AI, scientists unite to decipher old scrolls charred by Vesuvius
LONDON — Scientists hope a mix of artificial intelligence and human expertise will help decipher ancient scrolls carbonized by a volcanic eruption 2,000 years ago.
Hundreds of papyrus scrolls were found in the 1750s in the remains of a lavish villa at the Roman town of Herculaneum, which along with neighboring Pompeii was destroyed when Mount Vesuvius erupted in A.D. 79.
The library of what’s called the Villa of the Papyri has the potential to add immeasurably to knowledge of ancient thought if the scrolls, which have been rolled up into the size of a candy bar, could be read.
The heat and volcanic ash from Vesuvius destroyed the town and preserved the scrolls, but in an unreadable state, turning them into charred fragile blocks that disintegrate if unrolled physically.
Scholars and scientists have worked for more than 250 years on ways to decipher the scrolls, the vast majority of which are held in the National Library of Naples.
In 2023, several tech executives sponsored the “Vesuvius Challenge” competition, offering cash rewards for efforts to decipher the scrolls with machine learning, computer vision and geometry.
On Wednesday, the challenge announced a “historic breakthrough,” saying researchers had managed to generate the first image of the inside of one of the three scrolls held at Oxford University’s Bodleian Library.
University of Kentucky computer scientist Brent Seales, co-founder of the Vesuvius Challenge, said the organizers were “thrilled with the successful imaging of this scroll.” He said it “contains more recoverable text than we have ever seen in a scanned Herculaneum scroll.”
The scroll was scanned by Diamond Light Source, a lab at Harwell, near Oxford, that uses a particle accelerator known as a synchrotron to create an intensely powerful X-ray.
Scientists then used AI to piece together the images, search for ink that reveals where there is writing, and enhance the clarity of the text. The process led to a 3D image of the scroll that allowed experts to unroll it virtually, using a process called segmentation.