Interviewer’s note: Wulf Kansteiner (1964–2025) was a Professor at Aarhus University. This interview was conducted in June 2025, shortly before his sudden death, and is now published posthumously.[1] I was deeply saddened to learn of his passing, and I feel honored to have had the chance to conduct this interview and exchange such thought-provoking ideas with him.
Wulf Kansteiner specialized in memory studies and contemporary European history, his scholarly work focused on cultural memory, particularly how it is shaped by visual media such as television, film, and digital platforms. He was also deeply engaged with post-narrativist approaches to historical theory and has contributed extensively to research on Holocaust and genocide memory, history, and historiography.
Kansteiner pursued his academic training in history and German studies at Ruhr University Bochum and later at the University of California, Los Angeles. Over the course of his career, he held teaching and research positions at institutions including the University of Tennessee, Kent State University, Friedrich Schiller University in Jena, and Binghamton University (State University of New York), before joining Aarhus University in 2014.
He played a leading role in advancing the field of memory studies, having served as President of the Memory Studies Association (MSA). He was also a co-founder and co-editor of the journal Memory Studies (SAGE) and an active member of the international research network Mnemonics. Kansteiner led and contributed to several major research initiatives, including the Velux Foundation-funded project SoundTrak and the EU Horizon 2020 project UNREST, where he was responsible for a key work package.
In addition to his research and editorial work, Kansteiner served on multiple academic advisory boards, such as those of the Leibniz Research Consortium Value of the Past, the History of the Ruhr Foundation, and the International Network for Theory of History. He was also on the editorial boards of several academic publications and series, including Media and Cultural Memory (De Gruyter), Memory, Heritage and Public History in Central and Eastern Europe (CEUP), and the Cambridge University Press journal Memory, Mind & Media.
Key publications include the monograph In Pursuit of German Memory: History, Television, and Politics after Auschwitz (Ohio UP); the co-edited volumes Agonistic Memory and the Legacy of 20th Century Wars in Europe (Palgrave), Probing the Ethics of Holocaust Culture (Harvard UP), Den Holocaust erzählen? Historiographie zwischen wissenschaftlicher Empirie und narrativer Kreativität (Wallstein) and The Politics of Memory in Postwar Europe (Duke UP); and the articles “Digital Doping for Historians: Can History, Memory, and Historical Theory be Rendered Artificially Intelligent?”, “History beyond Narration: The Shifting Linguistic Terrain of Timothy Snyder’s Bloodlands,” “The Holocaust in the 21st Century: Digital Anxiety, Transnational Cosmopolitanism and Never Again Genocide without Memory,” “Genealogy of a Category Mistake: A Critical Intellectual History of the Cultural Trauma Metaphor,” and “Finding Meaning in Memory: A Methodological Critique of Collective Memory Studies.” Texts are available at https://www.researchgate.net/profile/Wulf-Kansteiner-2.
Kansteiner’s latest work dealt with the relationship between Holocaust memory and postcolonial memory, the role of sound and artificial intelligence in processes of social remembrance and social forgetting, the concept of agonistic memory, the relationship between academic history and the interdisciplinary field of memory studies, and the narrative and logical structure of professional academic writing.
Interview
Johanna Vollmeyer: At the Complutense University of Madrid, we lead the research group LEETHI, within which we have been developing the project REC-LIT Cultural Recycling: Transliteratures in the Postdigital Age. The project focuses on strategies and models of literary text recycling in hypermedia environments, as well as on the potential for the reappropriation of the textual-literary at a time when users increasingly take hypermedia products as their primary cultural frame of reference.
From this starting point, a number of fundamentally ontological questions emerge—questions concerning our being-in-the-world and the ways in which postdigital media and the broader postdigital condition shape this mode of existence. As is well established, identity and memory are deeply intertwined with our being-in-the-world. Today, I would like to reflect with you on how this condition is undergoing significant transformations in response to contemporary digital developments.
So let me begin by expressing my sincere thanks for your time and attention.
My first question concerns younger generations, as they are the ones who primarily receive information through digital media. I believe this shapes their consciousness of the world and, more specifically, their historical consciousness. I would like to know your opinion on how the consciousness of young people—who seldom consult sources beyond social media, if at all—is changing. How is their perception of history being shaped by this media use? And what impact do more recent AI-based sources have on this development?
