Mediating memory through immersive technologies and artificial intelligence: revisiting individual and collectively co-created pasts and futures

Introduction

Extended Reality (XR), encompassing immersive technologies, has transformed the ways in which memory, identity, and history are mediated and preserved. By enabling decentralised storytelling and expanded forms of participation XR creates new possibilities for how individuals and communities engage with the past. We define XR as an expanded format encompassing multisensory, spatial, and interactive technologies that connect embodied experience with narrative agency. As Slater (2009) argues, immersion is a factor of objective technology, whereas presence involves a willing perceptual relocation of consciousness into the designed spatial world. In XR experiences memory becomes both an artifact and a living process, inviting individual and community participation in constructing shared pasts and speculative futures.

Immersive environments enable memory to take material form, while co-presence generates affective insights that help build, share, and preserve memory. XR thus emerges as a powerful framework for structuring and retrieving memory. As Halbwachs (1992) theorised, individual memory is always recalled and anchored within the specific social groups and experiences to which someone belongs. XR extends these social frameworks through key affordances: embodiment, site-specificity, and real-time interaction. Virtual Reality (VR) forges shared witnessing within reconstructed spatial architectures, foregrounding the physical parameters of a memory through deeply visceral simulation. Augmented Reality (AR), by contrast, enables the layering of forgotten or marginalised memories directly onto physical sites. Through site-specific storytelling, users experience simultaneous co-presence in both the past event and the present physical landscape. Combined with AI, these dialogic loops, environmental plasticity, and real-time adaptation make visible that the nature of memory is inherently fluid – continuously rewritten rather than fixed. As Hoskins (2011, p. 78) observes, digital re-spatialisation and re-temporalisation produce both, “...a comprehensive ‘monumentalization’ of memory, powerfully fixing events in their multiple iterations within a single perspective or ‘timeline’,” while simultaneously creating accounts open to “...unprecedented scrutiny and challenge” (Hoskins, 2011).

XR and AI technologies have increasingly shaped participatory approaches to memory reconstruction in XR. Early XR works, particularly in non-fiction journalism, positioned users primarily as ‘virtual witnesses’ at the center of factual events (De La Peña, 2017; Dowling, 2021). As XR technology matured, creators designed experiences for participants that go beyond observation toward subjective, embodied, and fragmented expressions of memory. Abstract and symbolic spaces, combined with embodied cognition, shifted users from passive observers to active participants whose physical movements and affective responses became integral to the narrative architecture. Furthermore, today, Artificial Intelligence (AI)-driven adaptive environments further personalise these experiences by responding in real time to biometric and behavioural data. While this evolution enables deeper artistic expression and participatory memory work, it also raises significant ethical challenges around privacy, data ownership, and the translation of lived experience into digital forms.

As a Perspective article, this piece investigates how AI-XR enables new co-creative methodologies for reframing, reconstructing, and safeguarding memories, while simultaneously opening speculative horizons for alternate future realities. The primary contribution of this manuscript is a conceptual framework that maps AI-XR memory work across three levels: therapeutic-personal, community-cultural, and socio-political. Furthermore, it outlines a set of design commitments for AI-assisted memory reconstruction. Drawing on selected case studies, we examine both the transformative potential and the ethical complications introduced by these technologies. These specific four cases were chosen because they effectively map onto the different levels of memory mediation in our framework, offering a diverse but representative view of AI/XR use in contemporary digital heritage and artistic practice (Table 1). These case studies open questions on how critically engaged uses of generative AI actively intervene in the construction of memories and interrogate how community-led practices in XR challenge colonial control over historical memory. We propose framing AI-XR not merely as a combination of representational tools, but as speculative and co-creative methods. In doing so, we argue that technologically mediated memory can allow communities “to be believed,” grant them epistemic justice, foster restorative futures, and sustain meaningful dialogue between personal experience and collective heritage.

Table 1. Cross-case analysis of AI-XR memory mediation.

