Dimensions in Testimony

USC Shoah Foundation. Dimensions in Testimony. Ongoing initiative. https://sfi.usc.edu/dit

Concepts:  Interactive biography | Survivor-authored virtual testimony | NLP testimonial retrieval | Dialogue preservation | Survivor -co-author

Keywords:  Holocaust testimony | Interactive technology | USC Shoah Foundation | Virtual survivor | AI and memory

The USC Shoah Foundation’s Dimensions in Testimony is the foundational practice-based precedent for AI-mediated interactive Holocaust testimony. Developed over many years of collaboration with survivors — each filmed for approximately a week, answering around a thousand questions — it uses natural language processing to match user queries to pre-recorded survivor responses, enabling real-time conversational interactions with virtual survivor biographies in museum and educational settings. The initiative’s most significant feature for the log is its survivor-as-author principle: survivors were not subjects of the project but active participants who shaped what questions were asked and what answers they gave, bringing their accumulated knowledge of what had always been asked, what had been missed, and what needed to be said for posterity.

“Nothing can replace the ineffable, dynamic interaction between two human beings.” (USC Shoah Foundation, Dimensions in Testimony FAQ, n.d., n.p.)

“Dimensions in Testimony is a different type of engagement meant to preserve the opportunity to have a dialogic experience.” (USC Shoah Foundation, Dimensions in Testimony FAQ, n.d., n.p.)

“In our ten years of experience with interactive biographies, we have seen users absorb history while engaging in meaningful conversations with survivors.” (Clark, in Museum of Jewish Heritage, 2025, n.p.)

AI and the Holocaust: Rewriting History? The Impact of Artificial Intelligence on Understanding the Holocaust

UNESCO. “AI and the Holocaust: Rewriting History? The Impact of Artificial Intelligence on Understanding the Holocaust.” UNESCO Digital Library, 2024. https://doi.org/10.54675/ZHJC6844

Concepts:  AI threats | Value-sensitive design | Liar’s dividend | Data voids | Stakeholder collaboration 

Keywords:  Holocaust | AI ethics | Disinformation | Memory preservation | Education policy

This commissioned report constitutes the most authoritative institutional statement currently available on the intersection of AI and Holocaust memory. It identifies five major threat categories: AI inventing false narratives, deepfake technology, manipulation of AI to spread hate speech, algorithmic bias reinforcing dominant narratives while marginalising lesser-known histories, and AI oversimplifying the Holocaust’s complexity. Against these threats, it documents genuine benefits (large-scale archival processing, interactive testimony) and issues specific recommendations to policymakers, educators, archives, AI developers, and researchers. Its central normative position — that AI development in relation to Holocaust content must involve close collaboration with survivors and descendants as design partners, not merely as subjects — is the institutional articulation of the log’s foundational claim: AI can only do with testimony what human beings enable it to do.

“The threats associated with AI on safeguarding the record of the Holocaust are manifold.” (UNESCO, 2024, p. 3)

“Tech companies must work closely with the Jewish community, Holocaust survivors and their descendants.” (UNESCO, 2024, p. 1)

“AI can organize, categorize and search the content of a testimony.” (UNESCO, 2024, p. 13)

Generative AI and the Purpose of Legal Scholarship

Smith, Michael. “Generative AI and the Purpose of Legal Scholarship”. Massachusetts Law Review, vol 21, Issue 1, February 2026.

Concepts:  scholarly writing | AI non-instrumental scholarship | metrics vs. contribution | Legal scholarship | Cognitive deskilling 

Keywords:  Legal scholarship | Generative AI | Academic writing | Law schools | AI ethics

Smith argues that an instrumental view of legal scholarship — dominated by obsession with quantity, citations, and prestige hierarchies — obscures the detrimental effects that generative AI may have on the intrinsic value of writing and publishing. A deeper look at the fundamental purposes of legal scholarship, he contends, reveals that AI poses far more of a danger than a purely instrumental perspective suggests. This article makes the structural equivalent of Burzlaff’s argument for historians, in a legal studies context: the discipline’s essential contribution — interpretation, argument, ethical engagement with complexity — is precisely what AI cannot provide, regardless of its fluency at the surface level.

