Imagining the past — Generative AI as art, historical interpretation or fiction

8 min readFeb 11, 2025

While AI by now has made its entrance in the museum sector, discussions are growing about its use. One area of discussion concerns the use of generative AI in mediating historical events, people and places. A recent example is the exhibition Häxor (Witches), by artist Pompe Hedengren, currently on display at the Historiska museet in Stockholm, Sweden. Generative AI has here been used as a tool to visualize the past, and in this case a piece of very dark Swedish history.

Gertrud Svensdotter. Promptography by Pompe Hedengren.

The exhibition has been widely reviewed and praised for innovative exhibition aesthetics, and above all it has significanlty increased the visitor numbers to the musuem, indicating that it has reached a broader audience. At the same time, some reviews have specifically criticized the use of generative AI in the mediation of the witch trials:

“The exhibition ‘Witches’ at the Historical Museum mixes poorly executed AI images with sloppy historical representation. The finished result is not for those who are sensitive to details and historical authenticity”
By Karolina Uggla in Utställningskritik.

Or:

“The question is what AI adds to make us modern people understand the symbolic connection between witchcraft and feminism. On the contrary, the technology can be used to claim anything and is perfectly suited to fake news and the spread of rumors.”
Lisa Ehlin in Flamman.

The criticism in both reviews concerns, among other things, what is perceived as a lack of quality in the AI-generated images, and that what the viewer perceives as inaccuracies and unnaturalness can undermine the credibility and authenticity of historical exhibitions. Further arguments emerge that AI-generated images and props that do not match the period can lead to a superficial and inaccurate representation of history that can reduce its pedagogical value and credibility.

I ask Pompe Hedengren, producer, artist and artistic director, what he thinks about some of the strong reactions to the illustrations in Häxor:

– I understand them. They are used to a historical exhibition containing authentic objects with clear signs and order. But I have a completely different starting point. For me, the visitor is most important, not the antiquarian. I want to reach young adults, not the historian. I am competing with Netflix, not with the library.

In general, concerns and doubts about AI-generated images in the museum sector stem from, among other things, fears of copyright infringement and the risk of visual creators losing their jobs. Arguments are also raised, as in the case of Häxor, about the risk of lacking historical authenticity. The question of who owns the AI and how it is trained is also highly relevant.

However, Pompe makes an important point about the need for museums to reach new audiences. The question of generative AI and visual storytelling is highly topical and relevant and should spark a much broader discussion about the exhibition medium, and not stop at debating the authenticity of the images. Instead: How can story experiences create value for museums?

Pompe Hedengren has already shown the sector, with the previous exhibition Aquanauts, how immersive storytelling can become a gamechanger. It all begins with a story:

“Suddenly, nine suitcases and two wooden crates arrive at the Ontological Museum in Stockholm. The anonymous donation contains drawings, films, photographic negatives on glass plates, a peculiar skull, and dried scent traces. In September 1897, Beata Gardelius and Inga-Lisa Blomgren disappeared during an expedition to Siljansringen. The fate of these two women remains unknown to this day”

I asked Klaus Sommer Paulsen, founder and CEO of AdventureLAB , Denmark about the reactions to the Häxor exhibition. He agrees that the critical reviews are rooted in more than just the use of generative AI. The question that arises is fundamentally how to mediate history in museums, a discussion that can partly be linked to the fear of “ Disneyification ” and popularization. Something that Klaus also believes is a paradox since more and more museums are using spatial design principles from theme parks.

– It’s about integrating several new principles and methods that attract and engage the audience, in a balancing act with the museum’s voice, authenticity and credibility. It doesn’t have to be either or.

Though a development of the exhibition medium is a separate debate in itself, it is relevant to note the increased number of visitors to the Häxor exhibition and to the museum. Of course, the AI images have contributed here. Needles to say, this discussion should of course not be distinguished from the need for maintaining high-quality historical mediation, the very USP of museums. As Klaus suggests, museums can maintain their voices and still explore fiction and storytelling as tools to engage new audiences.

So perhaps the question to ask is not if the use of generative AI in the mediation of history is good or bad, but instead query:

In what ways can stories, art and fiction help audiences approach historical facts? In what ways can museums illustrate the past with the help of AI, while still maintaining their credibility? What value is created through AI-generated images?

There aren’t any simple answers as the use of generative AI in visualization is of course not a simple yes or no question. Visually representing history with generative AI is neither just about pressing a button, it is a process. As Pompe Hedengren describes the work with Häxor:

– I worked for two years and made a total of 4,000 images. Of these, I selected 50 images that I worked on further in eight different programs. So it’s actually 50 collages created from thousands of images. It’s a digital craft that requires patience.