Mediating memory with XR

Participatory and co-creative approaches to memory in XR are increasingly supported by empirical research on embodiment, perception, and episodic recall. XR enables conditions that are simultaneously controlled and ecologically valid – that is, realistic enough to resemble the settings in which memories actually form – through embodied, first-person enactment in space and time (Lenormand et al., 2024; Penaud et al., 2023). The phenomenological intensity of presence plays a central role in how memories are encoded, recalled, and emotionally experienced within XR environments. Bodily enactment is not incidental to memory formation but constitutive of it. VR research has demonstrated that different VR configurations shape, in different ways, the quality of memories formed and the phenomenological sense of remembering. Iriye and St. Jacques (2021) demonstrated that encoding events from a third-person avatar perspective produces measurably less vivid, less accurate, and less spatially rich memories, with effects persisting after a one-week delay. Even more, when participants are immersed in a virtual environment and their sensorimotor feedback is manipulated, the consequences are measurable and multidimensional. Disrupting bodily self-consciousness in a virtual reality experience affects spatial and perceptual memory but also the subjective sense of remembering (Penaud et al., 2023, 2022)

Bonnail et al. (2023). argue XR's capacity to leverage perceptual richness also renders it a potential instrument for what they term XR Memory Manipulations (XRMMs) — deliberate or inadvertent interventions in encoding, consolidation, and retrieval that can distort, suppress, or artificially strengthen personal memories. Given that autobiographical memory serves self-constitutive and social functions, these manipulations carry significant implications for personal identity and for the ethical design of XR systems.

These findings validate VR as an active environment for processes of self-enactment in which memory, and consequently identity, are potentially altered and reshaped. VR does not merely represent memories, but, as a performative medium, actively intervenes in their formation and recall, it shapes the encoding, modulates the phenomenological sense of remembering, and can strengthen or distort what is later recalled. This positions VR as a medium uniquely capable of supporting the recreation, imagination, and enactment of memories, including memories that are fragmented, undocumented, or have never existed in particular material or audiovisual forms. By generating immersive “states of memory” – understood here as experiential conditions in which the memories become accessible through a medium that actively mediates what is remembered, and the boundaries between recalling and reconstruction become permeable- XR can help extend or give form to experiences that exceed conventional documentary forms. Furthermore, the question of extending these “states of memory” into collective and community-based contexts remains an underexplored space. It is precisely at this intersection, where immersive enactment (propelled with generative AI) connects with participatory co-creation, that a different set of possibilities and risks begins to emerge: not merely the reconstruction of a single subject's synthetic “state of memory”, but collective histories built from multiple perspectives—collaborative synthesis of fragmented, undocumented, or contested pasts.

AI-XR and the collective mediation of memory

Today, the convergence of XR and AI is shifting how individuals and communities engage with personal and social histories. In therapeutic and spatial computing environments, dynamic AI-XR systems enable users to externalise internal landscapes (psychological, emotional, symbolic, mnemonic), transforming personal memory into shared participatory spaces. Here, physical presence in the XR environment moves users away from passive witnessing of trauma to active co-creative agency.

Operating at the socio-political level, Victoria Bousis’ Stay Alive, My Son VR (Ume Studios, 2024) exemplifies this shift. The XR memoir reconstructs Pin Yathay's experiences of surviving the Cambodian genocide by allowing users to embody the father as a hyperrealistic digital human. Traversing symbolic phases of grief after being forced to abandon his six-year-old son, users undertake an interactive pilgrimage. Rather than observing historical events from a distance, participants encounter survivor testimony as a visceral, embodied experience. AI is used to track participants’ gaze at crucial points in the narrative to ensure that the digital avatar of Yathay seems to directly engage with the player at all times. This AI-integrated approach to XR fosters affective empathy. As Bousis (2025) notes, “I wanted to use technology in a way that makes people more emotionally vested—so they could become active participants of change.”