Wulf Kansteiner: It is true that social media have a major impact and are certainly shaping entire generations in ways that we are now considering across different countries. There seems to be an emerging understanding that the kind of access to social media we have allowed young people may be problematic. I share this concern to some extent, although I would immediately qualify it: I don’t think that this is a problem of the digital in and of itself.
Rather, I think we do not know enough about the digital—specifically, how we have structured our transition from an analogue electronic to a digital media ecology. This transition has been framed in terms of power—organized within a neoliberal, highly capitalist framework. And only now, somewhat belatedly, are we recognizing how deeply problematic that framework is.
This has profound implications for our memory culture—with our post–Cold War memory culture, which, in my view, has naturalized the neoliberal-capitalist framing of our being-in-the-world. That naturalization, to use an ironic term, has proven to be very expensive for us as human beings.
One must immediately add that all media ecologies come with strengths and weaknesses. Social media platforms are not neutral containers of content. They can be fantastic in some respects and deeply flawed in others. But the core issue right now may be the way we have organized digital mediation: in very specific and problematic terms. We know very little about how to organize a digital culture in terms of public access and public mediation—such as we had during the electronic age with public television and radio. Those older forms of mediation were very powerful. Now, we are lacking these and rely on platforms that are highly profit-driven, centralized in private corporations, and largely unregulated. We are only just beginning to assert control over them.
That is our specific challenge at the moment.
So we should be cautious about making definitive statements about digital culture. Nevertheless, I believe we can speak to some of the concrete dangers concerning memory, education, and historical consciousness—especially among young people using specific platforms. These include a lack of democratic access and control, and a lack of transparency. Media literacy is crucial—but it is very difficult to teach without these fundamental structures in place.
So there are some clear problems, although they are not necessarily the ones that are often raised, addressed, or reflected. For instance, the question of facticity — I am thinking of deepfakes— does pose problems, I agree. But not in the way we generally assume. I would phrase these risks and dangers differently than some other theoreticians do, contextualizing and conceptualizing them differently.
JV: Am I right in hearing a critique here that many researchers are not addressing the fundamental problems, but instead focus on issues like deepfakes, as if these were the core challenges of digital media and the digital media ecology—while neglecting the actual underlying problems?
WK: I think it’s a common tendency to essentialize new phenomena—trying to fix their risks and benefits too quickly. Interestingly what’s often overlooked in this process is our personal relationship to media. But what we really need is to take a historical perspective. We need to compare transitional periods. We’re clearly in such a transitional period now—a radical shift. You can observe it even in the last six months with the breathtaking development of AI.
Of course, it is a problem that these platforms are “great at lying” and have a problematic relationship to historical truth. But what also emerges—what always emerges during a shift in media ecology—is a new understanding of previous ecologies. We suddenly realize we didn’t understand or underestimated their limitations. Now, we can appreciate those earlier parameters more clearly.
One insight I’ve gained is that factuality is not, in and of itself, a sufficient value for memory cultures. That’s a painful realization for historians. Philosophers and theorists have said this for decades, but it hasn’t changed historical practice—or the ways memory cultures are produced, especially in visual media. We assumed that honoring facts would lead to a decent memory culture. We’re learning now that this isn’t true. A memory culture can be factually accurate and still disastrous—or it can lack factual integrity and still be ethically responsible. This is also thinkable.
This teaches us a difficult but important lesson: there is a difference between facts and interpretations. They are interconnected but fundamentally distinct. Good facts do not necessarily produce good interpretations, and vice versa. I call this the factual vacuity of interpretation and the interpretive vacuity of facts. Facts themselves don’t tell us how to interpret them, and interpretation based on lies can still have ethical integrity. That’s a very provocative idea, but I think it’s one we need to understand. This applies to how we negotiate and engage in learning processes. It is a conversation we should have with younger generations, focusing on their concrete use of digital media and on where boundaries might need to be drawn. Because we need to establish limits—especially with regard to digital lying since the scale and speed at which lies spread is deeply concerning.
