There are 100,000 brass plates set into the sidewalks of Europe.
Each one is the size of a cobblestone. Each one is engraved with a name, a birth date, a deportation date, a destination. Hier wohnte—here lived. Gunter Demnig began laying the Stolpersteine in 1992, pressing them into the pavement in front of the last freely chosen homes of Holocaust victims. You stumble on them. That is the point. The German word Stolperstein means stumbling stone. You are walking to get coffee and your eye catches brass in the concrete and for a half-second you are looking at a name and a year and Auschwitz, and then you keep walking, and then you don’t forget.
This is what deliberate memory looks like. Not a museum you choose to visit. Not a documentary you choose to watch. A brass plate in the ground that interrupts your day without asking permission.
I wrote a paper about this in the spring of 2017. Intro to European Studies at UT Austin, twenty-one years old, trying to figure out how a country metabolizes something that cannot be metabolized. The paper was about Vergangenheitsbewältigung—Germany’s decades-long project of coming to terms with the past—and my argument was simple: collective memory is not a natural process. It is political technology. Germany proved this by building it from scratch. Forty-two billion euros in reparations paid. Fifty-six billion more expected through 2037. Mandatory Holocaust education. Speech laws criminalizing denial. The Memorial to the Murdered Jews of Europe—2,711 concrete stelae spread across 19,000 square meters in the center of Berlin, an entire city block given over to the act of not forgetting. I leaned on Wulf Kansteiner’s five-phase periodization of German post-Holocaust memory and Maurice Halbwachs’s framework for how societies construct what they remember. I got an A-minus. My professor noted that when the subject has both moral weight and structural complexity, I produce my best work.
What I could not see at twenty-one was the question nobody was asking: what happens when the architecture of memory is no longer attacked by forgetting, but by machinery that makes forgetting and remembering functionally indistinguishable?
1. What Germany Built
Start with what Germany actually constructed, because the scale of the project is easy to understate.
After 1945, the country that had engineered the most systematic genocide in human history undertook the most systematic project of collective remembrance in human history. These two facts belong in the same sentence. The engineering capacity is the same. The institutional coordination is the same. The cultural infrastructure is the same. Germany used the machinery of a modern state—education systems, legal codes, public architecture, diplomatic protocols, media policy—to ensure that what the state had done would never be structurally forgettable. This was not guilt. Guilt is a feeling. This was architecture.
The theoretical framework beneath it evolved across decades. Halbwachs, the French sociologist, laid the foundation in the 1920s: memory is socially constructed, shaped by the groups we belong to, maintained through institutional practice. What we remember as individuals is scaffolded by what our society makes available to remember. His framework was powerful but rigid—it treated collective memory as something a society has, like a possession, rather than something a society does, like a practice. Jan Assmann complicated this by distinguishing communicative memory—the lived memory of direct witnesses, lasting roughly eighty to one hundred years—from cultural memory, which is externalized into institutions, texts, monuments, and rituals and can persist indefinitely. Aleida Assmann added a further distinction: functional memory, which is active and identity-forming, the memory a society uses to know itself, versus storage memory, which is archived and dormant, available for reactivation but not part of daily self-understanding (Tamm, 2013).
Germany’s project was an exercise in keeping memory functional. Not stored. Not archived. Active. The Stolpersteine in the sidewalk. The speech laws that make denial a criminal act. The school curricula. The reparations payments that extend into 2037—a financial commitment that keeps the past economically present in annual budgets. Every mechanism was designed to prevent what the Assmanns would call the migration of Holocaust memory from functional to storage—from something Germany lives with to something Germany once knew.
The assumption beneath all of this was that the threat to memory was entropy. Time. Generational distance. The natural human tendency to let the uncomfortable fade. The entire architecture was oriented toward a single enemy: forgetting. Build enough memorials, fund enough education, maintain enough institutional commitment, and the memory holds. The question was never whether the architecture itself would come under attack. The question was whether the commitment to maintain it would endure.
That assumption was wrong. Not because the commitment failed. Because the threat changed.
2. The Siege From Within
In January 2017—the same semester I was writing my paper about the durability of German memory—Björn Höcke stood in front of an audience in Dresden and called Berlin’s Memorial to the Murdered Jews of Europe a “memorial of shame” (Hinterleitner & Sager, 2022). He said Germany needed a “180-degree turn” in its memory culture. He was a leader of the Alternative für Deutschland, a party that had been founded only four years earlier as a Eurosceptic protest movement and was rapidly mutating into something else.
