A thousand-year cultural conversation about value, authenticity and provenance is being reshaped by artificial intelligence running not in a laboratory supercomputer, but on a consumer gaming graphics card.
Researchers at University Putra Malaysia, working with UNSW Sydney, have developed a deep learning model that can analyse Chinese ceramics and predict their value at auction with extraordinary precision. Trained on real-world sales data from houses such as Sotheby’s and Christie’s, the system uses image recognition to classify artefacts and estimate price categories, all powered by an NVIDIA GeForce RTX 3090, a GPU more commonly found in gaming rigs than academic institutions.
The project challenges the idea that expert appraisal of cultural heritage must rely solely on subjective human judgement. By training on kiln-specific styles, vessel morphology and decorative motifs, the model achieved classification accuracy of up to 99 per cent. It was even able to estimate market value within close range of final hammer prices, placing one Ming Dynasty item just 30 per cent below its eventual sale.
An algorithmic eye for craftsmanship
The system was trained on 20 historically significant styles across seven major Chinese dynastic periods, from the Tang Dynasty (618–907 AD) to the modern era. These included distinct categories based on kiln origin and visual characteristics. Analysis extended to vessel types, such as bowls, cups, jars, and bottles, classified through modular morphological features like necks, spouts, and bases.
In parallel, the AI identified decorative themes by detecting six major motif categories: plants, animals, human figures, landscapes, crackled glaze effects and geometric designs. A YOLOv11-based object detection model was employed to annotate the most visually dominant features of each artefact, laying the groundwork for both historical classification and market prediction.
While AI has been used increasingly in museum digitisation and restoration, this model moves closer to a form of cultural decision-making, using algorithmic training to simulate expert judgement. For smaller institutions, collectors or digitisation initiatives lacking access to seasoned curators or appraisers, such tools offer a new way to scale knowledge without compromising precision.
From gaming hardware to heritage insights
Perhaps most striking is the system’s reliance on widely available hardware. The use of a GeForce RTX 3090, a standard GPU designed for consumer gaming, demonstrates how far accessible AI infrastructure has advanced. No data centre, cloud supercomputer or specialist chip was required to produce the results.
This democratisation of cultural AI challenges assumptions about the scale and cost needed to deploy advanced analysis in the arts. The research team argues that this approach could extend to other forms of visual heritage. Initial experiments are already underway on Cantonese opera costumes and historical murals, where pattern recognition and visual classification are similarly valuable.
As AI tools gain ground in fields once thought to require human intuition, their presence in cultural appraisal raises deeper questions about authorship, authority, and access. The value of artefacts, once determined behind closed doors by a small group of experts, may increasingly be shaped by models trained on data, not opinion.
In this case, a graphics card built to render digital battlefields is now parsing centuries-old craftsmanship, bringing new perspectives to one of humanity’s oldest and most global debates.