At the end of the day, I’m more frightened by harmful interpretations than by a lack of factual integrity. Take the example of Donald Trump in the U.S.: what makes him so dangerous is not merely his lies, but his complete lack of ethical integrity. His values are terrible. Yes, he lies constantly, but the real issue is his disregard for democracy and basic responsibility. As a side note, he is also consistently lying. In this case, these elements are combined, but they do not necessarily have to be.
JV: Would you say that digital or postdigital media favor people who lack ethics? There have always been unethical people, but perhaps the dynamics are now different. Where do digital media come into play?
WK: I’m very cautious about making essentialist claims that digital media inherently promote unethical behavior. It’s not the media themselves, but how we’ve framed, structured, and powered them. Let’s take an example: in analogue electronic media, we had gatekeepers, for better or for worse. Many people lacked direct access to television—then the dominant medium—and access was regulated. Some structures were unfair, others necessary. But they were always about power.
Now, we have a strange historical combination: on the one hand, decentralization, which can be democratizing—people can participate without needing a TV channel. But on the other hand, we have an extraordinarily centralized structure of power in the hands of private actors with dubious motives—mostly commercial. That combination is a recipe for disaster.
This is why I now advocate for media regulation. You could call it censorship if you like—but I would argue that we need democratically legitimized institutions with the power to enforce ethical standards on platforms. These decisions—about what content to remove, what lies to eliminate—are incredibly powerful and must be taken by legitimate public bodies.
At present, the entities setting the boundaries of digital discourse have no democratic legitimacy. The only major player with some legitimacy is the European Union, and even that is limited. We need public boards, elected institutions, and open debate. Young people should be part of this. They should vote on who serves on these regulatory boards—even at ages 14 or 15—because they experience digital media in ways older generations do not. But we are far from that reality. At the moment, our most powerful moderators are Silicon Valley companies and the Chinese government. That is unacceptable.
If we leave decisions about collective memory and forgetting to the likes of Elon Musk, we are heading for disaster—not necessarily because he is a bad person, but because it is simply a bad idea to give that much power to one individual or one group.
JV: You’ve raised several key points here. You mentioned gatekeepers like traditional TV, where everyone watched the evening news and had access to the same information. That shaped a shared information ecology. Now, with personalized platforms and algorithms, information is individualized. As McLuhan said, “the medium is the message.” Today, we have decentralization of content alongside centralization of power in private hands, which has a massive impact on checks and balances. You spoke of the need for limits—and I agree: only within limits can freedom exist. Without limits, there’s no protection for minority groups.
WK: Absolutely. Minority representation remains a challenge in digital media ecologies. That’s where we need to look closely—especially with AI. Even in the past, our gatekeeping systems were flawed. People may have consumed similar products, but their memory cultures differed. Beneath the surface, many alternative or even ugly memories were active. Public media were never perfect.
So we must invent ways of transforming mediation and limit-setting into an ongoing process of self-reflection. It will never be finished or perfect. But it is necessary. Each year we should ask: what injustices have occurred through censorship? But also how much injustice has been done by failing to capture, safeguard, and protect certain segments of the population—particularly minorities? This will never go away—and perhaps never did.
JV: I think Germany is a clear example for these “ugly memories” that you have mentioned. Despite a seemingly successful memory discourse, we now face the rise of the far-right party AfD (Alternative für Deutschland). This cannot be attributed solely to digital media. As you said, these currents were always there, just beneath the surface.
WK: Absolutely. Our research shows that beneath the apparent consensus of cosmopolitan memory, there were pockets of Nazi nostalgia and antisemitism. These are now visible on social platforms. And perhaps we should welcome their visibility—it gives us a chance to address them.
JV: So where is memory headed? These layers are now visible, and that can be an opportunity. What threats and opportunities do you see—especially with AI? And how should Memory Studies critically respond?
WK: In our current context, artificial intelligence is being used in highly problematic ways, essentially functioning as a form of digital communication “on steroids.” The speed and reach of communication—and the reproduction of memory structures—have increased dramatically, but these processes are often deeply troubling. For instance, certain types of interpretations, such as right-wing narratives, can be seductive, politically charged, and yield consequences that, at least from my perspective, are undesirable. Thus, the issue of framing is once again central to the problem.