I didn’t know about Höcke’s speech when I wrote my paper. If I had, it would have changed the argument. Because Höcke was not engaged in forgetting. Forgetting is passive—the thing Germany’s memory architecture was built to resist. Höcke was engaged in demolition. He was standing inside the architecture and swinging a hammer at the load-bearing walls.
The AfD’s assault on German memory culture is documented extensively in the scholarly literature, and what the literature shows is not a fringe movement but a structural attack on the postwar consensus. Alexander Gauland, the party’s co-founder, called the Nazi era a “Vogelschiss”—a bird dropping—in the sweep of German history (Shell et al., 2020). AfD figures appropriated resistance heritage, claiming the legacy of Stauffenberg and the July 20 plotters while simultaneously relativizing the crimes they resisted (Cento Bull & Clarke, 2020). Potter (2020) argues that populism was already reshaping German memory culture in ways that had “gone unrecognised.” Huyssen (2020) puts it most clearly: “The success of the AfD shows that memory politics in Germany has dealt better with the past than it deals with the political pressures of the present.”
The numbers track the rhetoric. AfD support nearly doubled from 10.3 percent in 2021 to 20.8 percent by 2025 (Yalcin & Onal, 2025). These are not protest voters. This is the second-largest party in the Bundestag. One in five German voters now supports a party whose leaders have called the nation’s central memorial a shame and its defining historical crime a bird dropping.
Then came January 2024. Correctiv, a German investigative outlet, exposed a clandestine meeting in Potsdam where AfD figures and associates discussed plans for mass deportation of people with migration backgrounds—including German citizens. Three million Germans took to the streets. They marched under a single motto: NeverAgainIsNow (Yalcin & Onal, 2025).
The instinct is to read this as the architecture working—the memory infrastructure activating in response to a live threat. And it is that. Three million people in the streets is not a society that has forgotten. But look at the structural situation clearly: the memory architecture that took decades to build now requires three million people marching in the cold to defend it against elected officials in the Bundestag. The architecture was designed to prevent forgetting. It was not designed to withstand demolition by a party that commands a fifth of the electorate.
My original paper, written the same month Höcke delivered his speech, worried about whether German students received enough in-depth study. I cited Lars Rensmann’s observation that many young Germans deflected with whataboutism: “The Americans did the same thing with the Indians, and the Israelis do the same with the Palestinians.” I framed this as an education problem—not enough depth before 10th grade, at which point two-thirds of students leave the formal system. The concern was real but the framing was wrong. The problem was never that students weren’t learning enough about the Holocaust. The problem was that an entire political movement had emerged to tell them the learning itself was the problem—that the architecture of memory was a weapon used against Germany, not a gift Germany had given itself.
3. The Machine That Remembers for You
Here is where the argument turns, and where the connection to the Comanche essay’s central paradox becomes structural rather than metaphorical.
Germany built its memory architecture with the tools of a modern state: mass communication, institutional coordination, cultural engineering. These are the same tools now being automated. And the automated versions don’t remember or forget. They do something for which we don’t have good language yet—they generate a continuous present tense of content that makes the distinction between remembering and forgetting functionally irrelevant.
Wulf Kansteiner published “Digital Doping for Historians” in 2022. The same Kansteiner whose five-phase periodization anchored my paper at twenty-one. The same scholar who mapped how Germany processed the Holocaust through distinct phases of denial, acknowledgment, marginalization, revival, and institutionalization. That Kansteiner is now warning that AI is “making history, literally”—that machine learning tools play “a key role in crafting images and stories about the past in popular culture” and that large language models are “structurally incapable of telling the truth and tracking pieces of information through complex causal chains” (Kansteiner, 2022).
Read that last clause again. Structurally incapable. Not occasionally wrong. Not prone to error. Structurally incapable. The scholar who built the framework for understanding how Germany remembers is now arguing that the dominant information technology of the present cannot, by design, participate honestly in the act of memory.
This matters because the act of memory has migrated. Neiger (2020) identifies media as the primary shaper of collective memory in the digital age, arguing that media practitioners not only shape audience memory but create self-referential memory cultures of their own. The Stolpersteine were designed for sidewalks. Holocaust education was designed for classrooms. The Claims Conference surveys were designed for populations that encountered the past through institutions. None of these mechanisms were designed for a world in which the primary site of memory formation is an algorithmic feed optimized for engagement.