This also raises other interesting questions: Can the creative process be part of the public mission of museums? In what way can the exploration of AI and the representation of the past also become part of co-creation methods at museums?

Kulturarvsresan — The Cultural Heritage Journey

Another example of using generative AI and visual communication of history is the project Smart Heritage Tourism in the High Coast , which I myself managed between 2021 and 2024. The project was facilitated by the Västernorrlands Museum in a collaboration with Danish AdventureLAB . The aim was to mediate cultural heritage stories from the region to visiting tourists, who are looking for information and experiences via their mobile phones. The Kulturarvsresan presents five media guides, each with eight stories, encouraging tourists to visit almost 30 places in the area. The newly produced stories invite the audience to explore history from the Middle Ages to the late 1600s. The stories are based on historical sources and facts and use fiction to connect with the audience and to evoke emotions.

Two of the media guides have been illustrated by professional artists, Sámi artist Per Elof Ricklund and illustrator and archaeologist Mats Vänehem.

Three of the guides were illustrated using generative AI. The media guide with the theme Water uses authentic illustrations from medieval fishing chapels along the High Coast, which have been processed through AI and then processed again by Klaus Sommer Paulsen in dialogue with antiquarians to achieve the final result. The illustrations of folklore creatures have been created more freely, following a vivid dialogue within the project team around the visual tonality.

Women marching towards the Häxberget (the Witches’ mountain) for their execution. The illustration is generated with Midjourney and processed in Photoshop and Adobe Lightroom.

One of the media guides depicts the witch trials of Torsåker in the 1670s and has been visualized using generative AI, entirely by the museum’s project team. The aim was to create a more direct visual appeal in all eight stories, to connect with the strong stories and capture the audience’s attention. The discussions in the team were about authenticity, not deviating from the creative strategy, or risking contributing to “dark tourism “, i.e. more sensational and where the deeper meaning and relevance are lost.

The result of the Kulturarvsresan in terms of visitor numbers is impressive. The media guides attracted the same amount of users as there were visitors to the museum’s permanent exhibitions during the summer of 2024. High quality stories and illustrations have contributed to this. But of course, questions remain: What does it mean that the images do not show a 100% accurate depiction of what actually happened? How much can the visual deviate from historical accuracy? And yes, there is plenty of room for further discussions here.

When there are no visual original sources

An important reason for using generative AI in the Kulturarvsresan was the lack of authentic illustrations, and the need for visual communication to lower the threshold for visitors to understand emotionally connect to what happened. It has also been a driving force for Pompe Hedengren, who describes his extensive preparatory work for the exhibition:

– I traveled to Älvdalen and waded out into the Österdal River right where Gertrud Svensdotter walked, herding the goats towards home. For this she was later sentenced to death, an event that triggered “The Great Noise”, the onset of major witch trials in Sweden. I traveled to Torsåker Church where Laurentius Hornaeus sentenced and led 71 women, men and children the five kilometers up to Häxberget where they were beheaded and burned. I discussed with historians, bishops and modern witches. After having learned about this very dark history, I wanted to honor these women in some way. It was frustrating that there were no drawings or paintings that portrayed them. There was no artist who depicted the countryside or the witch trials in Sweden, and the camera was invented 200 years later.

Illustration, Klaus Sommer Paulsen, AdventureLab. Made with Midjourney and processed with Photoshop and Adobe Lightroom.

What comes next

AI in its various forms can and will contribute to the development of museum mediation in the coming years. AI is here to stay, and the museum sector will benefit greatly by constructive and creative discussions about the mediaiton of history as well as the development of the exhibition medium. There are exciting examples of reconstructions or artistic interpretations of historical events, such as the short film Battalion by Dave Clarke.

Personally, I am not entirely comfortable with the combination of Pompe Hedengren’s innovative artistic creativity meeting the more traditional medium, the museum exhibition, with long formal texts, in the exhibition Häxor. This opinion is more about the challenges of the museum medium meeting the creative artist, than about the use of generative AI. I am more fond of the Aquanauts exhibition, which really went “all in” with an innovative immersive experience format and storytelling down to the smallest detail. There is so much here to be inspired by, in the museum sector.

The fact however that Häxor has attracted a large audience at Historiska museet, makes the use of generative AI and the development of the exhibition medium, worth a serious discussion, from many different perspectives some of which have been suggested above. For this I am also highly appreciative of people like artist Pompe Hedengren pushing the limits and showing different ways of telling stories. And hopefully sparking constructive discussions in the museum sector.

As John H. Falk says in his brand new book Leaning Into Value :

“…the value of anything, be it a good or a service, be it for personal use or public use, is always defined by the user/receiver, not by the creator/provider.” (2024)