ARTDESENT — Art, Design, Entertainment

AI Reads the Past: The Herculaneum Scroll and Art Preservation

The deciphering of the Herculaneum scroll with AI introduces new possibilities for cultural heritage preservation, raising questions about its place in the broader landscape of art.

By Ravi Iyer··2 min read
Paul Gauguin — Ia Orana Maria (Hail Mary)
Ia Orana Maria (Hail Mary), Paul Gauguin, 1891 · Paul Gauguin (Public Domain (CC0))

In October 2023, researchers partially deciphered a text from the Herculaneum scrolls, ancient manuscripts buried under volcanic ash since 79 CE. This breakthrough resulted from X-ray microtomography and machine learning. Led by Brent Seales, a professor at the University of Kentucky, the team employed custom neural networks to detect faint traces of carbon-based ink on the scrolls' fragile surfaces.

The scrolls, part of a library attributed to Lucius Calpurnius Piso Caesoninus, were discovered in the 18th century near Naples. Traditional examination methods have largely failed due to their fragility. Attempts to unroll or manipulate them caused further degradation. This new approach minimizes physical damage, allowing access to content once thought lost.

The AI-driven process began with non-invasive X-ray imaging to capture high-resolution scans. These scans fed into a machine learning model trained on synthetic data simulating expected ink patterns. Researchers identified characters and patterns invisible to the naked eye, marking the first step in translating the scroll's text.

“It's a profound moment for digital humanities,” Seales told Nature. “The neural network doesn't just read the ink; it helps us hypothesize about how the text might have been structured, giving scholars tools to reconstruct the document contextually.”

This breakthrough highlights AI's technical potential but raises critical questions. Whose narratives will AI prioritize in cultural heritage preservation? The Herculaneum library's dominance of Greek texts obscures other aspects of Roman life. AI systems trained on biased datasets risk amplifying these gaps.

Moreover, reliance on proprietary software raises concerns about access. The X-ray data and trained models are currently held by the University of Kentucky. Open access to these tools is essential for global research benefits. Esra Akcan, an architectural historian at Cornell University, emphasized at a 2022 symposium, “The ethics of access must evolve alongside our tools.”

The applications of this technology extend beyond text. AI is increasingly used in art preservation, reconstructing fragmented frescoes at Pompeii and digitally restoring faded medieval manuscripts. These advancements are transforming conservation practices from analogue to digital workflows.

However, as AI mediates our understanding of art and history, it raises fundamental concerns. Traditional restoration relies on human expertise, while AI operates on probabilistic models. The results, though authoritative in appearance, depend on algorithms and training data, significantly impacting historical accuracy.

The broader adoption of AI in art preservation intersects with funding and institutional priorities. AI-driven projects often attract resources that might otherwise support traditional methods. In a field with constrained budgets, this shift could sideline valuable but less glamorous preservation efforts.

The decoding of the Herculaneum scroll signifies a major advancement in using AI for cultural heritage. Technology can reveal the inaccessible, offering new insights into our shared history. As institutions adopt these tools, they must critically engage with the questions they raise about bias, access, and the evolving role of human expertise in interpreting history. The scroll may now speak, but the task of listening—and questioning—remains ours.

#ai#art preservation#cultural heritage#herculaneum scroll#digital humanities#technology#art history
Ravi IyerRavi Iyer writes on generative practice, video art and code-based work from Mumbai. Previously curated at the Khoj Studios.
Continue reading