A January 2025 Claims Conference survey across eight countries found “alarming gaps in knowledge, particularly among younger generations.” German memorial directors report that students displaying the Hitler salute and shouting “Sieg Heil” on school trips went from “one or two per class” to “common sense” in school groups. Meanwhile, younger participants show greater trust in digital platforms and AI-driven tools for learning about the Holocaust than in traditional institutional sources.
Lily Ebert, a Holocaust survivor, built a following of two million on TikTok to fight antisemitism and misinformation. She died in October 2024 at age one hundred—among the last direct witnesses. One in three TikTok videos about Holocaust memorial sites receives comments containing hate speech or Holocaust distortion (Manca & Raffaghelli, 2023). The platform’s algorithm does not distinguish between a survivor’s testimony and the comment beneath it calling her a liar. Both are content. Both are engagement. The algorithm is structurally incapable of valuing one over the other—the same structural incapacity Kansteiner identified in large language models, manifested in the recommendation engine.
Germany spent decades building memory infrastructure with institutional commitment—with values encoded into the architecture. Silicon Valley built memory infrastructure accidentally, as a byproduct of engagement optimization. Both shape what populations remember. One was designed. The other was not. And the undesigned one is winning.
4. Synthetic Remembering, Synthetic Forgetting
The CHI 2025 Best Paper from MIT found that AI-edited visuals increased false recollections by up to 2.05 times compared to control conditions. The highest confidence in false memories came from AI-generated videos of AI-edited images (Nightingale et al., 2025). This is not about misinformation in the political sense—the deliberate creation of false content to advance an agenda. This is about the cognitive architecture of memory itself. The technology literally manufactures false memories in the minds of the people who encounter it. Not false beliefs. False memories. The difference matters. A false belief is something you think is true. A false memory is something you remember happening.
Germany’s memory project was built on the premise that the historical record, properly maintained and widely disseminated, could anchor collective memory against the drift of time. The Stolpersteine, the memorials, the mandatory education, the speech laws—all of them assume a shared evidentiary standard. The Holocaust happened. The evidence is overwhelming. Denial is not a competing interpretation; it is a category error. The entire architecture rests on this foundation: that the truth of the past, once established and maintained, provides stable ground on which collective identity can stand.
The liar’s dividend dissolves this foundation. Not by proving the record wrong, but by making the concept of a shared evidentiary standard seem naive.
The liar’s dividend, as documented by the Brennan Center (2024), works like this: in a world where AI can generate convincing fake images, fake video, and fake audio, any real piece of evidence can be dismissed as fabricated. The defense was used in the trial of Guy Reffitt after January 6—authentic video evidence of the Capitol breach was challenged on the grounds that it could have been AI-generated. The claim didn’t need to be true. It just needed to be plausible enough to introduce doubt.
Apply this to Holocaust memory. Germany’s entire architecture depends on the premise that the photographic record, the documentary evidence, the survivor testimony, the physical sites constitute proof that sits beyond reasonable dispute. The liar’s dividend doesn’t attack this evidence directly. It attacks the category of proof itself. When anything can be faked, nothing can be proven. When nothing can be proven, every piece of evidence becomes contestable—not because it has been refuted, but because the technology exists to produce something that looks identical to it.
This is the inversion of Germany’s project. They built architecture to make the past undeniable. AI builds architecture that makes everything deniable. And the second architecture does not need to intend this outcome. The liar’s dividend is a structural byproduct of the technology’s capability, not an intended function. Nobody at OpenAI or Google DeepMind set out to undermine Holocaust memory. They built tools that generate convincing synthetic media, and one downstream consequence of that capability is the erosion of the evidentiary standard on which all institutional memory depends.
The connection to the Comanche essay is direct, and it is not a metaphor. The force that creates is the force that destroys. Mass communication, institutional coordination, cultural engineering—these are the capabilities that let Germany build the most sophisticated memory architecture in human history. The same capabilities—now automated, accelerated, and stripped of institutional commitment—are the ones dismantling it. The technology of remembering and the technology of forgetting share a root system. The Stolpersteine and the deepfake come from the same civilizational capacity: the ability to construct representations of the past and embed them in the physical and informational environment. Germany used that capacity deliberately, with moral purpose, over decades. The algorithm uses the same capacity without deliberation, without moral purpose, in milliseconds. The tool is always also the weapon.
