f, 2025g). Not only does this raise the spectre of “AI cannibalism”, where generative AI increasingly draws in generative AI-created text, but is also shows that generative AI models will not be immune from manipulation by mal-actors determined to flood the internet with misinformation material (Anderau, 2023).

The ever-increasing sophistication of digital image generation technology will soon pose considerable challenges to differentiate reality from illusion. Real objects, due to their physical nature, hold singular significance in an era marked by alternative truths, deep fakes and the erosion of reality. They can help communities maintain a coherent, factual narrative, especially when truths are contested or subject to misinformation campaigns. With their physical existence and provenance, objects function as anchors of truth and authenticity. They not only enable direct, sensory and personal engagement, but importantly carry inherent credibility because their origins and histories can be traced, verified and cross-checked against factual records. Moreover, physical objects not only provide haptic (both texture and weight) and olfactory feedback unique to the object but also carry contextual information, such as marks of usage, wear and layers of historical modification.

Setting aside that planting fake artefacts needs to be accompanied by a convincing creation of a chain of provenance, real objects also carry physical characteristics that are difficult if not impossible to convincingly fake to avoid detection through forensic techniques. These include microscopic examination, carbon dating, chemical analysis or spectroscopy. While single items could conceivably be forged with great effort and attention to detail, real objects rarely occur in isolation but exist in broader contexts, such as related artifacts and cultural patterns.

The past and present role of museums, archives and libraries has been to act as guardians of real objects, thereby providing reference points and trusted sources that audiences can rely upon for verification. To facilitate this, collections policies and curatorial standards have been developed and implemented (Garry et al., 2024). In an age of doubt and misinformation, it can be posited that the significance of that role will continue to increase. This, however, also presupposes that the public continues to accept a museum as a trustworthy institution and does not question the integrity and authenticity of the museum documentation. The increasing pervasiveness of alternative truths and deep fakes, coupled with the distrust in authoritative (as opposed to authoritarian) viewpoints is poised to challenge this, in particular among the younger and digitally-native population due to their exposure to misinformation in other aspects of their daily lives.

Museums that pivoted to virtual exhibitions during the COVID-19 pandemic and that continue to rely heavily on digital representations or virtual exhibitions face a greater vulnerability to a perception of deep fake interference or misinformation. Without doubt, any digital content needs stringent authentication protocols to prevent unauthorized manipulation or dissemination of false representations of collections. When considering the authentication of collections objects, well documented provenance is crucial, based on a triangulation of independent verification methods, such as historical records, eyewitness testimony and ontic characteristics (e.g. macroscopic, microscopic and spectroscopic data). In an age where distrust in authorities will only increase, it will become essential that museums not only continue to engage in, but actively and consistently demonstrate transparency and trustworthiness through rigorous documentation, including, but not limited to provenance verification, curatorial histories of objects and interpretive context in the museum as well as the public/academic space.

The history of conspiracy theories has shown that it cannot be predicted what shape future conspiracy theories may take. Consequently, it cannot be anticipated as to which items that are held in current collections may become significant for such theories, become the focus of irrational attention and have their authenticity questioned as they do not confirm with that theory’s narrative. It can be surmised, however, that it is less likely that this will affect the iconic collections items with their deep and well researched provenance, but involve items that have an uncertain provenance or where the chain of custody is broken.

In consequence, museums also need to continue to transparently and openly communicate their research methods related to provenance verification or object characteristics, as well as to transparently acknowledge any uncertainties or controversies surrounding objects in their collections.

If in 20 years’ time someone were to question the authenticity of an object they would need to distinguish between objects and documentation that existed, and have been demonstrated to exist before the rise of deepfake technology (2014–2017 onwards) (Goodfellow et al., 2014; Kim et al., 2018) and objects that came into existence, or have been documented for the first time, since then. Critical for current collections management is that original digital files, created pre-2017, and thus containing pre-2017 metadata, are not inadvertently modified (for example by opening and resaving) erasing evidence of their pre-2017 existence.

