How AI is connecting painted walls and echoes in Etruscan tombs

Tarquinia’s painted tombs preserve something a museum photograph cannot: the room around the image. UNESCO records some 200 painted tombs among the necropolis’s roughly 6,000 rock-cut graves. Their walls belong to architecture that people entered, crossed and occupied.[4]

Maurizio Forte and Jacqueline K. Ortoleva bring that spatial question to computation in a study published online in August. They combine acoustics, visual saliency and simulated movement in two fifth-century BCE tombs: the Tomba del Gallo and Tomba dei Demoni Azzurri.[1]

Start with a sound that can be measured

The acoustic foundation predates the AI study. Ortoleva’s 2023 investigation of the Blue Demons tomb used a recorded handclap as a repeatable impulse source. Measuring how that impulse decayed revealed a room whose response varied substantially across frequency.[2]

These numbers describe the measured acoustic response. They help explain why the same room can treat a low sound and a high one differently. A picture of musicians supplies iconography; an impulse response supplies a physical property that can be inspected independently. Bringing them together gives the interpretation a second kind of evidence.[2]

How a wall becomes a map of attention

Visual saliency is an older strand of AI. In their 1998 model, Laurent Itti, Christof Koch and Ernst Niebur combined image features across scales into a map of conspicuous locations. The idea is to select places worth further visual processing before trying to understand the entire scene.[3]

That distinction is useful in archaeology. A strong local contrast can stand out without a model knowing whether the painted figure is a deity, a musician or a relative. Meaning and visual prominence can then be investigated separately, rather than silently assuming they are the same thing.[3]

The tomb study uses transparent feature scores, spatial rules and simulated agents responding to visual, acoustic and architectural gradients. It reports stable attention hotspots near the Gallo threshold and the Blue Demon panel under parameter changes. These are hypotheses about sensory organisation, not measurements of ancient gaze or trance.[1]

A productive reading is to treat the model as an explicit question. If attention is pulled towards a particular wall, does the result survive a change in the lighting assumption? If a narrow passage changes the simulated path, is the effect driven by geometry or by the weighting chosen for it? Those questions turn an evocative interpretation into something another researcher can challenge.

The study also discloses that ChatGPT assisted with analysis code which the authors reviewed and ran. Its archaeological argument rests on the measurements and modelling choices.[1]

The interesting object is therefore the whole room: an image, a route and a sound field sharing the same space. AI gives researchers a way to vary their assumptions about that arrangement. The surviving tomb supplies the constraints.

Report an error or suggest a correction