Situated at the community-cultural level, Kasia Molga's How to Find the Soul of a Sailor (Molga (2025b)) uses AI as a speculative and co-creative method. By training generative AI on her late father's handwritten nautical journals and integrating real-time climate data, Molga projects his voice and maritime memories fifty years into the future. Rather than attempting literal resurrection, the work creates a metaphorical exploration of memory, loss, and ecological change within an AI-driven immersive environment. Individual memory is thus transformed into collective reflection and shared agency. As Molga (2025a) explains, “I use AI not as a gimmick, but as a bridge–a way to reach across time.”

Together, these works illustrate how AI-XR systems increasingly transform memory from a static record into an adaptive, relational, and participatory process. In these contexts, memory is not recovered in isolation but emerges through conversational feedback loops mediated with AI. Generative AI can visualise fragmented or undocumented personal histories by iteratively refining images and sounds until they align with the subject's internal image. This process operates, in the terms proposed by Bañuelos Capistrán et al. (2025), as a form of affective symbolic documentalism: the generated image does not claim historical accuracy but acquires testimonial value precisely through its acknowledged constructedness, functioning as what Barthes (1981) calls a synthetic punctum — an image that provokes genuine emotional contact with silenced or lost personal histories. During the experience, narrative agency becomes distributed among the AI system, the original creator, and the audience.

The interactive exhibition Synthetic Memories: Live Reconstruction (Domestic Data Streamers, 2024) exemplifies the therapeutic-personal level of our framework. In this project, testimonies engage in dialogue with a human interviewer and an AI system that helps to reconstruct visually undocumented memories. The project was developed within the context of initiatives designed to help people with dementia or other cognitive impairments to recall personal memories. The historical photographs generated as the result of the interaction loop with the audience are intentionally abstracted or blurred to support personal reminiscence without claiming documentary accuracy. By telling their stories and self-engaging with the creation of their memories, participants may experience positive cognitive and emotional affects, alongside forms of personal and collective validation.

In the related civic-themed pop-up installation Citizen's Office of Synthetic Memories (Domestic Data Streamers, 2024), a public municipal office was recreated where citizens/audiences could walk-in (Figures 1, 2), reconstruct lost personal histories with AI assistance, and receive printed physical photographs. Through these processes, deeply personal memories are transformed into tangible media artifacts that wider audiences can encounter, validate, and archive. Histories that would otherwise disappear are rendered visible, shared, and archived. Synthetic media thereby becomes a method for taking personal stories forward and fostering intersubjective comprehension (Abes et al., 2026). The Synthetic Memories project has also been extended into a fully navigable virtual environment, converting AI-generated images into explorable 3D spaces. The project demonstrated that spatial immersion amplifies the affective impact of synthetic memory reconstruction. Participants reported heightened presence and emotional resonance when static images were transformed into environments they could inhabit and navigate, with one participant describing the reconstruction of a Hurricane Katrina memory as a “transcendent and healing experience” (Abes et al., 2026). Crucially, however, the immersive expansion also revealed new tensions: sound design incorporating AI-generated voices raised participant expectations that the technology could not consistently meet, underscoring the ethical and perceptual limits of simulating intimate personal histories through algorithmic mediation.

Un archivo inexistente (Rivas San Martín, 2022) is a series of fictional photographs depicting working-class same-sex couples, queer and non-binary individuals set in early twentieth-century Latin America (Figure 3). Images were created using generative AI. The project uses AI to fabricate a fictional photographic archive of queer life in Latin America, a record that heteronormative culture structurally prevented from existing. Operating at the socio-political level of intervention, the project responds to what its author describes as a defining consequence of heteronormative culture: that the queer experience of the past could not leave any record or archive (Rivas San Martín, 2022).

Memories that do not exist in our mainstream cultural archives represent an important challenge for both individual and socio-political future forms of human-AI memory ecologies. Because of the absence of a visual photographic or audiovisual archive, AI models are not trained on these images and consequently are unable to generate them — as if these instances of the past did not exist in their latent space. Rivas San Martín (2022) contests this situation by developing what he calls “minority prompts” — a set of deliberately designed instructions that counteract the patriarchal and Western-centric biases embedded in AI training datasets. The results were unexpected and imperfect, but rather than concealing the constructed nature of the resulting images, the project preserves their imperfections as ethical markers and red flags: blurred details and anachronistic elements are visible signals that these images are not evidence but vestiges of an undocumented past that acquire testimonial value precisely through their acknowledged fabrication. This project reorders the past by proposing a critical reinterpretation of colonial visual regimes, and in doing so activates political questions of what can be seen, represented, and remembered.

