Is Artificial Intelligence Misrepresenting Roman Britain?

In-Depth Archaeological Analysis

Lisa Maurice's study, published in The Journal of Classics Teaching, presents a critical experiment designed to examine the impact of generative artificial intelligence (GenAI) images on the reception of the ancient world. The researcher input the phrase 'Roman Britain', referring to the Roman province of Britain (covering modern-day England and Wales between AD 43 and 410), into eight different free AI image generators. All resulting images contained details that contradict historical and archaeological evidence. For instance, some images depict Roman soldiers wearing armor that is either late Byzantine in style or entirely fantastical, while civilian attire appears heavily inspired by Hollywood cinema. These inaccuracies stem from AI systems learning representations of the ancient world solely from existing digital visual datasets, such as 19th-century paintings, modern reconstructions, or video game screenshots. Indeed, no photographs of Roman Britain exist; we only possess archaeological remains, architectural plans, mosaics, sculptures, and written sources. Consequently, GenAI systems attempt to generate 'realistic' images based on incomplete and biased datasets. The researcher emphasizes that while this technology could serve as an effective tool in classrooms, it is only beneficial when paired with critical thinking skills. Instead of students accepting AI-generated images as factual, they should be encouraged to ask, 'What is incorrect in this image?'

Key Findings

• Eight different GenAI tools produced historically inconsistent images when prompted with 'Roman Britain'.
• The accuracy of AI images is limited by the lack and bias of existing visual training data for ancient periods.
• GenAI images can be beneficial, particularly when used as a tool for critical thinking in education.

Historical Context and Connections to the Past

Roman Britain was a strategic province located on the northwestern frontier of the Empire, encompassing major urban centers such as Londinium (modern London), Eboracum (York), and Verulamium (St Albans). This region represented a unique blend of Roman architecture, local Celtic traditions, and military settlements. However, the modern visual perception of this era has largely been shaped by the works of 18th and 19th-century Romantic painters, 20th-century cinematic productions, and popular culture representations. This reality leads AI systems to learn from this distorted visual heritage. Recent studies published by Cambridge University Press also demonstrate that AI tends to reinforce existing popular narratives in archaeological illustrations rather than maintaining objectivity. Therefore, the images generated by GenAI reflect not historical reality, but the Roman myth embedded within collective memory.

Expert Perspective

This study serves as a critical warning for understanding the impact of the digital age on archaeology and classical studies. AI-generated images are increasingly being used in textbooks, documentaries, and even museum displays. However, presenting these images as 'authentic' may lead to the establishment of inaccurate historical perceptions, particularly among younger generations. As archaeologists, rather than rejecting this technology, we must consider how to adapt it as a pedagogical tool. For example, having students design their own prompts to test clothing, architecture, or daily life elements from different periods could enhance historical empathy and analytical thinking skills. However, for this process to be effective, it is essential to develop open-access, diverse visual datasets grounded in archaeological evidence.