“An instrumental view of legal scholarship... obscures the detrimental effects that AI may have on the intrinsic value of writing.” (Smith, 2025, abstract)

“AI poses far more of a danger than a purely instrumental perspective suggests.” (Smith, 2025, abstract)

“The fundamental purposes of legal scholarship demand what AI cannot provide.” (paraphrase of Smith, 2025)

Holocaust Memory in the Digital Age: Survivors’ Stories and New Media Practices

Shandler, Jeffrey. Holocaust Memory in the Digital Age: Survivors’ Stories and New Media Practices. Stanford, CA: Stanford University Press, 2017.

Concepts:  Mediated memory | Institutional testimonial production | Physical markers | Code-switching | Algorithmic curation 

Keywords:  Holocaust memory | Digital media | New media | Oral history | Technology

Shandler argues that Holocaust testimony is always shaped by the medium through which it is conveyed, and that digital formats do not inherently enhance authenticity. His analysis of the USC Shoah Foundation’s Visual History Archive reveals how institutional framing, indexing, and protocols mediate testimonial production — a concern especially relevant to AI-generated tools that incorporate algorithmic curation and simulated interaction. His identification of code-switching (survivors shifting to Yiddish during emotional moments) as a testimonial marker that AI cannot process illustrates Kraft’s core/narrative memory distinction from a linguistic angle: the very moments at which testimony exceeds the narrative register are the moments AI misses.

“Digital formats do not inherently enhance authenticity.” (paraphrase of Shandler, 2017, cited in Rosenthal, 2025, p. 6108)

“Testimony is always shaped by the medium through which it is conveyed.” (paraphrase of Shandler, 2017, cited in Rosenthal, 2025, p. 6108)

“AI’s inability to capture survivors’ code-switching into Yiddish during emotional moments.” (Shandler, 2017, cited in Rosenthal, 2025, p. 6108)

Can AI Make Hitler Cry? Exploring the Use of AI in Holocaust Education across Four Generations

Rosenthal, Aharona. “‘Can AI Make Hitler Cry?’ Exploring the Use of AI in Holocaust Education across Four Generations.” AI and Ethics 5 (2025): 6105–6126. https://doi.org/10.1007/s43681-025-00816-3

Concepts:  Generational Trust Gap | cultural memory transition | Holocaust-gamified memory | Inter-generational knowledge validation) | Hybrid educational model 

Keywords:  Holocaust education | Generational memory | AI ethics | Survivor testimony | Intergenerational transmission

Rosenthal presents the first empirical, multi-generational analysis (191 participants, four generations) of how Holocaust descendants perceive AI in Holocaust education. Her key finding — that concerns about AI are not age-based or tied to digital fluency but rooted in distinct ethical and moral perspectives — is the empirical complement to Burzlaff’s scholarly argument. Her Generational Trust Gap framework maps the progressive displacement of dimensionality as lived connection to testimony fades: from 1G’s absolute rejection of AI on grounds of authenticity, through 2G’s cautious ethical openness, 3G’s algorithmic bias concerns, to 4G’s gamification anxiety. She advocates for hybrid models integrating AI’s reach with human storytelling and scholarly oversight.

“A computer cannot replicate the survivor’s tears or a descendant’s passion.” (Participant 1-04(Q), in Rosenthal, 2025, p. 6118)

“AI must never be our only source of truth.” (Participant 2-48(Q), in Rosenthal, 2025, p. 6122)

“The Holocaust isn’t just data. Real pain can never be coded.” (Participant 4G, in Rosenthal, 2025, p. 6116)

Ethics of the Algorithm: Digital Humanities and Holocaust Memory

Presner, Todd. Ethics of the Algorithm: Digital Humanities and Holocaust Memory. Princeton, NJ: Princeton University Press, 2024. https://doi.org/10.1515/9780691258980

Concepts:  Algorithmic ethics | Instrumental reason | Human dignity | Algorithmic visibility/judgment | Humanist AI guidance 

Keywords:  Digital humanities | Holocaust memory | AI ethics | Algorithm | Memory work

Presner argues that algorithms lack ethical agency and must not be entrusted with morally consequential decisions about Holocaust memory. He critiques the legacy of “instrumental reason” that enabled technocratic elements of genocide, warning that computational tools — including algorithms and mass data collections — must acknowledge their historical role in past abuses. He stresses the need for transparency about training data sources and calls for humanists to actively guide the application of AI to ensure authenticity, accuracy, and interpretive freedom. His concept of “close and distant listening” to the Shoah Foundation Visual History Archive is one of the most developed methodological frameworks in the field.