Speculative futures and AI-XR reconstruction

Generative AI-XR systems synthesise multimodal data to bring coherence to fragmented, undocumented, or at-risk memories. By enabling real-time collaboration and embodied interaction, these systems align with participatory design principles and support “storyliving” — a process in which participants actively shape narrative architectures (Duguleana, et al. 2026). AI-XR also opens space for imaginative collaboration on “what if” scenarios. When combined with forensic, archival, and diagnostic frameworks, such processes facilitate worldbuilding from multiple perspectives (Ma et al., 2025; Yu, 2025). Particularly, in climate change and ecological storytelling, these approaches offer a way to address anthropomorphism by incorporating multi-species outlooks into the design of environments and landscapes, thereby broadening speculative memory work beyond purely human-centered narratives.

In the context of XR-AI experiences, speculative worldbuilding becomes an extension of memory reconstruction, allowing communities not only to revisit the past but to imagine alternative futures emerging from it. This positioning of AI-XR as a co-creative and speculative method opens important creative and political possibilities. It creates significant opportunities for intergenerational transmission, enabling younger audiences to inhabit memories actively and develop affective continuity rather than receiving them passively. However, these affordances also raise critical concerns. Most AI-XR tools emerge from Global North infrastructures, meaning that reframing Global Majority memories without genuine local control risks reinforcing digital colonialism (Ávila Pinto, 2018; Couldry and Mejias, 2018; Vijayakumar, 2024). Algorithmic and techno-cultural biases compound these issues. As Shan (2025) observes, dominant techno-cultural imaginaries often frame non-Western XR developments as derivative or lagging, thereby reinforcing North-South hierarchies. The obscuring of human labour in data training and model maintenance adds another layer of ethical complexity. Moreover, not all users enjoy equal agency, raising the danger of performative rather than substantive participation.

While immersion and co-presence can democratise access to history and personal past events, they also risk turning trauma into spectacle by collapsing critical distance, a critique identified in earlier immersive media scholarship (See Madary and Metzinger, 2016; Nash, 2018). The blurring of fact and simulation may exacerbate inaccuracies, undermine authenticity, and produce false recall. Highly affective experiences further heighten the potential for re-traumatization, particularly when consent protocols are inadequate. Further, affective proximity to the real can blur distinctions between fact and fiction, encouraging an over-attribution of authenticity to immersive experiences (Kalpokas, 2020; Kapur and Ansari, 2026).

Immersion is not limited to technological hardware. It also arises through the process of co-creation and the cognitive states that such experiences trigger. Even without headsets or fully digital environments, deep engagement can produce a powerful ‘illusion of nonmediation’ (Lombard and Ditton, 1997), generating strong narrative immersion. Ultimately, positioning XR, AI, and AI-XR as co-creative and speculative methods is theoretically productive, but risks techno-optimistic overreach unless tempered by robust participatory ethics and community governance over both process and output.

Discussion: AI-XR and future memory archives

Immersive technologies function as expanded media environments that enable diverse approaches to memory preservation and transmission. When mediated through XR and AI, personal memories become sites of dialogue, opening conversations and pathways from individual experience to collective understanding. By positioning AI and XR as speculative and co-creative methods rather than merely representational tools, this perspective contributes to understanding how co-creative immersive methodologies can bridge individual memories and cultural heritage with broader social and political memory work and futures, while underscoring the need for ethical frameworks in AI-assisted memory reconstruction.

We examined power asymmetries between human creators, AI systems, and participants and risks of bias; epistemic and ontological issues such as false recall and undermining the historical weight of trauma as well as ethical and material unevenness in how different users experience agency. The synthesis of individual and collective memory practices points to the importance of participatory ethics and new models for digital heritage preservation.