5. What Halbwachs Couldn’t See
Return to the theoretical framework. Halbwachs argued that memory is socially constructed—that what we remember is shaped by the groups we belong to and the institutions that maintain the past on our behalf. His framework assumed that the social groups were human and the institutions were governed by human judgment. Jan Assmann’s distinction between communicative and cultural memory assumed the same thing—communicative memory carried by living witnesses, cultural memory maintained by human-governed institutions.
The last Holocaust survivors are dying. Lily Ebert in October 2024. The communicative memory window—Assmann’s eighty to one hundred years—is closing. The Holocaust is transitioning from communicative to cultural memory in real time, which means its preservation now depends entirely on the institutional architecture Germany built. The memorials. The education system. The legal framework. The political will.
Every one of those institutions now operates in an information environment they were not designed for. The education system competes with algorithmic feeds for the attention of students who trust digital platforms more than classrooms. The speech laws criminalize Holocaust denial but cannot reach the algorithmic amplification that makes denial content go viral across borders. The memorials exist in physical space while the primary site of memory formation has migrated to digital space. The political will faces a party commanding twenty percent of the electorate that calls the memorial architecture a national humiliation.
The Shoah Foundation’s “Dimensions in Testimony” project already uses AI to create interactive conversations with recorded survivors—you ask a question, and an AI system selects the closest response from thousands of hours of pre-recorded testimony, creating the illusion of dialogue with a person who may no longer be alive. What category does this memory fall into? It is not communicative memory—there is no living witness. It is not quite cultural memory in the Assmann sense—it simulates the form of lived testimony rather than externalizing it into a monument or a text. It is something new. A third category that the frameworks don’t have a name for: synthetic communicative memory. The form of a living witness, maintained by a machine, responding to questions in something like real time.
This is not a dystopian scenario. It is a genuine attempt to preserve testimony beyond the lifespan of the witnesses. But it introduces a structural ambiguity that Germany’s memory architecture was designed to eliminate. The entire project of Vergangenheitsbewältigung was built on clarity: this happened, here is the evidence, here is the testimony, here is the physical site. The introduction of AI-mediated testimony—even well-intentioned AI-mediated testimony—blurs the line between record and reconstruction. And once that line is blurred in one direction, it becomes harder to police in every other direction.
6. The Whataboutism Machine
In my original paper, I cited Rensmann’s observation about German students deflecting from the Holocaust with comparative grievance: “The Americans did the same thing with the Indians, and the Israelis do the same with the Palestinians.” I treated this as a failure of education. In 2017, it was a minority opinion expressed by students who hadn’t received enough depth of instruction.
In 2025, it is an information architecture.
Algorithmic recommendation systems are, by structural design, whataboutism machines. Their function is to surface content that is contextually adjacent to what you have just consumed—not content that is analytically coherent, not content that maintains the moral distinctions a society has deliberately constructed, but content that is statistically associated with engagement. A student who watches a documentary about the Holocaust will be recommended content about other genocides, about colonial atrocities, about Israeli military operations, about American slavery—not because the algorithm has an agenda, but because those topics are contextually proximate in the engagement data. The result is a flattening of moral distinction that looks, from the inside, exactly like the whataboutism Rensmann documented. Except it is not coming from the student’s lack of education. It is coming from the information architecture the student inhabits.
Zeng and Kaye (2022) describe TikTok’s “visibility moderation”—the platform’s ability to algorithmically amplify or suppress content without users being aware of the intervention. This is memory architecture. It determines what content persists in collective attention and what content disappears. But unlike Germany’s memory architecture, which was designed with specific values and maintained by accountable institutions, visibility moderation operates through opaque algorithmic decisions made by engineers optimizing for a metric—engagement, retention, time-on-platform—that has nothing to do with historical truth or collective responsibility.
The whataboutism I identified as a student opinion in 2017 has become a structural feature of the information environment. Germany can fund every memorial, mandate every curriculum, enforce every speech law, and the algorithmic feed will still flatten the Holocaust into one item in an infinite scroll of comparable atrocities—not because the algorithm denies it, but because the algorithm cannot distinguish between events that a society has decided must never be equivalent and events that are statistically adjacent in engagement data.
Martell et al. (2024) documented how Twitter’s memory of January 6 split into competing frames—“insurrection” versus “censorship” versus “treason”—shaped not by the event itself but by platform dynamics. The same mechanism operates on every piece of historical memory that enters the algorithmic ecosystem. The memory is not lost. It is fractured into competing narratives, each reinforced by its own engagement feedback loop, each increasingly disconnected from the evidentiary base that institutional memory was designed to maintain.
