History is never simply an accumulation of dates, documents, and testimonies. It is also the result of a process of selection, interpretation, and transmission. For centuries, historians, archivists, teachers, journalists, and cultural institutions have played an essential role in shaping our shared memory. But a new player is now capable of intervening on a considerable scale: artificial intelligence.

With generative models capable of producing texts, photographs, videos, and artificial voices, the relationship between AI and collective memory is becoming a societal issue. These technologies can facilitate the exploration of vast archives, restore certain documents, or make cultural heritage more accessible. They can also produce perfectly plausible representations of events that never actually happened.

The real issue therefore goes beyond the simple question of whether a machine can tell history. It is about determining how we will continue to distinguish a historical document from a reconstruction, a hypothesis from an established fact, and an illustration from evidence.

This issue, at the heart of Léwis Verdun’s mini-book How Will AI Rewrite History?, leads us to a broader question: in the age of generative content, how can we protect the reliability of our collective memory without giving up the possibilities offered by artificial intelligence?

When Archives Meet Artificial Intelligence

Archives represent an immense source of material for understanding past societies. Libraries, museums, government institutions, and research centers preserve millions of photographs, manuscripts, newspapers, recordings, and documents, only a portion of which have been studied in depth.

In this context, artificial intelligence and history can form a particularly fruitful partnership. Computer systems can help researchers classify large quantities of documents, detect connections, transcribe texts, or facilitate searches within digitized collections.

The goal is not necessarily to replace the historian, but to enhance their capacity for exploration.

A machine capable of rapidly analyzing several thousand documents can, for example, uncover connections that would have required a considerable amount of time for a human team to identify. AI can also help make certain digital archives more accessible to the general public by facilitating their indexing, translation, or contextualization.

This development opens up interesting possibilities for museums and education. An exhibition could offer different visualizations of an era based on authenticated sources. A teacher could gain easier access to documents from several collections. A researcher could query an enormous corpus without having to consult every document individually.

But speed of analysis does not guarantee accuracy of interpretation.

An algorithm works from data and models. If its data contains errors, outdated representations, or imbalances, its results may reproduce them. An AI-generated representation of a historical figure or population may therefore appear extremely realistic while being based on questionable assumptions.

This is precisely where human vigilance becomes indispensable.

A Realistic Image Is Not Necessarily Historical Evidence

We have long given photography and video a special status. Although image manipulation existed long before artificial intelligence, generative tools are profoundly changing its scale and accessibility.

It is now possible to create a historically plausible scene even though it was never photographed. A deceased person can appear to speak in front of a camera. An old street can be reconstructed in minute detail. A synthetic voice can give the impression that a public figure is saying words they never actually spoke.

Historical deepfakes therefore weaken a deeply rooted mental association: “I can see it, so it must have happened.”

The danger lies partly in realism. A poor-quality representation naturally encourages caution. A photorealistic image, accompanied by a credible caption and shared thousands of times, can instead acquire the appearance of documentary evidence.

Repetition then reinforces the phenomenon. When the same image circulates on social media, appears in a video, and is subsequently reproduced on different websites, it can gradually come to be perceived as a reference, even when its origin is artificial.

Future generations may therefore face a new difficulty: their digital environments will simultaneously contain authentic documents, restorations, educational reconstructions, creative works, and deceptive fabrications.

The question will no longer simply be: “What does this image show?” but also: “Where did it come from, who created it, from which sources, and for what purpose?”

This reflex is one of the foundations of the digital literacy that citizens and students will increasingly need.

Can Algorithmic Biases Alter Our View of the Past?

Generative artificial intelligence is not a neutral window onto history. Its outputs depend on the data used for its training, the design of the system, and the way requests are formulated.

Yet human archives themselves are incomplete.

Some populations have left behind enormous amounts of written or visual evidence, while others have been poorly documented. Certain narratives dominated institutions for decades. Testimonies have disappeared, languages have been marginalized, and inaccurate representations have sometimes been repeated until they became familiar.

Algorithmic biases can therefore amplify historical biases that already exist.

Imagine a system tasked with producing an image of an ancient population based on an unbalanced or scientifically outdated set of references. The result may be visually spectacular while still giving the public a false impression of that population.

The problem becomes particularly delicate when AI automatically fills in what it does not know. Generative models are designed to produce coherent responses, not to turn every uncertainty into silence. Invented information can therefore slip between several accurate facts and benefit from their credibility.

In the field of history, this ability to produce convincing “hallucinations” imposes an essential distinction between exploration and validation.

