AI Photo: VCG

A Chinese research team has launched an artificial intelligence (AI) model designed specifically for mineral exploration in the southern parts of Northwest China's Xinjiang Uygur Autonomous Region, marking what developers describe as the region's first large language model dedicated to the mining sector.

The "Congling Zhixun" AI mineral exploration model was unveiled at the 16th Kashgar Central and South Asia Commodity Fair, held recently in Kashi, Xinjiang region, the Science and Technology Daily reported Monday. 

The model is designed for the metallogenic belt spanning southern Xinjiang and Central Asia, a geological zone where conditions favor the formation of mineral deposits.

Unlike general-purpose AI models, the model is trained for a specific industry and uses specialized datasets and knowledge. In this case, the developers have trained the model on geological data from southern Xinjiang to help identify areas that may contain mineral deposits.

Ding Haifeng, a researcher and head of the "Congling Zhixun" project, said the model needed to be developed locally because geological conditions and the processes through which minerals form vary greatly from one region to another, the Science and Technology Daily reported. 

"The unique geological structures of the metallogenic belt spanning southern Xinjiang and Central Asia, as well as the rich iron ore deposits in the Taxkorgan Tajik autonomous county, mean that the model must be developed on the ground in southern Xinjiang to ensure the accuracy of its predictions," Ding was quoted as saying.

To train the model, Ding's team worked with the Kashgar Geological Brigade and used high-grade iron ore deposits in the Taxkorgan Tajik autonomous county as an initial sample. They built a specialized geological dataset, addressing what the team described as the major challenges in the sector: difficulties in sharing geological data and a shortage of high-quality data for AI training.

The resulting system is intended to support several stages of mineral exploration and geological work, including identifying potential exploration targets, geological administration and science education.

The developers say the model goes beyond conventional AI-assisted prospecting by incorporating what they call a "world model" of geological mineralization. In simple terms, this means the system is designed not only to identify where minerals may be found, but also to analyze the geological processes that could have produced a deposit.

"Congling Zhixun" cannot only accurately identify mineralization targets, but also infer the entire process, from the sources of mineral-forming materials and their migration pathways to mineral accumulation and subsequent geological alteration, Ding said. 

Ding added that this approach follows the reasoning of geological experts, making the model's predictions "explainable, traceable and verifiable."

The project also relies on substantial computing power. Kashi is home to Xinjiang's first 10-million-kilowatt-scale photovoltaic renewable energy base and has built southern Xinjiang's first computing center with a capacity of 4,000 petaflops, according to the developers. The combination of renewable power, computing infrastructure and AI is intended to provide relatively low-cost, low-carbon computing resources for training and updating the model.