Editor’s Note
From pioneering the jeweler’s loupe in the 1930s to now integrating artificial intelligence, the Gemological Institute of America continues its legacy of innovation. This article explores how GIA is leveraging AI to advance the science of diamond grading.
In the 1930s, the Gemological Institute of America (GIA) introduced the first high-quality jeweler’s loupe. This small, handheld device magnified gems 10 times, allowing jewelers and gemologists to examine a stone’s color and clarity. It revolutionized gemology and remains in widespread use today.
This spirit of exploration and innovation has always driven GIA, so it’s no surprise the organization saw artificial intelligence (AI) as the ideal way to optimize the diamond grading process. GIA Chief Operating Officer Pritesh Patel explains:
Assessing a diamond’s clarity is a complex process. Gemologists use a loupe, microscope, or images to meticulously examine each diamond for inclusions—tiny characteristics hidden within the stone’s structure. These can be minute internal spots or cracks that penetrate the surface.
Flawless diamonds, those without any inclusions, are extremely rare. Less than half of one percent of diamonds graded by GIA worldwide fall into this category. The vast majority have one or more inclusions, the combination of which makes each diamond unique.
With this in mind, Patel approached IBM Research with a vision: he believed that with the right skills and technology, the power of cloud-based AI could be harnessed to grade diamond clarity. IBM Research agreed, and the two organizations formed a strategic partnership: GIA provided expert images and data from tens of millions of diamonds examined by its diamond experts, while IBM provided AI capabilities and computing power. The result was GIA’s adoption of a cloud-based AI approach to diamond grading.
After IBM Research successfully developed a proof of concept (POC) demonstrating that AI could indeed help automate the grading process, the IBM Global Cloud Accelerator Team (GCAT) stepped in to move the project to the next stage. The GCAT team collaborated with the GIA Engineering DevOps team to guide the solution from POC to a production-ready environment with separate development, testing, production, and disaster recovery clusters.
Today, the solution is in testing and nearing full deployment. GIA laboratories upload professional images of each diamond to an IBM Cloudant® database on IBM Cloud®. The system’s middleware layer consists of IBM Cloud Kubernetes Service clusters.
GIA’s clusters consist of 3 NVIDIA K80 GPUs, each with 1 shared virtual node and 1 bare metal node in a serverless architecture. The NVIDIA GPUs, capable of rapidly processing high-resolution images, are well-suited to GIA’s needs, helping to speed up the entire process and reduce the time needed to validate AI algorithms.
The new solution analyzes each diamond using two custom algorithmic models. An automated plotting AI model visually maps a diamond’s inclusions, while a grading model assesses the diamond’s overall grade. This information is then sent via an iPad app to GIA’s gemologists, who can evaluate the automated plot and make modifications if necessary. These changes are fed back into the system to re-evaluate the grade and can be used to retrain the AI models for improved accuracy.
GIA’s mission to protect consumers drives its innovative thinking. But another goal for the project was achieving the highest level of data security to protect the integrity of diamond grading. Diamonds are high-value commodities, making it crucial to safeguard the grading process.
IBM Cloud App ID helps ensure the upload and processing of diamond images is secured with advanced security features, including multi-factor authentication, single sign-on, and user-defined password policies.
The images are stored using IBM Cloud Object Storage, a cost-effective storage solution that meets all of GIA’s requirements for scalability and accessibility. IBM Cloud Object Storage also provides built-in encryption and policy-enabled, lockable, Write Once Read Many (WORM) storage capabilities.
GIA grades millions of diamonds annually. Patel anticipates the new solution could handle 70–80% of those assessments, primarily focusing on smaller-sized diamonds submitted, thereby freeing human graders to concentrate on more complex cases where human evaluation is critical for determining the grade.
But reaching that point takes time. Currently, GIA is using the solution in 2 of its 11 labs, performing AI and human assessments in parallel as the team refines the algorithms. Ultimately, with ongoing support from the GCAT team, GIA intends to deploy the new solution across all its laboratories.
While the solution is still in its early stages, Patel already sees several key benefits. The first, he says, is efficiency.
The solution will also improve accuracy and repeatability. Although human diamond graders undergo rigorous, guided training, their work is still within the limits of human senses. When two graders look at the same diamond, their assessments can vary slightly. With AI, these subtle differences can be virtually eliminated, helping ensure diamonds are accurately valued when they enter the market.
GIA is staking its reputation on the integrity and accuracy of the new solution. Just as the introduction of the jeweler’s loupe in the 1930s transformed diamond grading, this project will take the precision of the grading process to an entirely new level.
