hanna-hofmann.com

Introduction

When I was 12, I got my first netbook – a term used in the early 2000s to describe a small, portable laptop. This small device had a webcam and when there’s a camera, there’s a way to make a movie. In my case, this meant that I started carrying this device around as if it were a professional camera, producing short films about eating snacks or painting dolls with nail polish. Now imagine that I shared all these videos online, and that one of the big data companies scraped this data to include in the training dataset for an AI system. This system now generates new content based partly on that input. Suddenly, my very personal experience – a memory that I am recounting here – became the basis for an algorithm that can produce things at an unprecedented speed, using my past to fuel it.

If we look at the current media landscape it’s not that far-fetched to imagine a story being written right now that goes exactly like this. We live in a world where data agency is shared the moment you decide to share something about yourself or your work online – even if you don’t explicitly consent to this. Huge amounts of the Data that are fed into Large Language Models (LLM) were never intended to serve this purpose. Moreso, the way AI works means, that this Data is not simply stored and retrieved from an archive, like you would search for a book in the library, but this data becomes the basis for creation of something new: “AI memories are generated rather than retrieved” (Hoskins, 2024, 3).

This process of transforming the past into something new is not only reminiscent of the way human memory works – fluid, reconstructive, and constantly reshaped through remembering – but also signals the emergence of a new form of memory. As Hoskins argues, AI generates memory-like outputs unlike anything humans have previously encountered, raising urgent questions about how we maintain agency over our past in an increasingly automated future.

Individual Memory and Cultural Memory

Memory is something that, most of the time, does not have a body in the physical world. It exists as an inner capacity – our ability to retain and recall a past we have personally lived through. This is the domain of individual memory, and it is central to how we construct a sense of self. Humans differ from most other living beings not merely because we can remember events, but because we use memory to build a continuous narrative of who we are. While many animals are capable of retaining information, their memories do not form the basis of a reflective identity in the way human autobiographical memory does.

This so called individual memory is therefore personal, embodied, and deeply tied to perception, emotion, and bodily presence. It is also remarkably fragile. Cognitive research shows that each act of remembering subtly reshapes the memory itself, influenced by the context in which it is recalled, as well as by desire, emotion, or social cues (Echterhoff, 2008). Due to its dynamic and reconstructive nature, individual memory is rarely stable or fixed and is never identical from one person to another. It is a living process rather than an archive: shifting, selective and intertwined with the present.

In contrast to this anthropology also defines broader memory systems by looking at how groups of people remember. Maurice Halbwachs introduced the term collective memory to describe the ways in which individual recollection is socially shaped and framed within living communities. Collective memory is situated, interactive, and limited to the time span of contemporaries who share experiences and communication (Halbwachs 1985, 22). Jan Assmann later refined this framework by distinguishing two forms within what Halbwachs had called collective memory: communicative memory, which comprises the short-term, everyday memories circulating within social groups; and cultural memory, which stabilizes meaning across generations through symbols, rituals, monuments, texts, and “institutional carriers” such as priests, griots, archivists, and teachers (Assmann 1992, 48–56; Assmann 2011, 17).

As my artistic project focuses on the more long-term and symbolically mediated transmission of memory through stories, artefacts, and rituals rather than the fleeting, interpersonal dimension of group memory, I adopt cultural memory as my primary analytical term. At the same time, my understanding is informed by – and indebted to – research on collective memory, since many foundational concepts concerning the social framing of recollection, the role of shared narratives and the communicative basis of remembrance, remain essential for understanding how cultural memory emerges and operates.

Our human memory is profoundly subjective. Research in cognitive psychology shows that every act of remembering subtly alters the memory itself, as the brain reconstructs the past through emotion, context, and present concerns rather than retrieving a fixed record (Echterhoff 2008). A past remembered by someone is closely tied to that person’s perception of the world and is therefore shaped by their emotional reality, social environment, and embodied experience. When such subjective recollections extend beyond the individual and become shared, retold, and symbolically reinforced, they form the basis of cultural memory. In this broader context, the past becomes less of a literal record and more a narrative — sometimes even a myth — structured around the messages, values, and events that a community decides are worth preserving and passing on. Cultural memory emerges when lived experience is symbolically encoded through stories, rituals, and objects and is kept alive through continual reactivation, reinterpretation, and transmission across generations.

We currently find ourselves in a period of profound media-technological transformation, during which new modes of producing, storing and circulating memory are proliferating at an unprecedented scale, driven by accelerated digital connectivity. Digital media have expanded the scope of memory practices far beyond handwritten documents and oral storytelling. High-resolution recordings, databases, digital archives, photographs, videos and infinitely reproducible files enable us to store, replay and remediate experiences on an unprecedented scale. The scope of what can be “remembered” is expanding.