For an HCI readership, these cases reveal necessary design implications. First, designers handling consent in AI-assisted memory reconstruction must implement dynamic, ongoing consent protocols, particularly for vulnerable populations. Second, synthetic images should often remain visibly imperfect (as demonstrated in Rivas San Martín's work) to avoid false recall and visually signal their constructed nature. Third, to avoid trauma spectacle, systems must prioritize user agency and emotional scaffolding over hyper-realistic simulation. Finally, robust community governance over prompts, models, and data outputs is required to ensure epistemic justice and prevent the extraction of marginalized heritage.

Drawing on Bañuelos Capistrán et al.’s (2025) research on AI and affective memory, the AI-XR case studies examined across this article help identify new paradigms of mediated memory at three distinct but interconnected levels, each with its own affordances and ethical stakes, and each requiring a different set of design commitments to fulfil its promise without reproducing the exclusions it seeks to address.

At the therapeutic-personal level, the co-constructed image-making of Synthetic Memories — and its extension into navigable VR environment — demonstrates that AI can function as a prosthetic method for memory activation and intervention, enabling a mode of embodied reenactment of the past that current documentary forms and media-participation practices cannot provide. The empirical evidence reviewed in this article also confirms that VR does not merely represent states of memory but actively intervenes in its formation, and generates affective states that can complete, extend, or give form to memories that have no tangible material trace. This positions XR not as a playback technology of the past but as an enactive medium — one in which memory, identity, and selfhood are continuously at stake.

At the community-cultural level, projects such as How to Find the Soul of a Sailor or Un archivo inexistente enable the recovery and transmission of collective memories that exist outside official archives, generating new forms of collective authorship through which communities actively construct, validate, and transmit their own histories of the past not present in the archives. Here, the technology functions less as a representational tool of a particular state of memory than as a bridge between oral memory — the lived, face-to-face transmission of experience within families and communities — and cultural memory, the formalised and institutional construction of shared identity through the communities to which individuals belong; as a relational medium through which memory emerges from sustained collective dialogue. Technology here becomes the interface through which participants contribute their testimonies, negotiating which memories are worth materialising, and validating the generated images against their own lived knowledge of the past. How to Find the Soul of a Sailor operates at the intersection of training generative AI on an individual story (the artist's father's handwritten journals) but projecting his voice and memories fifty years into the future and creating a space in which familial memory becomes a vehicle for collective reflection on ecological loss and cultural continuity.

Finally, at the socio-political level, works such as Stay Alive or My Son VR or Un archivo inexistente demonstrate that AI-XR can function as an instrument for redistributing the narrative regimes that determine whose histories are seen, believed, and preserved. This is not primarily a question of representation — of adding marginalised voices to an existing archive — but of contesting the archive itself as a political action. Yet this redistributive potential is conditional upon AI systems infrastructures that are trained on culturally biased datasets, risking the reproduction of narratives they seek to challenge, unless communities retain governance over both process and output. At this level, the politics of the prompt i.e., who controls the instructions given to the generative system and whose cultural logic those instructions encode, becomes a significant site of intervention and consequential contestation.

What ultimately emerges from the convergence of AI and XR in recalling the past, is a new imaginative and affective territory for mediated memory: the combination of latent memory spaces of AI models with the affective memory embodied experiences that XR generates, holds the potential to produce individual and collective audiovisual encounters with the past that are entirely unprecedented — experiences that are at once personally enacted and collectively shared, technically mediated and affectively real, opening possibilities for documentary and audiovisual creation that have not yet been imagined. Ultimately, immersive memory co-creation serves not only as a means of revisiting the past but as an operational space for plural and equitable futures.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. AI assisted for copy editing.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Julia Scott-Stevenson, University of Technology Sydney, Australia

Reviewed by: Yufeng Wu, University of Technology Sydney, Australia

Justin Harvey, University of Technology Sydney, Australia