Algorithms trained on Google Maps satellite imagery no longer merely sort data — they are beginning to notice what has eluded archaeologists' eyes for decades. In Sardinia, such a system has identified six probable new Bronze Age nuraghes, confirmed by preliminary field verification.
The AIOH model, developed by specialists at the CNR Institute of Heritage Science, was trained on 1612 images of known nuraghes. These conical stone towers, unique to the island, were built from roughly 1800 BCE. They served as dwellings, fortresses, and symbols of power. The neural network, based on Vision Transformers, learned to recognize their characteristic shapes and shadows even in complex terrain and under varying lighting.
In the northwestern part of Sardinia, near Porto Torres, Olmedo, and Stintino, the model flagged areas with a high probability of monuments. Field checks confirmed six previously unrecorded sites and eight false positives. This is not a "rediscovered"-style sensation, but a systematic tool that helps fill in the blank spots on the map of already known heritage.
Such technologies are changing the very logic of archaeological research. Previously, the search relied on chance finds, oral traditions, or labor-intensive ground surveys. Now algorithms scan vast territories in hours, identifying priority zones for expeditions. At the same time, the model's accuracy on test data approached 99,9 %, although real-world conditions always introduce corrections.
For Sardinia, where more than seven thousand known nuraghes are counted, such finds matter not only for science. They help us better understand how ancient inhabitants organized space, defended themselves, and interacted with the landscape. Each new site is another fragment of the mosaic showing how complex the society was that left behind these silent towers.
The main lesson here is not in the number of new points on the map, but in how modern tools allow us to look more closely at what is already familiar. Satellite imagery, available to everyone, combined with trained models, becomes a kind of magnifying glass for the past.