To safeguard against future arguments of manipulation of evidence, digital photos can be taken now, with the photos authenticated by blockchain technology (Bralić et al., 2020) and digital certificates, thereby anchoring them in time and, via EXIF and other metadata (such as embedded geolocation) also in space. While objects or sites themselves cannot be authenticated via blockchain, they can be authenticated by proxy. It is possible, for example, to take a photograph of an object and then blockchain that image. The only way to withstand future claims of fabrication or falsification of objects is to implement an unbroken chain of evidence that will withstand future forensic examination. That would, for example, entail to document an excavation step-by-step, from the object encountered in situ through all its subsequent processing stages such as removal from the ground, in its uncleaned state, its cleaned state, in its various conservation phases and, finally, as accessioned with an inventory number. The formal forensic crime scene documentation process can provide guidance on how to establish and maintain an unbroken chain of evidence (Bilal et al., 2025; D’Anna et al., 2023).

Likewise documentation needs to occur as other objects enter the collection, from donation (or collection in the field) to accessioning. Objects that are already in the collection can be documented step by step, as they are undergoing cleaning or restoration, documenting the various steps of the process (and discovery of undocumented elements, if any). Each of these images should be protected through blockchain technology. In the case of archival files it may well involve block-chaining each individual page scan to avoid future claims that a file has been tampered with.

Without doubt, such a process dramatically adds to the curatorial, or at least administerial, workload, which has resource implications that cannot be underestimated. Yet, this process goes a long way to ensure that authentication of objects can be verified at some future point in time if the need were to arise.

The geometrically accelerating pace of generative AI developments is forcing the hand of cultural institutions, from academics such as the author, to museum curators. We are literally at the cross-roads of history. The old firm concepts of interpretation and historiography, based on tangible evidence of sources, are challenged by digital factors where manipulation and falsification of data can provide “evidence” for alternative “realities”. That future appears inevitable.

On that trajectory, museums, libraries and archives, may need to expand their role as depositories of cultural memory, by explicitly addressing misinformation, deep fakes, and alternative truths as a key educational theme, empowering visitors to critically navigate digital information landscapes. Now is the time to prepare for that future. This includes, inter alia, to establish clear guidelines on the use of AI, augmented reality, and other digital tools, not only explicitly avoiding practices that could be interpreted as misleading or deceptive, but also to transparently document in detail any interpretive process from conceptualisation to delivery. While most museums are doing some of this, and some museums are doing all of this, there is a need to make this an industry-wide standard.

Moreover, now is the time to ‘future truth’ the existing collections through blockchain anchored documentation. There is a perpetual discussion as to what extent museums and other collections-based institutions should collect and curate duplicates or whether duplicates held in collections should be sentenced according to a set of rules and deaccessioned to make space for new collections material, or simply to reduce curatorial costs (Morgan and Macdonald, 2020; Ćosić, 2018). In a future age where perception will be even more detached from reality, well authenticated duplicates may gain considerable significance. They can act as sacrificial objects, which visitors can access on demand and not only view, but also handle/touch, thus individually verifying the tangible nature of these authenticated objects in all their ontic “glory”.

While it is possible to conceptualise the future relationship of AI-native generations (sensu Inayatullah, 2004) with tangible material culture and the significance authentic objects may hold in that relationship, its extent and impact cannot be quantified at this point in time. In view of current trends of an increasing alienation from nature and the disinterest in heritage exhibited by the smart-phone and social-media native generation (Richardson et al., 2018; Kesebir and Kesebir, 2017; Zhang et al., 2024), there is a distinct possibility that this or the following generation may well prefer to primarily interact remotely with objects rather than visiting a museum. In that case, as suggested by forecasting scenarios created by generative AI (Spennemann, 2025a, 2025b, 2025c, 2025d, 2025e, 2025f, 2025g), physical artefacts will continue to exist, but many will be digitally twinned, allowing visitors to interact with them through holograms and VR. In a doomsday scenario that may well lead to a collapse of museum visitation as the older, museum-native generation does out – this may well place significant pressures on funding and political support, forcing museums to justify the existence of extensive collections of physical objects. While, again, iconic exhibits will be immune, the pressure will be on all duplicates as well as most “mundane” objects in the collection.

Soon the profession will reach the Rubicon. Quo vadis?