7. The Tool Is Always Also the Weapon
Germany’s memory architecture was the most deliberate act of civilizational self-knowledge in modern history. A nation looked at the worst thing it had done, and instead of burying it, built an entire institutional apparatus to ensure it would never be forgettable. Forty-two billion euros. A hundred thousand brass plates. 2,711 concrete stelae. Mandatory curricula. Criminal statutes. Diplomatic posture calibrated to a debt that extends into 2037. This was not guilt performing itself. This was engineering.
The capabilities that made it possible—mass communication, institutional coordination, cultural engineering, the ability to construct and disseminate representations of the past at scale—are the same capabilities now operating in their automated form. And the automated form does not remember. It generates. It does not maintain. It produces. It does not distinguish between a survivor’s testimony and a synthetic video of a thing that never happened. It cannot, any more than a printing press can distinguish between a constitution and a forgery. The technology is the same. The institutional commitment is not.
This is the paradox the Comanche essay identified at the level of civilizational creation and destruction, now operating at the level of civilizational memory. The force that creates is the force that destroys. Germany used mass communication to build collective memory. Algorithms use mass communication to dissolve it. Germany used institutional coordination to maintain the historical record. Platforms use institutional coordination—corporate policy, content moderation, algorithmic tuning—to optimize for engagement, which structurally undermines the historical record as a byproduct. Germany used cultural engineering to make the Holocaust unforgettable. AI uses cultural engineering—recommendation engines, generative models, synthetic media—to make everything equally weighted, which makes nothing unforgettable.
The AfD is not the deepest threat to Germany’s memory architecture. The AfD is a political movement that can be opposed by political means—and three million people in the streets in January 2024 demonstrated that the will to oppose it exists. The deeper threat is the one that operates beneath politics: the migration of collective memory from governed institutions to ungoverned algorithms. From architecture designed with values to architecture designed without them. From a system that was built to ensure the past remains undeniable to a system that makes everything—past, present, real, synthetic—equally deniable.
Huyssen was right that German memory politics has dealt better with the past than with the pressures of the present. But the present is not just the AfD. The present is an information architecture that no nation’s memory project was designed to withstand. Germany built for entropy—the slow fade of time and generational distance. What arrived instead was velocity—the instantaneous generation and dissemination of content that mimics, distorts, flattens, and ultimately displaces institutional memory at a speed no human institution can match.
I wrote at twenty-one that “some incidents must never become lost in the sands of time.” I still believe that. But the sands of time were the wrong metaphor. Sand erodes slowly. What is happening to collective memory in the age of AI is not erosion. It is replacement. The institutional memory Germany built—deliberate, funded, maintained, accountable—is being replaced by algorithmic memory that is none of those things. Not destroyed. Not denied. Replaced. Outrun.
The person I keep coming back to is not the politician or the scholar or the protester. It is the German student on a school trip to Sachsenhausen who pulls out a phone and gives the Hitler salute for a TikTok. Not because he believes in National Socialism. Not because he is ignorant of what happened there. Because in the information architecture he inhabits, the Holocaust is content. The memorial is a backdrop. The salute is engagement. The algorithm that serves him his next video does not know or care what Sachsenhausen is. It knows he spent fourteen seconds filming, and it will serve him something that keeps him filming for fifteen.
That student is standing inside the most sophisticated memory architecture ever built by a nation-state. And he cannot see it. Because the architecture he actually lives in—the one that shapes what he remembers, what he attends to, what he treats as real—is not the memorial. It is the feed. The memorial was designed to make him stumble. The feed was designed to make him scroll. The memorial was built to interrupt. The feed was built to continue. Both are memory architecture. Both are political technology. One was designed to serve the truth. The other was designed to serve the metric. They are built from the same capabilities. They produce opposite outcomes.
That is the contradiction. Not between remembering and forgetting. Between two architectures of memory—one deliberate and one accidental, one governed and one autonomous, one built to hold the past in place and one built to generate a perpetual present—both constructed from the same human capacity to shape what populations know about the world they live in.
Germany built a machine to remember. The age built a machine that makes remembering and forgetting structurally identical. The same civilizational capacity produced both. The tool is always also the weapon. And the weapon does not know it is a weapon, which is what makes it so much harder to disarm than the AfD politician who at least has the decency to say what he is doing out loud.