“Algorithms lack ethical agency and must not be entrusted with morally consequential decisions.” (Presner, 2024, cited in Rosenthal, 2025, p. 6109)

“Algorithms are not neutral tools but ‘sociotechnical ensembles’ embedded with human decisions.” (Presner, 2024, cited in Rosenthal, 2025, p. 6120)

“Humanists must actively guide the application of AI to ensure authenticity, accuracy, and digital provenance.” (Presner, 2024, cited in Rosenthal, 2025, p. 6119)

Transmitted Wounds: Media and the Mediation of Trauma

Pinchevski, Amit. Transmitted Wounds: Media and the Mediation of Trauma. Oxford: Oxford University Press, 2019.

Concepts:  Transmitted wounds | Digitally engaged ‘witnessee’ | Digitial Testimony re-performed | digital testimony Risks | narrative rupture vs. digital predictive smoothness

Keywords:  Trauma | Media theory | Holocaust testimony | Digital media | Witnessing

Pinchevski argues that the digital mediation of Holocaust testimony reconfigures witnessing by shifting the focus from the survivor to the digitally engaged “witnesses.” In this participatory model, testimony is actively re-performed, raising profound concerns about authenticity, agency, and trauma mediation. He identifies four specific risks: reducing suffering to data, overidentification with survivors, replacing embodiment with user interaction, and flattening the emotional depth of trauma. He contrasts analog media’s capacity to convey narrative ruptures and silences — what the log calls core memory — with digital media’s predictive smoothness, which erases the affective gaps that give testimony its force.  

“Digital media’s predictive smoothness... may erase... affective gaps.” (Pinchevski, 2019, cited in Rosenthal, 2025, p. 6109)

“Reducing suffering to data, overidentification with survivors, replacing embodiment with user interaction.” (Pinchevski’s four risks, 2019, cited in Rosenthal, 2025, p. 6109)

“Algorithmic tagging of meaningful silences, turning ethically charged disruptions into semantically quotable formula.” (Pinchevski, 2019, cited in Rosenthal, 2025, p. 6121)

USC Digital Repository Powers New AI-Driven Holocaust Testimony Installation

Museum of Jewish Heritage. “Museum Launches New AI Installation and Web Resource on International Holocaust Remembrance Day.” Press release, January 28, 2025. https://mjhnyc.org/press/museum-launches-new-ai-installation-and-web-resource-on-international-holocaust-remembrance-day/  

Masters, Nathan. “USC Digital Repository Powers New AI-Driven Holocaust Testimony Installation.” USC Libraries, February 7, 2025. https://libraries.usc.edu/article/usc-digital-repository-powers-new-ai-driven-holocaust-testimony-installation

Concepts:  Non-generative AI | Collective testimony | Unmediated survivor voice | Questions | Intergenerational bridge 

Keywords:  Holocaust | Survivor testimony | Interactive installation | Non-generative AI | Museum of Jewish Heritage

These paired press releases document the January 2025 launch of Survivor Stories: An Interactive Dialog, developed by the Museum of Jewish Heritage, USC Shoah Foundation, and USC Libraries. The initiative enables visitors to converse with ten Holocaust survivors — nonagenarians ranging from 90 to 98 years old — through a non-generative AI system that retrieves and plays back unedited, pre-recorded video responses. Crucially, the technology does not interpret, imagine, manipulate, or generate responses: it matches and plays back what real survivors actually said, unedited. 

This is the practical implementation of the log’s core argument: AI can only do with testimony what human beings enable it to do, and this initiative demonstrates what enabling looks like when it is done with ethical care. The survivors participated as co-authors, shaping the interview questions and determining what future generations would hear.

“The technology does not interpret, imagine, manipulate, nor generate new or composite responses.” (Museum of Jewish Heritage, 2025, n.p.)

“Meeting a survivor and hearing from that person a firsthand account of lived experience has a profound impact on young people.” (Radensky, in Museum of Jewish Heritage, 2025, n.p.)

“Survivors have been the Museum’s North Star since its founding.” (Kliger, in Museum of Jewish Heritage, 2025, n.p.)

