Artificial intelligence is interested in ever more varied fields. In addition to sectors directly linked to high-tech, such as facial recognition, others are also concerned. DeepMind demonstrates this by putting a new AI model at the service of historians and archaeologists. This one reports The Vergehas proven its potential by suggesting particularly complete interpretations.
The potential of DeepMind’s machine learning applied to history
DeepMind has been working on this challenge for a few years now. In 2019, the company unveiled Pythia, an AI now used in debates between archaeologists. Now it’s the turn from Ithaca to take his place.
Harnessing facial recognition to diagnose rare diseases
Ithaca should provide more comprehensive results than Pythia when faced with language interpretation challenges. In 2019, DeepMind stated: “ One of the problems with discerning meaning from incomplete text fragments is that there are often multiple possible solutions. In word games, players guess letters to complete a word or phrase. The more letters specified, the more limited the possible solutions become. But unlike those games, where players have to guess a sentence in isolation, historians restoring a text can estimate the likelihood of different possible solutions based on other context clues in the inscription, such as grammar and linguistic considerations. , layout or historical context. »
With Ithaca, DeepMind has just demonstrated an artificial intelligence capable of deducing, from these elements, a missing text, but also the geographical origin of a fragment or its dating to within thirty years. The interest for archaeologists is obvious. As reminded Ars-Technica, many ancient sources are so damaged that a large part of their inscriptions are completely unreadable. The question of their geographical origin is also a challenge, each fragment having been moved over the centuries. As for the question of dating, current methods (when possible) take the risk of further damaging the remains.
A “complementary tool” and nothing more… for now
Ithaca uses the principle of machine learning. Before being put to the test, DeepMind’s artificial intelligence was exercised on 78,608 ancient Greek inscriptions, each tagged with metadata describing where and when they were written, as far as historians know. From there, the software looked for patterns in this information and used them to build algorithms.
Once launched, the scientists who created it claim thatIthaca is 62% accurate in restoring letters in damaged texts. She would be able to assign an inscription to any of the 84 regions of the ancient Greek world with an accuracy of 71%. Promising statistics, but which remain perfectible. Ithaca cannot do without human expertise, and it should be noted that its algorithms are based on human knowledge which may, itself, turn out to be wrong overnight.
For Thea Sommerschield, historian and machine learning expert, Ithaca should be seen more as “ a complementary tool to help historians and not as a fully automated historian. Ithaca’s source code can also be improved: its code is available in open-source.