On May 27, France's National Resistance Day, the French government released a video recounting the life of a Resistance member during the Occupation. Created with the help of artificial intelligence (AI), the video contained historical inaccuracies. Notably, it ended with a scene of the Liberation featuring a soldier from the Wehrmacht, the German army during World War II.

The video was criticized for blurring the line between truth and falsehood and was subsequently withdrawn. A few weeks earlier, the government's communication service had already used AI for commemorative purposes: An "artificial" post on TikTok commemorated women's right to vote. It is still online. The post depicted a stylized France in 1945 that was devoid of any trace of war. Among other inconsistencies, it showed a "France-Journal" newspaper that never existed.

To understand these absurd representations, it helps to recall how generative AI works. These systems are created using massive datasets, following a training phase in which statistically significant patterns are identified. These patterns are then used to generate text, images, and videos. It's a probabilistic logic; the systems are stochastic parrots – unpredictable by nature.

Generative AI tools can produce documents that mimic traces of the past. Thus, they introduce a new kind of mediation between us and our history that is fundamentally different from the work of archivists and historians. For instance, a prompt to generate a video about a French Resistance fighter will produce patterns related to World War II. The appearance of a cheerful Wehrmacht soldier would be statistically relevant but entirely incoherent in the context of the Liberation.

You have 51.21% of this article left to read. The rest is for subscribers only.