AI can suggest a lead. It can facilitate research or propose a hypothetical reconstruction. But a historical claim still requires identifiable sources, contextualization, and, when the subject demands it, human expertise.

International discussions on the ethics of artificial intelligence also regularly emphasize principles such as transparency, accountability, human oversight, and public awareness.

Applied to collective memory, these principles become particularly concrete: we need to know when a representation has been generated, be able to trace the sources on which it is based, and prevent a reconstruction from being presented as an original document.

Our Personal Memories Are Becoming Generative Too

The issue does not concern only major national archives or historical events. Artificial intelligence is also beginning to influence the way we preserve and represent our individual memories.

Based on a personal account, tools can create an image of a childhood place, visually represent a dream, or reconstruct a scene for which no photograph exists. These uses can have an artistic, educational, or even therapeutic dimension.

But they also blur a fundamental boundary: the one separating a memory from its representation.

Human memory is already not a perfect reproduction of the past. It evolves over time, with emotions and with the successive stories we tell about an event. If a generated image then gives that memory a highly convincing visual form, it could in turn influence the way the memory is recounted.

On an individual scale, the phenomenon is fascinating. On the scale of millions of users, it could change our collective relationship with traces of the past.

This is why AI and collective memory should not be considered solely from the perspective of misinformation. They raise a deeper question: how does a society construct its memories when an increasing proportion of the images it encounters may be synthetic?

How Can We Preserve Reliable Collective Memory in the Age of AI?

Banning all use of artificial intelligence in the field of history would be both difficult and counterproductive. These tools can offer genuine opportunities to researchers, teachers, cultural professionals, and citizens. The goal should instead be to establish practices that allow us to benefit from their advantages without confusing creation with documentation.

Here are some useful practices:

  • Identify the origin of content. A striking image should be traceable to a clearly identified source, author, institution, or method of creation.
  • Distinguish documents from reconstructions. An AI-generated representation should be identified as such, particularly when it concerns a historical figure or event.
  • Cross-check sources. An important claim should not rely exclusively on an answer produced by a generative model.
  • Examine the data used. Whenever possible, we should consider the quality, diversity, and age of the references used for a reconstruction.
  • Teach digital literacy. Knowing how to find the origin of an image, verify information, and identify signs of manipulation is becoming an essential cultural skill.
  • Maintain human oversight. Historians, archivists, researchers, and specialists remain indispensable for interpreting sources and their contexts.
  • Develop transparency rules. Institutions, media organizations, and platforms have an interest in clearly indicating when historical content has been generated or substantially transformed by AI.

For readers confronted with surprising historical content, a simple method is to ask four questions: What is the original source? Is the content authentic or reconstructed? Are there other independent sources that confirm it? What information is missing to understand its context?

These few seconds of verification can make a considerable difference.

The goal, therefore, is not to develop systematic distrust toward every digital image or text. Rather, it is to acquire a new culture of evidence suited to the age of generative content.

Collaboration Between Humans and Machines Could Also Enrich History

The risks associated with AI should not obscure its potential. Used appropriately, it can become a powerful tool for accessing cultural heritage.

It could facilitate the exploration of collections that were previously difficult to consult, help compare bodies of documents, encourage the translation of historical materials, or create educational experiences that allow the public to better understand a particular era.

The essential condition is to preserve the distinction between what we know, what we deduce, and what we reconstruct.

From this perspective, the future of artificial intelligence and history does not have to be reduced to a competition between humans and machines. It can instead be based on complementarity: the processing power of computer systems on one side, and human critical thinking, contextualization, responsibility, and expertise on the other.

Our collective memory has always evolved alongside its media: oral traditions, manuscripts, printing, photography, cinema, databases, and digital networks. Artificial intelligence represents a new stage, but with one major difference: it no longer simply preserves or distributes traces of the past. It can now produce new ones.

This is precisely why critical thinking, data quality, transparency, and media education are becoming decisive. The goal is not to choose between innovation and memory, but to ensure that the former helps enrich the latter without erasing its points of reference.

This reflection lies at the heart of How Will AI Rewrite History? by Léwis Verdun. Through the issues of archives, algorithmic hallucinations, deepfakes, biases, and the reconstruction of memories, this mini-book invites us to consider what we want to preserve at a time when machines are becoming capable of creating credible representations of the past.

Discover How Will AI Rewrite History? by Léwis Verdun now in the Nouveaux Horizons collection from FIVE ÉDITIONS.