Pritesh Patel, senior vice president and chief operating officer at the Gemological Institute of America (GIA), busts some of the myths associated with automated diamond grading.
This article was first published in September 2022 in a special Rapaport supplement, sponsored by the GIA and titled Future Proof.
Although artificial intelligence (AI) has been around for decades, many of the ways in which it works remain murky in the eyes of much of the world. And while that is not a problem for technology companies or labs such as the Gemological Institute of America (GIA), it can be for many consumers.
The industry continues to make strides in using AI for a number of diamond-related advancements, including mining, cutting, polishing and grading. But when it comes to consumer-facing segments of the industry, such as diamond grading, several myths prevail that could ultimately harm the trust bond that’s so important to the trade. And like most myths, what people believe to be true and what the truth is can often be two very different things.
Why do these misconceptions happen? Pritesh Patel, senior vice-president and chief operating officer at the GIA, believes much of it comes from the hype surrounding AI and the fact that it is a bit of the “flavor of the month.” Because AI is often presented as something beyond its current capabilities in many popular movies and books, this can drive confusion. And, in general, fear of the unknown can also play a factor.
The human element
One of the most prevalent myths surrounding diamond grading is that AI can – and eventually will – replace skilled experts, says Patel. The truth is that, while AI can support diamond grading by automating evaluation of the 4Cs, there will always be a need for human graders with experience to detect things such as treatments and other complex issues that require grading skills that technology just cannot duplicate.
“There’s an important role technology plays along with the human,” Patel explains. “In certain [areas], the human element is still very much required. Technology is not at the stage yet where you can completely rely on it 100%.”
Treatment deduction is a core part of the grading process, and key to consumer protection. Today, there are more sophisticated treatments being applied to diamonds than ever before, Patel notes. The speed at which they evolve makes them too difficult to automate. “What we are dealing with is a human element that only a human can verify, and that’s a core part of grading beyond AI,” he adds.
Colored diamonds also present a problem for AI to grade, as they are more complex and nuanced, while differentiating lab-grown diamonds from natural requires specialized analytical instruments. “AI is not yet to the point where it can handle the complex interaction of sophisticated instruments and visual observations and analysis by trained experts to determine the growth method and detect treatments,” Patel asserts.
Most importantly, and contrary to popular belief, not only will technology not usurp human jobs, but throughout history there is evidence that technology increases jobs, according to Patel. He points out that GIA expects AI to take on tasks that require repeatability and speed, which will free expert graders to move to newer and more complex challenges.
Man vs. machine
The debate over which is better — human or AI grading — is another highly contentious subject. For some, the myth is that humans are less fallible than machines, while for others the opposite rings true. The reality lies somewhere in the middle, notes Patel.
“For most 4Cs grading – cut, color, clarity and carat weight – AI is as good as human graders, and perhaps more consistent,” he says. “However, more complex grading tasks, such as higher clarities, treatments and challenging stones, still require trained, experienced staff.”
At this time, AI is very close to being expert at grading the 4Cs but is not yet ready to handle the observation and analysis needed for more complex tasks, and it may take some time, he comments.
Part of the fear, in terms of AI, is the worry that technology can be more easily manipulated or corrupted in some way. For this reason, Patel sees the necessity for human and technology to work together. “Any grading system needs checks and balances, and rigorous quality control to ensure that its standards are consistently and accurately applied, whether by graders or AI,” he says.
The idea that somehow AI and primarily automated grading will solve the perceived issue of inconsistency in grading between labs is another myth Patel has heard numerous times. While AI can advance consistency and repeatability within each laboratory, it is less likely to bring about the same stability between different labs. The reason inconsistencies occur, he explains, is because different labs use varying standards and application of those standards when they train AI.
“Our technology is very, very sharp,” says Patel. “And the key part here is we’re giving it the standards and millions of data points to let it learn over a period of time. So, technology is an advancement of learning at a very, very high level and pace that is harder for humans. It takes a lot of data to train AI. GIA has a unique and unrivaled dataset of information from millions of diamonds. As the technology develops, the time to train AI will decrease, but the need for high-quality, consistent, independent and trusted data will remain.”
Patel points out that, although it takes a long time to train AI, the result, when it comes, provides a more consistent outcome on a more regular basis. “It’s so important to get that consistency in grading and, you know, there is human subjectivity,” he explains. “GIA’s role in
developing these technologies is to use our decades of understanding – based on the millions and millions of diamonds that have flowed through our laboratories – and impart that knowledge to the AI so it can work with us in a more consistent manner.”
Trust the process
Ultimately, the main goal of any lab is consumer trust. Without that, there is no business. But while consumers want to trust the fashion in which a given lab is grading their diamonds, they don’t want to sacrifice efficiency. Patel believes that using AI in grading can provide both. “The key question in grading is always consumer trust. GIA has been working for many, many years to ensure we bring the technology and the efficiency but protect consumers’ trust at the same time.”
In the end, Patel hopes the technology will speak for itself in ensuring trust, but the GIA is ready to work with consumers and the trade to show how AI grading works and get the message across that the standards and systems it adheres to are the same as they have always been.
“It takes a long time to earn consumer trust, and you need to work every day to maintain it,” he says. “We are working cautiously and deliberately to leverage AI to enhance, improve and extend GIA’s diamond grading. The more we learn – and we are learning a lot very quickly – the more we will know how to use AI to apply GIA’s standards to extend our consumer protection mission.”
Main image: The GIA Match iD inscription-matching machine.