Kairotic Entanglement: Kairos, Artificial Intelligence, Persuasion, and the Search for Meaning — A Literature Review

Moore, Jonathan T. “Kairotic Entanglement: Kairos, Artificial Intelligence, Persuasion, and the Search for Meaning — A Literature Review.” Spectra Undergraduate Research Journal 3, no. 2 (2024): 21–33. https://doi.org/10.9741/2766-7227.1030

Concepts:  Kairos as opportune timing in persuasion | Captology (computers as persuasive technology) | Kairotic entanglement (AI-human persuasive interaction) | Human agency in AI encounters | Design ethics for persuasive technology

Keywords:  Kairos | AI persuasion | Captology | Rhetoric | Human-computer interaction

Moore conducts a literature review examining Kairos — the classical rhetorical concept of opportune timing — in relation to AI-driven persuasive technologies. He proposes “Kairotic Entanglement” to describe the emergent relationship between human agency and AI-driven persuasive encounters. This is the only existing academic work to join the terms Kairos, AI, and meaning in a scholarly context, and it is included in this log as a boundary entry: Moore’s rhetorical kairos (when to act to persuade most effectively) is categorically different from the Baertian kairotic framework (what conditions allow genuine understanding and transformation). The two frameworks share an ancient word; everything else diverges. Moore’s kairotic AI is an optimised persuader; the log’s kairotic space is precisely what optimized persuasion closes down.

“Kairos plays a substantive role in persuasive technologies.” (Moore, 2024, abstract)

“Human agency is suggested to play an equally important role in how persuasive technologies may engage people.” (Moore, 2024, abstract)

“A greater emphasis on design ethics and data privacy is required in relation to persuasive technologies.” (Moore, 2024, abstract)

A Learning Ecology-Based Approach for Enhancing Digital Holocaust Memory in European Cultural Heritage Education

Manca, S., J. E. Raffaghelli, and A. Sangrà. “A Learning Ecology-Based Approach for Enhancing Digital Holocaust Memory in European Cultural Heritage Education.” Heliyon 9, no. 9 (2023): e19286. https://doi.org/10.1016/j.heliyon.2023.e19286 

Concepts:  Learning ecology | AI tools | Participatory memory-making | Passive-active learning-shift | Holocaust memory-representation

Keywords:  Holocaust education | Digital memory | Cultural heritage | Learning ecology | Europe

Manca et al. propose a learning ecology framework in which AI tools, social media, and simulations contribute to flexible, digitally integrated Holocaust education, marking a shift from passive learning to participatory memory-making. Their framework is cited by Rosenthal as part of the generational shift in Holocaust learning practices and by UNESCO as evidence of AI’s potential to support new methods for representing Holocaust memory. Their work is inherently multi-modal and human-designed, leaving space for kairotic encounters…  

“AI tools, social media, and simulations contribute to flexible, digitally integrated Holocaust education.” (paraphrase of Manca et al., 2023, cited in Rosenthal, 2025, p. 6119)

“A shift from passive learning to participatory memory-making.” (paraphrase of Manca et al., 2023, cited in Rosenthal, 2025, p. 6108)

“AI’s ability to support new methods for representing Holocaust memory.” (paraphrase of Manca et al., 2023, cited in Rosenthal, 2025, p. 6108)

Artificial History? Inquiring ChatGPT on Historiography

Leme Lopes, A. P. “Artificial History? Inquiring ChatGPT on Historiography.” Rethinking History 27, no. 4 (2023): 709–749. https://doi.org/10.1080/13642529.2023.2234227

Concepts:  historiographical synthesis | historical debate | AI mimicry | probabilistic language | summarizing-positioning

Keywords:  AI and history | ChatGPT | Historiography | Historical methodology | Digital humanities

Leme Lopes demonstrates with ChatGPT that AI can synthesize existing narratives, summarize debates, and mimic argumentative tone, but cannot position itself within a historical conversation. This is not a technical failure but a structural limitation of probabilistic language: it cannot rupture, argue, or position itself within complex debates. 

Burzlaff builds directly on this 2023 finding, confirming it still holds two years and several model updates later, while extending the argument from AI’s limited capacity for historiographical positioning to its more fundamental failure to engage with testimonial fracture, silence, and ethical weight.

