A bit like a volcano, AI walks through the door
A dog walked into a tavern and said, “I can’t see a thing. I’ll open this one.” That’s it — that’s the joke. To get the humour, one would need cultural context from one of the oldest civilisations, Sumer (c. 3300-1900 BCE). Missing context is one of the ways in which an ancient text, though present, may still be “lost”. In other cases, the text’s physical condition may prevent it from being read. For instance, the eruption of Mt Vesuvius in Italy in 79 CE led to a whole library of scrolls being carbonised and preserved, but left too fragile to open. Until AI walked in and said, “I’ll open this one.” A scroll from the Herculaneum library has now been “unwrapped”, though not literally — it was scanned using high-resolution X-rays, virtually reconstructed and flattened so that a machine-learning tool could bring out the traces of ink, allowing scholars to read the text. It turned out to be a treatise on ethics, perhaps authored by the Stoic philosopher Chrysippus. Hundreds of scrolls remain to be opened, and both the data and the code are open to all. Classicists are chomping at the bit, talking about all the ancient texts they dream of rediscovering. Some have brought up the lost dialogues of Aristotle, called a “river of gold” by Cicero. The nerdiest may yearn for the dictionary of the Etruscan language written by the nerdiest of emperors, Claudius. This speaks to the duality of AI in academia: While it has arrived as a menace in the classroom, the possibilities it opens up for both scientists and humanities researchers are mouth-watering. It’s destruction and preservation wrapped up in one — a bit like a volcanic eruption.
- 1While this editorial is cultural rather than constitutional, it intersects with governance debates over regulating artificial intelligence, an area where India has so far preferred a light-touch approach under the IT Act, 2000 rather than a dedicated statute. The government's 2023 stance favoured not regulating AI growth prematurely, even as the proposed Digital India Act is expected to address AI harms. Policymakers must balance innovation, illustrated here by scholarly breakthroughs, against risks like deepfakes and academic dishonesty.
- 2The breakthrough is a global scientific collaboration: the scroll was decoded through the Vesuvius Challenge, an open competition launched in 2023 that used X-ray tomography from a particle accelerator and machine learning to read the Herculaneum papyri. Such open-science cooperation, with data and code freely shared across borders, mirrors international frameworks like UNESCO's 2021 Recommendation on the Ethics of Artificial Intelligence. It shows how cross-national research infrastructure can unlock heritage inaccessible to any single country.
- 3Legally, AI-assisted recovery of ancient texts raises questions of copyright and cultural-heritage law: works this old are firmly in the public domain, but the software, scans and reconstructions may attract fresh intellectual-property claims. India's Copyright Act, 1957 protects original expression, while the machine-learning models involved sit in a contested space that courts worldwide are only beginning to address. The editorial's note that the data and code are 'open to all' reflects a deliberate choice to keep such heritage a commons.
- 4The scale is striking: the Herculaneum library holds an estimated 1,800 carbonised scrolls, of which only this handful has been virtually read, and the buried text dates to the Vesuvius eruption of 79 CE, nearly two thousand years ago. The Vesuvius Challenge offered prize money exceeding one million US dollars to spur decoding. With machine learning now proven, hundreds of scrolls remain to be opened, potentially expanding surviving classical literature by a significant fraction.