Yet, as Zierold notes, memory scholarship remains ambivalent: contemporary media systems are “examined only briefly, fragmentarily, and often with a very pessimistic view,” with fears that digitalization may cause societies to “forget their past” (Zierold 2008, 399). The empirical picture is uneven. Linguistic diversity and oral traditions are disappearing rapidly (UNESCO), while digital media simultaneously allows endangered knowledge to be archived, indexed, and redistributed. Cultural memory is not simply dying; it is being redistributed, unevenly—hyper-archived in some areas and endangered in others.
The question of how societies negotiate their past within today’s media system is one that is still being explored.

This tension becomes even more pronounced in emerging debates on AI and memory. Recent work argues that generative and agentic AI systems fundamentally reshape how memory is produced, curated, and validated. Hoskins suggests that AI’s capacity to extract, remix, and replay “shards” of human interaction produces a new synthetic memory ecology – one that destabilizes the possibility of shared reference points and undermines the conditions for collective memory formation (Hoskins 2026, 102156). AI systems do not “possess” cultural memory in a human sense, but they increasingly act upon it: they inherit large-scale cultural archives, transform them, and feed them back to users in ways that may amplify, distort, homogenize or fragment what communities remember. In this sense, AI becomes a new mediating layer through which cultural memory is stored, reassembled, and circulated, raising urgent questions of provenance, authority, and mnemonic agency.

Memory (Storage), Memory (Remembering)

Human memory has never been solely contained within the mind; it has always been extended, supported, and reshaped by the tools we create. I think this is why it is so important to rethink memory during the digital turn. Modern media isn’t simply external storage – it actively reshapes and influences our process of memory through that. As Fawns notes, whenever we write a note, record a message, or take a photograph, “we distribute some of our remembering agency,” extending memory into the world and entangling it with social and material environments (Fawns 2022, 3). In this sense, digital media have become integral mnemonic technologies that reshape the very conditions under which memory is formed, accessed, and shared. This technological shift has deepened a new “memory revolution,” (Merrill 2025, 173–176) where digital devices, platforms, and algorithms have become pervasive mnemonic infrastructures that mediate not only access to information but access to memory itself.

These developments have produced what Merrill titled as cyborgian remembrance, which is referencing Donna Haraways idea of the cyborg: a hybridization of human and machine memory in which remembering is distributed across neural, social, and computational systems (Merrill 2025, 181). While earlier technologies served primarily as memory aids, contemporary digital systems – cloud storage, social media archives, algorithmic feeds, and now AI – shape how memories are selected, retrieved, narrated, and valued. Our memory is increasingly dependent on digital infrastructures.

AI – the foil of Cultural Memory

As previously defined in this paper, cultural memory depends on the long-term institutional stabilization of meaning, supported by specialist human carriers such as priests, griots, archivists, and teachers, and anchored in durable symbolic forms, such as texts, monuments, rituals, and artefacts that enable intergenerational continuity (Assmann 2011, 16–18). Cultural memory is therefore socially shared, materially embedded, and historically oriented. It presupposes a community that remembers, a temporal arc across generations and a structured relation between past, present, and future (Erll 2008).

AI systems (largely referencing also LLMs here) essentially possess none of these characteristics. As Hoskins argues, AI fundamentally disrupts the conditions for collective memory by remixing, extracting, and recombining cultural material in ways that sever provenance, erase historical situatedness, and produce “synthetic pasts” that never existed through something called hallucination (Hoskins 2026). Rather than recalling a shared past, AI generates statistically plausible narratives from decontextualized data. These outputs lack embodied experience, social anchoring, and verifiable lineage – features that constitute the very basis of cultural memory.

Furthermore, cultural memory requires embodied, communal, and ritualized practices of transmission: commemoration, storytelling, ceremony, pedagogy, and shared remembrance. AI has no body, no community, no mortality, and no generational cycle; it does not inherit a past nor anticipate a future. Its “memory” is not memory at all but a computational recombination of patterns (Zierold 2008). In anthropological terms, cultural memory is lived and enacted, whereas AI pseudo-memory is performed without being experienced.

Zierold observes that contemporary media already accelerate the fragmentation and instability of shared memory frameworks, raising concerns that digital media may contribute to societal forgetting (Zierold 2008, 399–402). AI intensifies this fragmentation by producing an endless flow of unverifiable, individualized, memory-like content what Hoskins describes as a “black-box memory ecology” in which humans cannot trace or authenticate the origins of what appears to be remembered knowledge (Hoskins 2026).

Thus, while AI may simulate the idea of cultural memory – through narrative generation, stylistic imitation, or voice synthesis – it cannot fulfill its criteria. AI as an algorithmic systhem is not merely incapable of preserving cultural memory: it is structurally its antithesis. Instead of continuity, it produces recombination; instead of shared remembrance, it outputs individualized memory simulations; instead of authenticity, it generates plausible fabrications. AI therefore creates pseudo-memory: a surface resemblance without the cultural depth.

A puppet for a grandma

Grandma is sitting in an old wooden chair in the middle of the countryside. Her local village church is behind her, as is the bus stop which doesn’t really run a bus anymore, since the village doesn’t have enough inhabitants to justify a public transport route.