There are also challenges that may be inherent to the technology itself, and I believe it is crucial to recognize that we are still at the very beginning of this development. It is a revolution already underway, but nevertheless still in its early stages. I anticipate that certain aspects will continue to evolve. Data centers for AI are massive and energy-intensive, built by just three major players. The goal is to create AI that mimics human reflection. The current design of artificial intelligence involves a peculiar combination: it fundamentally operates on probabilistic calculations, followed by rather crude filtering processes. This approach represents only an initial step. At present, the machine’s reliance on probability, combined with subsequent filtering, tends to produce homogenous and often problematic outcomes. However, it remains to be seen what forms of reflective capacity might emerge with the development of new machines, particularly as quantum computing technology advances. It is conceivable that future AI systems could be embedded within programs capable of relatively sophisticated ethical self-reflection and self-criticism.
It’s too early to judge whether this will be ethically disastrous or beneficial. Humans are pragmatists and often fail to follow their own ethics—we need institutions to help us stay accountable. Maybe machines, if designed with reflective ethics, can do better. It could very well be that artificial intelligence is much better in, for example, governing scarce resources and providing people with access to resources in a way that is much fairer.
At present, we are confronted with significant challenges due to the rapid proliferation of highly problematic content. We also face a complex situation—one I am not yet certain should be regarded as entirely disadvantageous. We have already entered a state in which memory cultures are no longer exclusively human; a substantial portion of the texts that the average individual in digitized societies interacts with and consumes daily is AI-generated. Thus, memory culture has become partially nonhuman, representing a form of human–nonhuman collaboration. Whether this hybrid condition constitutes a problem ultimately depends on its impact—for example, on the survival of the planet and perhaps also of the species, depending on one’s anthropocentric perspective. The verdict on whether this is ultimately problematic remains open.
What we can assert with confidence is that the current manner in which AI is being used poses problems. Presently, forces opposing environmental protection and an ethics of care are more strongly entrenched within AI systems than their opposites. This imbalance largely stems from the fact that these systems are built around short-term profit rationales.
JV: Human beings tend to think hermeneutically, whereas machines—leaving aside quantum computers—must sequentialize and essentialize the world to make it computable. This is fundamentally different from how we act. It’s as if machines must “freeze” an image of a person or event in order to process it, whereas humans operate within indeterminacy. This indeterminacy fosters development, because it allows for open possibilities. That’s a big contrast with what AI does. So do you really think that AI can provide us with so many opportunities and foster a positive development?
WK: I’m not convinced that the human track record gives us much to celebrate. Just look at the centralization of power, industrialization, and the ecological damage we’ve done. Our species has often made poor collective decisions.
Indeed, there seems to be different between quality knowledge and quantifiable knowledge and that’s what your dichotomy is kind of based on. But I’m not sure it will hold in the future.
More powerful computing might one day approximate forms of ethical reasoning or creativity that we’ve traditionally considered uniquely human. We’re just at the beginning. Judging AI now is like evaluating the car industry based on prototypes from the 1890s. I’m sceptical of the claim that powerful machines can never match human-level qualitative understanding. We’re already seeing that AI can anticipate and replicate many forms of human behavior—at least collectively.
And here’s the question: what are we missing in this process? Memory scholars often assume that good memory fosters good societies—democratic, sustainable, ethical ones. But what counts as a “good” memory is under question. Take Holocaust memory: we believed it would promote ethics and care, yet these memory cultures are sometimes co-opted in ways that reinforce racism—conscious or unconsciously.
Still, one thing we do know—at least about humans—is that identity is built through narration. Narratives follow patterns, and AI is extremely good at recognizing and generating patterns. If ethical or “good” memory is based on strong narrative patterns, then AI could be a powerful tool. Today’s language models can already reproduce genres and narrative structures with impressive fluency. So if patterns is at the heart of our good interpretations then this is a potent tool.
JV: That raises a crucial point: semantics. In terms of semantics, computers still have a long way to go—even though they’re improving rapidly. However, semantics is a vast field of research, and machine capabilities remain quite limited compared to human understanding. This is why I remain cautious. I agree, patterns matter, and AI is becoming more grounded, but their grasp of semantics is still shallow. That’s why we sometimes end up with discourses that feel overly simplistic.
