Images, AI, and Historical Misinformation: Navigating New Challenges in History Education

[Un article de The Conversation écrit par Mathieu Marly – Responsable éditorial de l’Encyclopédie d’histoire numérique de l’Europe (Sorbonne Université – Éducation nationale), agrégé, docteur en histoire contemporaine et chercheur associé au laboratoire SIRICE, Sorbonne Université ; et Gaël Lejeune – Enseignant-chercheur, Sorbonne Université]

Each day, images related to historical events are put online without being referenced – with their author, their date, their location, their place of conservation – and even less contextualized by a historical comment. This is the case, for example, of this photograph, most often published to represent the abuses of the “hands cut” in the Congo of Léopold II, at the turn of the XIX^e XX^e century. In 2024, it is found on thousands of web pages without the author of this photograph being always mentioned and without being the subject of an appropriate historical comment.

This photograph seems to illustrate the abuses committed by rubber companies in the Congo in a striking way. But what can we see of this photograph without knowing its history? The spectator is constrained here to interpret the photographic message to the filter of his own representations, seized by the contrast between the victims and the Europeans dressed in white to the colonial helmet, which seem to justify the corporal punishments by their hieratic installation.

In reality, this photograph was taken in 1904 by a Protestant missionary, Alice Seeley Harris, to denounce this violence and the two men on photography participate in this act of photographic resistance which will help to mobilize European public opinion against crimes committed in the independent state of the Congo. The identity and the intentions of the photographer are not a detail here: they account for a more complex historical reality, that of a “moral polyphony” of European societies at the end of the 19th century^e century, divided on the merits and the drifts of colonization.

A digital fog of decontextualized images

Examples like this one, there are thousands of them on the web, the publications and sharing of images generating a fog of decontextualized photographs, made viral by the algorithms of search engines and digital social networks.

Take the example of this tweet from Eric Ciotti posted on July 16, 2024 in commemoration of the Vel d'Hiv roundup:

The posted photography does not have much to do with the roundups of July 16 and 17, 1942: it is actually a snapshot showing the French suspected of collaboration locked up at the Vel d'Hiv after the Liberation. This Twitter account is not the only one to reproduce this error; It should be noted that Google Images algorithms have long placed this photograph in the first search results as a result of keywords “Rile of HIV Vel”.

Is the rectification of this error only the affair of historians concerned with the identification of sources? In reality, this error contributes to the historical ignorance of the Vel d'Hiv roundup. As historian Laurent Joly shows, there is only one photograph of the roundup, taken on July 16, 1942 for propaganda purposes and yet never published in the press. This detail is not trivial, he reveals that the German authorities have prohibited the publication of the roundup photographs, alerted by the disapproval of the Parisian population.

A challenge for teachers

These few examples must alert us to the illustrative use of photography still too present in school edition. In the absence of space, the textbooks are content, most often, of a simple legend without comment to shed light on or confirm the teacher's course.

The use of photographs by historians has however evolved in recent years, now considering it as real archives to which the basic rules of sources must be applied. Such use would certainly benefit from being generalized in the teaching of history to make students aware of documentary criticism – most often summarized by the Sandi method (source, author, nature, date, intention). Because, if this method is sometimes considered artificial by the students, it finds a justification, so to speak immediate, in the criticism of the photographic archive.

Indeed, the gaze carried by the students on the photographic image changes radically once known its history.

This documentary approach is all the more necessary since students and students are now informed on social networks, networks where photographs are relayed by army of unscrupulous methodological accounts and sometimes oriented by conspiracy readings of the past.

We must also add another data to understand the educational issue that awaits history teachers in the coming years: by 2026, according to a Europol report, the majority of the content available on the web will be generated by the 'Ia. This will probably imply the publication of false photographs increasingly credible and sophisticated, which will take place of evidence for fictions disguised as history.

The proliferation of generative AI, the acceleration of exchanges of invented, diverted or decontextualized photographs constitute a real educational challenge. How to teach the history to students without transmitting the tools to confront online historical disinformation? How to explain to students the digital environment in which they are immersed (AI, algorithms, vitalities of images) without proposing a history course which is also a digital history course?

A project to combat the virality of historical disinformation

To respond to these challenges, the ENCLOSE ENCLOPEDY DIGURE OF EUROPE (EHNE-SORBONNE University) and the teams of computer scientists from CERES-SORBONNE University are developing a new tool that targets a public of teachers as well as publishers And journalists: the Virapic project, a digital platform whose objective is to identify viral photographs (reproduced online on a very large scale and/or on a reduced period of time) when they are invented, diverted or decontextualized historical events that they claim to illustrate.

The goal is double. It is a question of injecting historical content around viral photographs (source, legend, historical commentary) and of analyzing the digital viralities of photographs (which publishes them? On which supports? With what temporality?). The Virapic project addresses above all the problem of historical disinformation by a pragmatic approach: to fight against the virality of disinformation by the referencing of historian work on search engines.

The originality of this tool is indeed due to the possibility of acting directly on the practices of Internet users thanks to the referencing of the Ehne encyclopedia whose web pages appear in the first results of search engines. Internet users thus looking for photographs to illustrate historical events will see the ehne/virapic web pages appear in the first search results like Google Images.

By constituting a base of referencing of viral, diverted, decontextualized or invented photographs around historical events, the Virapic project will allow quickly access to solid and critical historical content on the images that students, teachers or publishers wish to publish online online Or use in class.