11.18. Ghosts in the Machine: AI and the Politics of Historical Memory

As artificial intelligence (AI) becomes increasingly involved in cultural and mnemonic processes, questions arise not only about what is remembered but also who—or what—remembers. This paper develops a geopolitical critique of AI and memory that foregrounds the uneven global distribution of algorithmic infrastructures, data, and agency. Building on recent calls to deblackbox AI, the paper asks: how do global asymmetries shape the hybrid forms of mnemonic agency emerging at the intersection of humans and machines? Drawing on critical AI studies and theories of digital colonialism, I argue that generative AI reproduces hegemonic forms of remembrance while marginalizing others, enacting a new “mnemonic imperialism.” Large language models trained primarily on English-language data consolidate epistemic hierarchies that silence Global South memory cultures. From a Latin American perspective, these dynamics resonate with long-standing struggles over epistemic sovereignty and the right to narrate histories that have been systematically silenced or appropriated. Yet the entanglement of human and non-human agents also invites us to imagine new, distributed, and potentially decolonial modes of remembering. The paper situates this reflection within broader debates on hybrid and posthuman agency, drawing inspiration from Indigenous epistemologies that conceive intelligence as relational and collective rather than individual. Such perspectives open possibilities for reimagining AI not only as a site of technological domination but also as a potential field of epistemic negotiation and creative resistance. Ultimately, it reflects on the ethical and political stakes of treating AI as a geopolitical mnemonic actor, calling for decolonial approaches that recognize the multiplicity of ways of remembering.

This paper discusses how Argentina, Chile and Mexico have used imaging technologies (from ID photographs and surveillance cameras to biometric databases, drones, and AI systems) to monitor, control, and persecute political activists from the 1950s to the present, and how these images are being contested and resisted in visual counter-archives and artivism. The military dictatorships of the Southern Cone deployed identification photographs to target, kidnap, and disappear political activists. Within clandestine detention centres, perpetrators photographed detainees to file them, paradoxically creating visual records meant to remain unseen. Decades later, survivors, human rights organisations and visual artists have reclaimed these same images as legal and memorial evidence. In Mexico, continuities between Cold War profiling and current state violence are made visible in the aftermath of the 2014 Ayotzinapa disappearances, where groups such as Forensic Architecture in collaboration with the Equipo de Antropología Forense used synthetic and AI-generated imagery to reconstruct events the state sought to erase. Drawing on theories of ‘operational images’ (Farocki, 2000), ‘extractivist seeing’ (Henning 2025) and the idea that biometric images replace likenesses with likelinesses (they are ‘after-images’ produced through processes of ‘prompting’ that generate data visualisations with a photographic style) (Steyerl, 2025), the paper interrogates how past and present imaging technologies transform citizens into data, and how the absent bodies of traumatic histories and the ghosts of past perpetrators (once hidden behind cameras) intersect with today’s algorithmic spectres, from invisible data workers to online haters spreading ‘fake news’. Ultimately, the paper asks how memory studies can inform more ethical uses of technology, responding to Hito Steyerl’s (2025) call to make visible the hidden operations behind images of power, and to imagine resilient, small-scale, and equitable technological futures beyond monopoly and exploitation.

The starting point for the project on Holocaust-related AI slop on Facebook was earlier research analyzing audience reactions to commemorative posts about individual victims published on the public profile of the Auschwitz Museum. During that research, the scholar observed that similar content began appearing in her news feed, originating from other profiles, bearing signs of AI generation, and often only loosely inspired by historical facts. Initial observations suggested that this content was being mass-produced and disseminated, characterized by low informational quality—marked by high emotional appeal but lacking verifiable details—and accompanied by AI-generated images. This led to the formulation of several research questions regarding the nature of the phenomenon: What is Holocaust AI slop? (What narratives and visual materials are used to construct such content?) How does it spread on Facebook? (Which profiles distribute it, and at what scale?) How do users respond to this content? (Including private individuals, unverified public profiles, and institutional accounts such as the Auschwitz Museum?) And finally, what are the potential long-term social consequences of this phenomenon? To address these questions, the study employed several netnographic research methods, including algorithm training, netnographic observation, and content analysis. After training the algorithm, several hundred screenshots of Holocaust AI slop content were collected over six months (from May to October 2025), along with a comparable sample of user reactions and comments. Analysis of the collected material led to a characterization and typology of Holocaust AI slop content. The study of user responses and comments revealed that some audiences interpret such content as factual historical accounts, while critical voices are relatively scarce and receive limited visibility. As a result, this phenomenon may have significant disinformative and manipulative potential, with serious social consequences—particularly in the realms of education and historical policy.