WK: But who decided that simplicity is bad? You’re right to highlight semantics. The question is whether computers are inherently limited in semantic depth due to their architecture. Many would say yes, since quantification is their foundation. That is the computer’s DNA.
But consider how societies function: powerful interpretations of the past are rarely semantically rich. Identities—even those key to democratic societies—are not highly complex in practical terms. Complex ideas must be translated into simpler narratives for democratic engagement. If they remain too intricate, they lack political traction. That’s a human limitation. So, I don’t think semantic simplicity necessarily works against machines.
In fact, digital culture and AI might help us include more people in decision-making processes by making core values and stories more accessible. Complexity isn’t inherently virtuous—sometimes it paralyzes action. And in our current crises, we can’t afford inaction.
Take Anthropocenic memory as a test case. It doesn’t take a scientist to understand we’re using resources unsustainably. That’s a simple truth. Yet we’ve failed to turn that truth into policy or lifestyle changes—even in societies that claim to have a well-developed ecological conscience, like the Scandinavian countries. There’s a disconnect. And the problem isn’t a lack of understanding or complexity. But we need to change.
JV: That makes me think: perhaps we’re focusing too much on what machines are or can do—and not enough on the narratives we construct about them. We tend to anthropomorphize them instead of seeing their unique capabilities and limits. Maybe we need new narratives that don’t just mirror us but clarify where machines excel and where we do.
WK: Exactly. That’s what I’ve been trying to emphasize. One advantage machines might have is consistency. Compared to human culture, machines could follow values and goals more coherently. Humans often build beautiful memory cultures—like ethical commemorations—then act in ways that contradict them. We admire the memory but ignore the lessons. Machines might avoid that inconsistency.
Ironically though, as machines become more “human-like,” they might also become less consistent. That’s possible. But I don’t feel equipped to make that judgment yet. So I choose to remain hopeful.
JV: You mentioned narrative’s importance to memory. With the rise of strongly text-centered, generative AI – think of Large Language Models – , would you say we’re experiencing a new linguistic turn? We’ve had the visual turn before. And AI is also becoming better and better in producing images. Since you are also an expert in visual media I would like to know from your side if you would agree with this statement that we living now a new linguistic turn.
WK: I think this is just a temporary phase. It has to do with technological development. There’s a kind of window, and we don’t know how long it will last—but my guess is, not very long. Especially in the academic world, there is a return to the text. Many academics never really transitioned into visual culture. Think of historians: they never made the shift. We’ve never reached a point where the next generation of historical researchers delivers their PhD thesis in the form of a documentary film. We are still writing texts.
So, many academics, including historians, are quite happy right now, because suddenly text and textual qualities have regained value. Artificial intelligence is really good with texts—but that’s just a phase. That’s because texts, compared to other types of cultural mediation, are simple. They can be counted, measured, calculated. That’s why text is beautiful—and indeed, we’re seeing a wave of research institutions returning to text analysis through digital means. That’s perfectly fine.
But when it comes to collective memory and popular culture, we will never return to text alone. It’s not going to happen. Social media platforms—the powerful ones—already operate on different layers of communication. Take TikTok or Telegram: they are fabulously multilayered, and also, very ironic. That’s important. People are using these platforms to subvert the old, unironic gatekeeper culture. This new culture is highly visual and highly ironic.
There’s a constant tension between different media layers—between text and image—and this tension is exactly what people are working with today. I don’t think you can build a popular platform for the crafting of collective memory that works only textually. If you want to reach young people on these platforms, visual elements are essential.
Very soon we’ll be able to produce film with artificial intelligence at the level we currently produce text. Right now, we’re limited by computing power. How long that limitation lasts—I don’t know. But the machines being built are specifically meant to break that barrier. The goal is to generate visual culture with AI that is as compelling as what we already consume. In some media contexts—like video games—it’s already happening.
And that’s the metric: it’s always about whether people feel the image is “authentic” or “believable.” There are fantastic opportunities here. Imagine developing a visually and ethically compelling piece on the Holocaust. That still hasn’t happened because we haven’t made the transition yet.
