Parallel Minds and Papyrus: How GPUs Help Scholars Read the Past
When poets etched their lines on brittle clay they could not have imagined a future in which video-game chips would help us hear their voices again. Yet the most powerful allies of today’s classicists sit not in dusty libraries but in racks of humming graphics processors. The story of how those processors entered the seminar room shows that progress often arrives sideways—and that the humanities thrive when we notice the detour.
Graphics processing units began life as the arithmetic engines that turn polygons into dragons. They conquered artificial intelligence once researchers saw that a neural network learns by performing the same multiplication thousands of times. A single processor grew impatient with that drudgery, but thousands of tiny GPU cores were happy to oblige. Scale, not cunning, moved the field.
Scholars soon borrowed that scale. In 2022 a team from DeepMind and the University of Oxford trained a model called Ithaca on fragmentary Greek inscriptions. The network suggested missing letters, proposed dates, and even guessed a stone’s birthplace. The recovery of fourteen characters on a weather-beaten decree from Delphi nudged Athenian legal history a century forward. Machines did not replace epigraphers; they gave them better questions.1
The same arithmetic drives the Vesuvius Challenge, where scrolls carbonized at Herculaneum are reconstructed slice by slice. Tomographic images feed a network that teases ink from charcoal, letting papyrologists trace the hand of an unknown Epicurean.2 Similar projects read Syriac palimpsests, cuneiform tablets, and medieval marginalia. Everywhere the texts crumble; everywhere the GPUs pick up the crumbs.
If your department hopes to join the effort, look first at your workload. A laptop will classify a few shards, but a rack of GPUs will handle libraries. Shared clusters now run from Ann Arbor to Abu Dhabi. Collaboration, as ever, beats isolation.
The moral is plain. Follow the mathematics, respect the craft, and never dismiss the line cooks of computation. They are busy stirring the past back to life.
1 DeepMind Blog, “Ithaca: Restoring and attributing ancient Greek inscriptions,” 2022; Nature 603 (2022): 202-206; The Guardian, “AI helps historians piece together ancient Greek texts,” 9 March 2022.
2 Nature 620 (2023): 12-15; Vesuvius Challenge Official Site, “The race to read the Herculaneum scrolls,” 2024; The New York Times, “A scroll erased by Vesuvius reveals its first word through AI,” 12 October 2023.