This is exacerbated as they are incorporated into generative AI datasets. Beyond the use of outmoded data, generative AI visualizations of the past can be extremely poor.
Even when more plausible details are included, they can be seamlessly integrated with other highly inaccurate elements. For example, it is impossible for viewers to disentangle the data-led from the so-called hallucinations (mistakes) produced by AI.
Highlighting uncertainty is of central importance and concern among archaeologists. Archaeological illustrator Simon James noted that reconstruction artists have used strategically placed clouds of smoke to obscure unknown elements.
As a digital archaeologist, I have made virtual reconstructions of many different sites and subjects. I know there is often estimation and guesswork involved in making holistic representations.
Indeed, photorealistic accuracy is not always the paramount consideration in visualization—particularly when exploring different hypotheses or addressing young audiences. But knowing what is backed by archaeological data and what is more speculative is key for authentic visual communication.
THE DANGERS OF PSEUDOARCHAEOLOGY
This is particularly important at a time when pseudoarchaeology is increasingly prevalent in popular media, such as the Ancient Apocalypse show on Netflix. The celebrity host and author Graham Hancock asserts there was a lost ice age civilization of Atlantis that had advanced technology. But this claim has been thoroughly repudiated by archaeologists.
Arguably, hoaxes will be much easier to perpetuate using generative AI. Beyond the high potential for misinformation about archaeology, the use of generative AI for archaeological visualizations can actually be harmful for archaeological knowledge production.
My research has shown that crafting reconstructions and illustrations in archaeology is incredibly important for understanding and interpreting the past. Creating visualizations based on science—and indeed soundscapes, smellscapes, and other interpretations based on multiple senses—is very helpful for generating new questions.