JV: I thought of that when I saw the Sophie Scholl Instagram account. To me it seemed like a missed opportunity. From a media perspective, it was technically well done— with these videos where Sophie is filming herself, which of course was impossible in the 1940s and, however, there was no friction or incoherence in this media use. But the message conveyed was ethically questionable.
WK: I absolutely agree. That example really highlights the tensions between two media ecologies. In the old analog or early electronic gatekeeper culture, irony and sarcasm were excluded. In the new media ecology, irony and sarcasm are the focal points. To work today, Holocaust memory culture would have to include irony—and that was, of course, completely taboo for the creators of that account. They simply took an analog narrative and imported it into a digital environment. Maybe it works for us “older” people—we recognize it as a kind of television miniseries. But it doesn’t resonate with the younger generation crafting and sharing images. They’re working with irony, ambiguity, and undecidability.
It has long been assumed that irony is incompatible with serious topics such as Holocaust memory—yet perhaps this assumption merits reconsideration. The central question concerns how to navigate this transition responsibly. I respect the decision of major video game companies to refrain from creating Holocaust-themed games, as no one has yet devised an ethically sound approach to such representations. It may be that digital cultures are ill-equipped to address questions of cosmopolitan narration or the binaries of victim and perpetrator.
However, digital culture might be particularly suited to engaging with what Michael Rothberg terms implicated subjectivity. Digital platforms may be capable of negotiating irony and undecidability, especially within the ethical gray zones where roles and responsibilities resist clear definition. It is conceivable that digital media’s strength lies in its capacity to explore this implicatedness.
Consequently, there is a pressing need to establish explicit ethical boundaries within digital culture. For instance, is it ethically permissible to assume the role of a mass murderer in a digital game? Should such portrayals be prohibited? Alternatively, might it be acceptable to simulate the biography and process of radicalization of a perpetrator, but only up to the point before any atrocities are committed? This demarcation—the simulation of all actions except the moment of violence—could constitute an ethical boundary. These questions are crucial for determining the limits that digital media must observe.
JV: In the specific case of Holocaust memory, there’s another issue: we know how it ends. There’s no good ending, no escape. That makes it extremely difficult to create a game around it.
WK: A compelling game might move players from the collaboration end of the spectrum to the resistance end. There are independent games that explore this gray zone. This War of Mine is one of them.
JV: I would like to turn to the question of artificial intelligence in research. You have noted the importance of simple patterns and narratives in memory culture—an observation with which I concur. However, it is equally imperative that we critically reflect on our own practices as researchers. The increasing integration of digital tools and AI across disciplines—including history and memory studies—has brought a pronounced emphasis on quantitative analysis. For instance, within literary studies, numerous colleagues prioritize digital text analysis to the extent that qualitative interpretation is often marginalized. But the outcomes of their analysis and interpretations of, let’s say the Gimm’s fary tales are quiet poor. While I do not oppose the use of such tools, I maintain that rigorous qualitative interpretation remains indispensable.
WK: There are two critical points to consider here.
First, we are currently gaining new insights into our own research practices across diverse media ecologies. The ability to analyze texts such as the Grimm’s Fairy Tales quantitatively and in great detail does not, in itself, guarantee that the results are meaningful. While such analyses yield robust data—such as the frequency of specific words or their evolution over time—these results invariably require interpretation. Importantly, interpretation is always situated within particular perspectives, which are underpinned by ethical and political frameworks that cannot be derived from the texts themselves.
This dynamic is also evident in digital history. As new subfields emerge, there is often an assumption that factual data alone suffice. Although it is possible to generate extensive factual information, facts in isolation possess limited interpretive and ethical significance. Therefore, the crucial question becomes: from which perspective, and with which underlying value system, do we interpret these findings?
Historically, researchers have often been reluctant to make their values explicit, perceiving such disclosures as unscientific. Yet this reticence is not unique to the humanities; even the natural sciences have witnessed the production of excellent research that has served terrible masters. Within digital humanities, a comparable misconception persists: a naïve belief that the mere generation of facts guarantees sound research. This was a mistake made in previous media ecologies, and it is being repeated. However, research and memory are fundamentally about values.
