December 13, 2022

The digital "super-expert" putting gemstones under the microscope

Determining if a gemstone is authentic and what it is worth requires years of experience and extensive specialist knowledge. CSEM has partnered with Gübelin’s Gem Lab in Lucerne, CH, to finalize work on a high-performance AI authentication solution. In the future, this digital "super-expert" will work together with human experts to analyze gemstones.

Gems in a black box

For the average joe, it is easy to be overwhelmed by precious stones’ dazzling colors and sparkling appearances. But this is not the case for experts based at Gübelin’s Gem Lab in Lucerne, who are connoisseurs in distinguishing and understanding these precious minerals. Indeed, Gübelin’s gem lab is regarded as the international gold standard for checking the identity, authenticity, and origin of high-value, colored gemstones like emeralds, sapphires, and rubies.

 

Gemstones are sent from all over the world to Gübelin's different locations – Lucerne, Hong Kong, and New York – for authenticity certification. "An essential aspect of our work is in guaranteeing consistency in our results," explains Dr. Daniel Nyfeler, Managing Director of the gemological laboratory. “This means that every time Gübelin’s experts put a particular gemstone under the microscope, they must make the same assessment. Deviations in the procedure should only arise if the stone has been altered in the intervening period,” he stresses.

A question of expertise

Gübelin is known for the high-quality standards it applies to its analyses. And while cutting-edge, highly complex analysis methods are utilized in its Lucerne lab, a large portion of the data interpretation relies on human evaluation by experienced experts. To ensure a high level of consistency in the results, Gübelin joined forces with the Swiss technology innovation center, CSEM, back in 2020 to develop a high-performance machine learning software (to the corresponding press release). The system helps relieve the experts from the arduous task of interpreting large volumes of data. A team led by Philipp Schmid, Head of the Industry 4.0 and Machine Learning Section at CSEM, helped guide the complex "Gemtelligence" project towards fruition. “Today, we have been able to bring the project to a successful conclusion, and together with Gübelin, we have been able to launch the first digital 'super-expert' for gemstones,” says Phillip Schmid. “This takes care of the extensive interpretation of analytical data, supporting the human experts in their labor-intensive work,” he adds.

So, how was this super-expert developed? “The foundation of any artificial intelligence is annotated data,” explains Schmid. The project team, made up of employees from CSEM and Gübelin, were able to tap into a veritable treasure trove of data for this project: The data register contained the comprehensive analyses of tens of thousands of customer gemstones, amassed by Gübelin's gemological laboratory since the 1970s. This data was enriched with findings from Gübelin's unique collection of reference stones, which comprises more than 28,000 gemstones from every commercial mine worldwide. “This means our application can combine optical spectral analyses with chemical properties and undertake large artificial intelligence extractions,” explains Schmid. As a result, a gemstone’s origin can be precisely determined, and the system can also detect if a gemstone has undergone any heat treatment. Both these factors have a strong bearing on the value of a stone.

"Gemtelligence," has received government funding from Innosuisse, with the technology representing an innovation with huge potential for application in multiple domains. Similar approaches could advance medical diagnostic techniques and make industry products safer thanks to 100% quality control. The project even offered up some exciting discoveries for the researchers from CSEM: “It clearly demonstrates that it's not enough to simply combine unfiltered datasets with AI,” notes Philipp Schmid. Only when the data is enhanced with important additional information, such as the experience of human experts, can the "ground truth" be digitally mapped and utilized. “Everything else is, and sadly always will be, essentially junk data,” Schmid concludes.