“ChatGPT can synthesize existing narratives, summarize debates, and mimic argumentative tone.” (Leme Lopes, 2023, cited in Burzlaff, 2025, p. 6)

“It cannot position itself within a historical conversation.” (Leme Lopes, 2023, cited in Burzlaff, 2025, p. 6)

“A structural limitation of probabilistic language: its inability to rupture, to argue.” (Burzlaff’s synthesis of Leme Lopes, 2025, p. 6)

Archival Memory: Representations of the Holocaust in Oral Testimony

Kraft, Robert N. “Archival Memory: Representations of the Holocaust in Oral Testimony.” Poetics Today 27, no. 2 (2006): 311–330. https://doi.org/10.1215/03335372-2005-006

Concepts:  Core memory | Narrative memory | Archival memory | lived atrocity | Memory function 

Keywords:  Holocaust testimony | Oral testimony | Survivor memory | Trauma representation | Video testimony archive

Kraft’s cognitive analysis of Holocaust oral testimony introduces the foundational distinction between core memory (the original phenomenal experience — vivid perceptual images, deeply felt emotions, and bodily sensations) and narrative memory (the episodic integration of core memories into communicable story structures). Core memories are extraordinarily persistent; time does not diminish their potency. A third register, archival memory, describes testimony oriented toward historical documentation and ethical transmission rather than personal narrative closure.

Kraft’s contribution is conceptually decisive: it explains, from cognitive science, exactly why AI flattens testimony in the way Burzlaff demonstrates. AI operates at the narrative-memory level and is structurally unable to access or transmit core memory. The scenes that ChatGPT omits from Fortunoff testimonies — Luisa D.’s blood, Samuel W.’s stoic silence — are precisely core-memory content: vivid, embodied, emotionally charged, resistant to narrative integration. Kraft’s 2006 framework thus provides the cognitive explanation for Burzlaff’s 2025 observation.

“Core memory consists of representations of the original phenomenal experience.” (Kraft, 2006, p. 311)

“The core memories of atrocity are extraordinarily persistent.” (Kraft, 2006, p. 311)

“Narrating the traumatic events ascribes meaning... to the function of memory.” (Kraft, 2006, p. 311)

Toward the Recognition of Artificial History Makers

Hughes-Warrington, Marnie. “Toward the Recognition of Artificial History Makers.” History and Theory 61, no. 4 (2022): 107–118. https://doi.org/10.1111/hith.12278

Concepts:  Artificial history makers | AI as recursive, revisionary historical discipline | History-shaped-by-belief | Recognition of non-human historical agents | Epistemological challenge of AI to historiography

Keywords:  AI and historiography | History theory | Historical thinking | Artificial intelligence | Epistemology

Hughes-Warrington’s 2022 article poses the question: if AI learns by revisiting the past, are we prepared to call it historical thinking? Her answer, both daring and unsettling, is that artificial intelligence may already be a kind of historical discipline: recursive, revisionary, and shaped by belief. Cited in Burzlaff as a precedent for thinking about AI’s historiographical capacity, this earlier article sets up the more ethically urgent 2025 intervention. For the log, its significance lies in the question it forces: if AI can perform the surface operations of historical thinking, what remains that is distinctively human?

“If AI learns by revisiting the past, are we prepared to call it historical thinking?” (Hughes-Warrington, 2022, cited in Burzlaff, 2025, p. 4)

“Artificial intelligence may already be a kind of historical discipline, functioning as a historical discipline, recursively drawing on prior beliefs, revising them, and using those revisions to shape present and future judgments.”

Ethics for Artificial Historians

Hughes-Warrington, Marnie. “Ethics for Artificial Historians.” History and Theory 64, no. 2 (2025): 159–177. https://doi.org/10.1111/hith.12377

Concepts:  Artificial historians | accountability, in AI | Recursion | AI-generated historical knowledge | artificial HISTORY Responsibility 

Keywords:  AI and history | Historical ethics | Historiography | Artificial intelligence | Memory

Hughes-Warrington’s 2025 article sharpens the stakes for artificial historians: AI systems, she argues, force us to confront not only the epistemology of AI but also its ethical orientation — its absence of accountability, memory, and obligation. While generative models operate through recursion, they lack reflexivity: they learn from the past without taking responsibility for it. This distinction — between memory and repetition, between meaning and recombination — is precisely where ethics enters, and precisely where historical thinking must stake its ground. For the log, Hughes-Warrington’s 2025 article is a companion to Burzlaff’s, addressing the same problem from a philosophy-of-history rather than a practitioner’s perspective.