Second, digital media—and particularly artificial intelligence—offer new possibilities, including within hermeneutic practices. AI enables us to construct ethically sensitive counterfactual scenarios previously beyond reach. For example, in examining moments of extreme violence, we have traditionally relied on survivor testimony, which is invaluable but constrained by ethical considerations that prevent certain lines of inquiry. AI, by contrast, can simulate such moments based on extensive source data, providing novel perspectives.
These inquiries are fraught with ethical risk: for instance, exploring the mindset of individuals on the verge of committing mass atrocities. It would be unethical to question survivors in this regard due to the risk of retraumatization, yet AI might allow us to probe this space. Similarly, AI can be tasked with generating disturbing right-wing historical interpretations—not to endorse them, but to understand their appeal and structure. In some cases, AI-generated extremist narratives may prove more persuasive than those offered by human interviewees, yielding valuable insights.
Thus, while these tools entail significant risks, they also permit textual—and soon visual—experiments that would have been ethically untenable previously. This capability is especially pertinent to the memory of the Anthropocene, where counterfactual thinking—imagining alternative futures or preventable outcomes—is central.
If AI enables the construction of counterfactual narratives with ethical sophistication, it holds considerable promise. We have yet to determine how to employ these technologies responsibly, but this challenge lies at the heart of scholarly inquiry. It also underscores the necessity of maintaining research environments that afford intellectual freedom—including the freedom to fail. Unlike popular memory culture, where public failures are intolerable, research demands the capacity to experiment, evaluate results, and, if necessary, conclude that certain findings should not be disseminated. This is the essence of rigorous scholarship.
Memory culture at large may not tolerate widespread errors; here, a potential disconnect emerges. While this may sound elitist, it is less about elitism and more about scale: errors must be confined to the relatively small scale of research rather than proliferating broadly within public discourse.
JV: That connects to your earlier point about memory not being fact-based. I’m a literary study scholar and literary products are appreciated as conveying a kind of truth, despite being fictional text. They might convey more “truth” than a fact-based historical text.
WK: Yes, but this concerns a different mode of truth—an interpretive, narrative truth. In fact, I would argue that the term “truth” itself may be somewhat misleading in this context. We are not primarily addressing empirical facts; rather, we are engaging with ethical truths—values to which we consciously or unconsciously adhere. Accurate factual information does not inherently produce sound or meaningful interpretations. As you rightly noted, storytelling conducted with integrity is far more likely to cultivate a meaningful culture of memory. The factual basis of a narrative may be secondary—whether it is fictional, counterfactual, or even predicated on errors. What ultimately matters is the integrity and ethical coherence of the narrative itself.
JV: What about the limitations of AI, particularly regarding interpretation? For instance, in the case of the Grimm tales, AI can perform quantitative analyses such as word counts and pattern recognition, yet it remains incapable of interpreting words with the nuanced understanding that a human can provide. This is especially true in poetry and fiction, where words carry multiple layers of meaning and connotation. Currently, AI excels in syntactic processing but remains limited in semantic comprehension. While this may evolve over time, it is essential to remain attentive to the current capabilities of these tools and to continuously reassess their potential as they develop.
WK: You raise a compelling technical issue, and although I do not have a definitive answer, I can outline the problem. Currently, machines lack what might be called vertical memory: they calculate probabilities but cannot specify the exact data underpinning those calculations. Achieving this level of transparency would require computational resources far beyond what is presently available. In the future, however, machines may be capable of greater transparency—disclosing the specific texts they were trained on, their age, and their provenance.
Such transparency would represent a significant shift. It would allow us to compare AI systems and conduct ethical evaluations, enabling judgments such as whether one system produces more ethically sound outcomes than another. At present, these models function as black boxes—not only to the general public but often even to experts in the field.
This situation may evolve, positioning us better to evaluate, curate, and shape AI tools responsibly. It might even become feasible to train large language models exclusively on texts demonstrating high interpretive integrity. Although no one has attempted this yet, and it remains uncertain whether it would succeed, the possibility warrants exploration.
JV: Thank you very much, Wulf, for taking the time to answer these questions. You’ve shared so many valuable insights—we truly appreciate it.
[1] My warmest thanks to Wulf Kansteiner’s wife, Sonja Wolf, who kindly granted permission for us to publish the interview.