Grandma is sitting in a white cube, dressed in traditional clothes, decorated by Edelweiss, her hair freshly out of hair rollers, her face a mask of peaceful indifference.

For me as an artist who is handling the topic of AI it is important to reclaim agency and state my position when collaborating with Neural Networks. To me there is no good AI art, without a moral code or reflective handling of the material, just as I would expect every other art form to also question itself about sustainability and intent.

First, behind every step of the process is a human. In the description of my work, it is left open for interpretation whether the story that the digital double tells is generated or self architected – and the truth is, it’s both. I did info dump random conversation bits I had with my grandma into a pre-trained algorithm that I was running locally on my computer via LM Studio. The voice was synthesized using ElevenLabs, with the compositing and final audio composition done in Premiere Pro.

The digital double of my grandmother was scanned and animated in Blender by hand. In a way, I am merging a physical work into a digital one, and this is also the metaphorical core of the artwork. As humans, we have been merging with technology long before AI became widespread and user-friendly. I do not want to condemn every form of AI, as I believe it has a place in artistic practice. The question, however, is how that place can be shaped responsibly: not through environmental exploitation, the erosion of provenance, or the extraction of human experience without consent, but through critical engagement with the technologies we create and use.

Black Box Memory is therefore not an attempt to replace human memory with machine-generated narratives. Rather, it stages an encounter between the two. By allowing an AI-generated voice to speak through a digital representation of my grandmother, the work makes visible the tensions between lived experience and synthetic generation, between remembrance and simulation, and between cultural memory and its computational approximation. The installation asks what is gained, what is lost, and who retains agency when technologies increasingly mediate our relationship to the past.

Conclusion / Discussion

The notion of listening to an elder recount stories from the past may seem outdated in an age where a search engine or AI chatbot can provide information instantly, often with greater speed and breadth. This paper is not an argument against these technologies, but a reflection on what may be lost when they become our primary mediators of memory.

If provenance becomes obscured, the consequences extend beyond cultural memory to individual memory and, ultimately, to our sense of self. The question is therefore not whether AI can generate convincing narratives of the past, but how societies can maintain agency over what is remembered, preserved, and transmitted. As generative systems become increasingly embedded in everyday life, these concerns form part of a broader ethical debate about responsibility, transparency, and the governance of AI technologies.

Through Black Box Memory, I explore these questions by placing a synthetic narrative alongside the embodied presence of a grandmother figure. The work does not reject AI, but examines the tensions that emerge when computational systems begin to participate in processes traditionally associated with human remembrance. In doing so, it asks where cultural memory resides when the distinction between recollection and generation becomes increasingly difficult to discern.

Donna Haraway describes the cyborg as a hybrid of machine and organism, a creature of both social reality and fiction. Perhaps AI occupies a similar position within contemporary memory culture: it is already entangled with how we record, access, and narrate the past. Yet this entanglement also demands critical reflection. How much authority are we willing to grant systems whose operations remain largely opaque? At what point does generated memory begin to displace remembered experience?

Once we delve into a black box to uncover the past, maybe it is time to step away from the screen for a moment and focus on the flesh-and-blood around us. My grandmother, for example, can also tell me crazy stories – and she doesn’t even need AI prompts for that.

References

1. Hoskins, Andrew. 2024. “AI and Memory.” Memory, Mind & Media 3: e18.
2. Echterhoff, Gerald. 2008. “Language and Memory: Social and Cognitive Processes.” In Cultural Memory Studies: An International and Interdisciplinary Handbook, edited by Astrid Erll and Ansgar Nünning, 263–275. Berlin: De Gruyter.
3. Halbwachs, Maurice. 1985. Das Gedächtnis und seine sozialen Bedingungen. Frankfurt am Main: Suhrkamp.
4. Assmann, Jan. 1992. Das kulturelle Gedächtnis: Schrift, Erinnerung und politische Identität in frühen Hochkulturen. Munich: C. H. Beck.
5. Assmann, Jan. 2011. Cultural Memory and Early Civilization: Writing, Remembrance, and Political Imagination. Cambridge: Cambridge University Press.
6. Zierold, Martin. 2008. “Memory and Media Cultures.” In Cultural Memory Studies: An International and Interdisciplinary Handbook, edited by Astrid Erll and Ansgar Nünning, 399–407. Berlin: De Gruyter.
7. Hoskins, Andrew. 2026. “AI & Collective Memory.” Current Opinion in Psychology 67: 102156.
8. Merrill, Samuel. 2023. “Artificial Intelligence and Social Memory: Towards the Cyborgian Remembrance of an Advancing Mnemotechnic.” In Handbook of Critical Studies of Artificial Intelligence, edited by Simon Lindgren, 173–186. Cheltenham: Edward Elgar Publishing.
9. Haraway, Donna. 1985. “A Cyborg Manifesto: Science, Technology, and Socialist-Feminism in the Late Twentieth Century.” In Simians, Cyborgs, and Women: The Reinvention of Nature, 149–181. New York: Routledge. Add page numbers for your quotes.