“Artificial historians force us to confront... its absence of accountability, memory, and obligation.” (Hughes-Warrington, 2025, p. 159)

“Learning from the past without taking responsibility for it.” (Hughes-Warrington, 2025, cited in Burzlaff, 2025, p. 4)

“The distinction between memory and repetition, between meaning and recombination, is precisely where ethics enters.” (Hughes-Warrington, 2025, cited in Burzlaff, 2025, p. 4)

Navigating the Meta-Crisis of Generativity: Adapting Qualitative Research Quality Criteria in the Era of Generative AI

Hasan, Md. Kamrul, Md. Nurul Haque, et al. “Navigating the Meta-Crisis of Generativity: Adapting Qualitative Research Quality Criteria in the Era of Generative AI.” Frontiers in Research Metrics and Analytics 10 (2025). https://doi.org/10.3389/frma.2025.1685968

Concepts:  generativity | Generative transparency | Ethical relationality | Postmodern pluralism (GenAI provocateur) | representation

Keywords:  Qualitative research | Generative AI | Research ethics | Epistemology | Methodology

Hasan et al. introduce the concept of the “meta-crisis of generativity” — a convergence of uncertainties arising when human researchers collaborate with GenAI to produce knowledge, entangling the foundational crises of representation, legitimation, and praxis identified by Denzin and Lincoln. They propose three interconnected principles for qualitative researchers: generative transparency, ethical relationality, and postmodern pluralism that embraces GenAI as a critical provocateur. For the log, this article provides the qualitative research methodology parallel to Burzlaff’s historiographical argument: the same concerns about AI’s effects on meaning-making, interpretation, and ethical engagement arise in qualitative social science as in historical writing.

“The meta-crisis of generativity: a convergence of uncertainties arising when human researchers collaborate with GenAI.” (Hasan et al., 2025, p. 1)

“Ethical relationality: prioritizing human accountability over AI efficiency.” (Hasan et al., 2025, p. 1)

“Embraces GenAI as a critical ‘provocateur’ challenging human-centric epistemology.” (Hasan et al., 2025, p. 1)

Humanities in the Time of AI

Dubreuil, Laurent. Humanities in the Time of AI. Minneapolis: University of Minnesota Press, 2025.

Concepts:  Poiesis vs. simulation | AI mimicry | Literary intelligence | humanity-fluency

Keywords:  Literary theory | AI and humanities | Language | Meaning | Simulation

Dubreuil distinguishes between poiesis — the act of making meaning, the originary creative act through which something genuinely new comes into being — and simulation: the recombination of encoded patterns that produces the appearance of creation without its substance. This distinction, cited in Burzlaff’s second principle (create, do not reproduce), is foundational to the log’s argument that what AI produces when it processes testimony is not understanding but an elaborate simulation of understanding. The historian’s task, and by extension the dialogue facilitator’s and educator’s task, is poietic — genuinely compositional, risky, and unrepeatable.

“Not poiesis... but simulation: not the act of making meaning, but the mere reproduction of patterns already encoded.” (Dubreuil, 2025, cited in Burzlaff, 2025, p. 7)

“AI recycles insight rather than generating it.” (paraphrase of Dubreuil, 2025, cited in Burzlaff, 2025, p. 7)

“What AI gives us is simulation, not the act of making meaning.” (Burzlaff’s synthesis of Dubreuil, 2025, p. 7)

Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence

Crawford, Kate. Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence. New Haven, CT: Yale University Press, 2021.

Concepts:  AI power | Extractive infrastructure | obscuring responsibility | harm scale | AI-Non-neutrality

Keywords:  AI politics | Technology critique | Labor | Environment | Infrastructure

Crawford argues that AI systems are not neutral tools but powerful systems of statistical pattern recognition built on extractive logics that concentrate power and obscure responsibility. Far from artificial or intelligent in any humanistic sense, today’s dominant AI models codify social and environmental harms into scalable infrastructures. Her characterization of AI as “a registry of power” (p. 8) is cited by Burzlaff to establish that AI’s limitations in processing Holocaust testimony are not technical accidents but structural features of how such systems are built and deployed.

