Listening after the end: The ethics of algorithmic memory

Last week, I “chatted” with Anne Frank. On the website DeepAI.org, the interface invites you to ask her anything: about the diary, the annex, her hopes, her death. Responses to my prompts arrived in clean prose, often emotionally attuned, always grammatically correct. The answers felt plausible—but I know firsthand that they are often fabricated. A disclaimer below the chatbox reads that Anne Frank is an AI language model, not a real person. These initiatives have been described both as violations and as tools for education. What few discussed was the mechanism itself: the system behind the voice, the prompts that shaped her answers, the archive it drew from—and what it left out.

It is a window into what Todd Presner, in Ethics of the Algorithm: Digital Humanities and Holocaust Memory, calls the post-testimonial condition: a world in which Holocaust memory is increasingly mediated—shaped, filtered, and reassembled—by machine-learning systems optimized for scale, searchability, and user interaction. The book is the most sustained scholarly attempt to date to think through this epistemological and ethical shift. Rather than treating digital tools as neutral or inherently corrosive, Presner interrogates their architectures: how algorithms sort what we hear and how computational methods can either deepen or distort our relationship to the past. The result is a rare book—one that bridges Holocaust historiography, critical data studies, the politics of design, oral history, interface theory, and archival ethics. As memory becomes computational, computation becomes an interpretive act—and the future of historical understanding depends on how these systems are built.

The book’s central claim is deceptively simple: algorithms are not neutral. They filter, rank, and shape what becomes legible in a digital archive—and what remains unsaid. In the context of Holocaust testimony, where silence, uncertainty, and affective rupture often matter more than declarative content, such mediation is never just technical: it is ethical. Rather than treat computation as either dangerous or inevitable, Presner insists that design decisions must be guided by ethical commitments from the outset. He calls this heuristic humanism—a framework that privileges attentiveness, care, and interpretive openness over efficiency or prediction. The problem is not computation itself, but the way it often flattens or forecloses interpretive possibility.

To illustrate the stakes, Presner aptly moves between philosophical framing and historical casework. He links modern algorithmic logic to earlier bureaucratic forms of sorting and simplification, drawing on thinkers like Hannah Arendt, Zygmunt Bauman, and Simone de Beauvoir. These comparisons do not equate computation with genocide, but trace the persistence of bureaucratic logics that render people as data and complexity as noise. A concrete case study involves Jacobus Lentz’s identity-card system in Nazi-occupied Netherlands, whose precise taxonomies enabled the deportation of over 100,000 Jews. Presner reads this as a prototype of bare data, a computational counterpart to Giorgio Agamben’s “bare life.” In contrast, he highlights reparative digital memorials—Amsterdam’s Names Monument or the participatory archive Let Them Speak—which resist closure and instead build what he calls “generous interfaces”: systems designed for ethical encounter, not efficient query.

For Presner and his team, the archive is not just a repository of testimony, but a system of knowledge. The way testimony is segmented, indexed, and retrieved profoundly shapes what kinds of memory survive. One example is the Shoah Foundation’s Visual History Archive—the largest digital collection of Holocaust testimony—where interviews are tagged by hundreds of standardized terms, from “deportation” to “liberation.” These categories make content accessible—but also impose a logic of coherence and containment. Emotional dissonance, narrative fragmentation, and tonal complexity are harder to classify, and often go unindexed. As a result, some of the most ethically charged dimensions of testimony—stammered recollections, moments of silence, gestures that exceed language—are rendered effectively invisible.

The book’s convincing response is not to abandon metadata, but to reimagine it through subjunctive metadata: indexing practices that emphasize ambiguity, relationality, and the unsaid. Rather than affirm what a testimony “is about,” these systems invite users to interpret how meaning is made. Presner is also attuned to the risk of what he dubs post-canonical listening: the tendency of search tools to steer users toward the most recognizable figures—Wiesel, Levi, Frank—at the expense of lesser-known voices. When interface design flattens difference into familiarity, the archive becomes a memory machine that echoes what it already knows.

One of the most original contributions lies in how the book retools computation as a method of ethical listening. Rather than treating algorithms as instruments of speed or scale, Presner and his team explore how they might amplify fragility, ambiguity, and multiplicity—qualities often lost in structured archives. One technique he and his collaborators use is semantic parsing: breaking testimony into subject–verb–object fragments to trace forms of agency across narratives. Even a sentence like “We had to wear the yellow star” becomes analytically rich—revealing coercion, collective identity, and spatial control. Repeated across thousands of interviews, these fragments form what Presner calls testimonial ensembles: constellations of related experience that resist reduction to typology.

This approach culminates in a digital reconstruction of Mala Zimetbaum’s life, remembered by other survivors at Auschwitz but never recorded in her own voice. Presner’s team extracts over two thousand narrative fragments about her, assembling a memory structure organized by themes like solidarity, care, escape, and death. The result is not a biography, but a reassembled voice: polyphonic, precarious, ethically unfinished. In another example, he turns to voice analysis: mapping pitch shifts, pauses, and breath patterns to register affective disruptions. These vocal textures—so often flattened in transcription—emerge as new objects of attention. The book’s algorithms do not promise closure; they prompt attunement and ask us to listen not just for what is said, but for what flickers at the edge of speech.

As survivor voices fall silent, their presence will increasingly take mediated form: holograms, scripted avatars, chatbots. This is the post-testimonial condition—a moment when memory is no longer narrated but computed, retrieved, and staged. One example is Dimensions in Testimony, a project that allows users to “converse” with recorded survivors. The illusion is convincing, but the system is powered by keyword-matching, not understanding. What looks like conversation is a curated playback—shaped by what was recorded, and what the interface decides to reveal.Presner is less concerned with deception than with design. Who decides which responses are indexed, which clips remain buried, and what kinds of questions are encouraged? Often, the most conflicted or emotionally complex answers go unplayed—not because they are absent, but because no one asks for them, or the system fails to match them.

This same logic extends beyond institutional archives to Instagram memory projects, Twitter bots, and AI-driven grief apps. Each offers connection; each depends on simulation. What they share is a premise: that memory can be externalized and made interactive. Presner and his collaborators do not reject these tools; they ask us to examine their architecture. As memory becomes a matter of interface, the ethics of presence—who speaks, and how—becomes an urgent question of infrastructure.

Ethics of the Algorithm is not a warning about technology, nor a celebration of innovation. It is a demand for scholarly responsibility in a time when historical understanding is increasingly shaped by systems we neither fully see nor control. What used to be a matter of archive and witness is now also a matter of code, interface, and infrastructure.What makes the book so striking is its refusal to separate the moral from the technical: metadata is not neutral, search design is not innocent. Listening is anything but passive—it is an act structured by what the system allows us to hear.

At its core, the book offers a quiet but urgent confrontation between quantification and interpretation—not only a diagnosis of risks like datafication, scale, and algorithmic opacity, but also a proposal for ethical design, metadata reform, and responsible infrastructure. Rooted in Holocaust studies, the book extends outward—offering an interdisciplinary model for how humanistic scholarship might help shape the ethical futures of computation. In this newage, we need an approach that combines large-scale computational methods with a values-based, humanistic commitment to the dignity of individual voices. The near future will only sharpen how memory is reshaped by algorithmic mediation. We ought not to romanticize our distance from the events we study. Instead, the book urges us to acknowledge that distance—and to take responsibility for the systems through which it is now navigated.