For the log’s concerns, Crawford’s work provides the political economy of flattening: it explains not only that AI smooths and reduces, but why — because systems built to extract, scale, and concentrate cannot by design attend to the particular, the irreducible, or the ethically entangled. Her work also connects to the log’s archive theory dimension: AI can only ingest the record, not interrogate its construction — and it was created by the same logic of extraction that shapes the record it ingests.

“A registry of power.” (Crawford, 2021, p. 8)

“They are powerful systems of statistical pattern recognition built on extractive logics.” (Crawford, 2021, p. 8, cited in Burzlaff, 2025, p. 9)

“Far from neutral tools, they codify social and environmental harms into scalable infrastructures.” (Crawford, 2021, p. 8, cited in Burzlaff, 2025, p. 9)

Fragments, Not Prompts: Five Principles for Writing History in the Age of AI

Burzlaff, Jan. “Fragments, Not Prompts: Five Principles for Writing History in the Age of AI.” Rethinking History: The Journal of Theory and Practice, September 14, 2025. https://doi.org/10.1080/13642529.2025.2546174

Concepts: Flattening (AI’s conversion of complexity into coherence) | Fracture (what testimony preserves; where meaning lives) | Poiesis vs. simulation (meaning-making vs. pattern recombination, after Dubreuil) | Ethical entanglement (remaining inside moral ambiguity; refusing algorithmic resolution) | Testimonial geography (how location of testimony shapes its texture)

Keywords:  Historical writing | Holocaust testimony | Generative AI | Historiography | Narrative ethics

Burzlaff argues that generative AI exposes — by negation — what is irreducible in historical writing: interpretation over description, creative risk over reproduction, archival singularity over pattern, ethical entanglement over algorithmic resolution, and authorial presence over fluent output. Drawing on five Fortunoff Video Archive testimonies (all recorded 1995, in La Paz, Kraków, and Connecticut), he proposes that the historian’s task is not to outperform AI but to be constitutively unlike it. Working from the case of ChatGPT’s omission of a Holocaust survivor’s most intimate and irreducible memory (Luisa D.’s mother spreading blood on her lips), he shows that AI is structurally trained to privilege the probable over the profound.

For this literature review’s governing concerns, Burzlaff’s article is the primary foundational text. Though addressed to historians, its five principles translate across disciplines as conditions for maintaining the kairotic capacity for genuine encounter with testimony: hesitate, return, interpret; create rather than reproduce; use corpora without becoming one; refuse algorithmic ethics; write as a person not a prompt. His argument that historical writing is “not an output but a risk — a form of presence, of dwelling, of making meaning” (p. 2) is the log’s operating principle extended beyond its disciplinary origin.

“AI renders the past too smoothly, rendering history as something already digested, already known.” (Burzlaff, 2025, p. 5)

“Description is what machines do. Interpretation is what historians must continue to reclaim.” (Burzlaff, 2025, p. 5)

“It is not about outperforming the machine but about sounding nothing like it.” (Burzlaff, 2025, p. 2)

History and Morality

Bloxham, Donald. History and Morality. Oxford: Oxford University Press, 2020.

Concepts:  Best bias | Ethical stance | Objectivity | Moral attention - Proximity 

Keywords:  Historical ethics | Objectivity | Morality | Atrocity history | Methodology

Bloxham argues that ethical stances in historical writing are unavoidable — especially in histories of atrocity, In his article, the Holocaust demands moral attention. Objectivity here is not the absence of perspective but the cultivation of what he calls the “best bias”: the most self-aware Bloxham’s framework argues that algorithmic ethics — which seeks resolution and avoids moral ambiguity — is a form of distortion parallel to, and only somewhat milder than nationalist revisionism.

Quotes: 

“Objectivity is... the cultivation of the ‘best bias’ — the most self-aware stance a historian can adopt.” (Bloxham, 2020, cited in Burzlaff, 2025, p. 12)

“Ethical stances are unavoidable, especially when we write histories of atrocity.” (Bloxham, 2020, cited in Burzlaff, 2025, p. 12)

“What matters is how open we are about our stance, and how we undertake it.” (Burzlaff’s synthesis of Bloxham and E.H. Carr, 2025, p. 12)