Discover how AI and machine learning revolutionize diamond grading. Explore automated color grading, clarity analysis, and cutting-edge technology from GIA, Sarine, and DeBeers labs.
11 Minute Read
Diamond grading stands at a technological crossroads where artificial intelligence meets centuries-old gemological expertise. Since GIA introduced the first official grading system in 1953, the industry has transformed from subjective visual assessments to sophisticated automated analysis. Today's laboratories employ spectrophotometers, 3D mapping systems, and machine learning algorithms that can analyze thousands of diamonds with remarkable precision.
This article explores:
Before 1953, diamond grading lacked global consistency. No officially recognized system existed, and jewellers applied different standards across organizations. This inconsistency created confusion in the trade and complicated transactions between dealers. GIA's introduction of a standardized grading system marked a pivotal moment, establishing clear parameters that laboratories could replicate worldwide and giving consumers confidence in quality assessments.
Gemologists have long envisioned a single "black box" capable of performing every analysis task, from stone identification to synthetic detection, origin determination, and treatment indication. Complete color and clarity grading would round out its capabilities. Unfortunately, this unified device has not materialized. Instead, the industry has developed numerous specialized devices focusing on different analytical areas.
Modern technology enables unprecedented diamond analysis:
The substantial investment in diamond research has pushed this field ahead of all others in automation and AI adoption.
Since the 1950s, GIA employed colorimeters to assign color grades. These early tools operated within limited ranges, typically working only on diamonds with yellow tints. Laboratories never relied on these devices for final grades; human graders always maintained priority. As technology improved, this fundamental reliance on human vision persisted, though automated systems gradually gained more influence.
Master stones represented one of the first steps toward standardized grading. Every laboratory maintains a set of diamond masters spanning color grades from D through L or beyond. Graders carefully select these diamonds, usually at 0.50 carat or one full carat, within 0.10 carat of each other.
Ideal master stones meet specific criteria:
These master stones technically form the basis for all grading developments, extending right up to modern AI-enhanced systems.
Clarity grading occupies a gray area in standardization. Producing useful master sets for each clarity grade proves impossible due to the extensive range of inclusion combinations falling into each grade. Instead, the industry has always relied on training by experienced graders and comprehensive textbooks. Publications like "Diamond Grading ABC" by Pagel-Theisen and "Photo-Masters For Diamond Grading" by Roskin convey possibilities through multiple images, providing the visual education that master stones cannot deliver for clarity assessment.
Proportion assessment was the first area benefiting from technological assistance. Anyone who has examined a diamond understands the difficulty in accurately assessing angles or sizes. For example, crown and pavilion angles for a well-cut round brilliant should measure 30-35 degrees and 41-42 degrees respectively. Accomplishing this by eye proves extremely difficult.
The proportionscope, introduced in 1967, made the task considerably easier. GIA and Gem Instruments Corporation developed this back-projection tool for determining angles and sizes. Graders simply positioned the diamond in the holder, aligned the stone's silhouette with guide lines, and read off measurements.
In 1992, proportion grading became fully automated through the Sarine DiaMension, followed by the OGI Systems Scanox S. These machines analyze every facet, angle, and proportion of a diamond in seconds, employing lasers and reflections to measure to tenths of a degree and hundredths of a millimeter. Though not true AI, these systems rely on algorithms to determine deviations from specified tolerances and assign overall grades.
Color grading represents a prime area for AI and machine learning utilization. Graders initially performed this assessment by eye, using a white grading tray as background and comparing the diamond to this shade or to master stones.
This traditional approach suffered from several issues:
Automated color grading offers consistency, repeatability, and profitability advantages. Computers never take vacations, can operate continuously 24/7, and never complain.
Initial automation attempts employed colorimeters, either spectrophotometers or tristimulus machines. Spectrophotometers measure absorption at selected wavelengths or across the entire spectrum. Research and development has focused mainly on spectrophotometer techniques due to their relative simplicity.
Early models included the Eickhorst Diamond-Photometer, which analyzed absorption of yellow and blue light as it totally internally reflected through the diamond. The system converted the ratio between these absorption levels into a color grade. The drawback was that analysis limited to blue and yellow light meant only Cape series and Canary yellow diamonds could receive grades.
Robert Shipley of GIA developed the Shipley Electronic colorimeter in the 1940s, working along similar lines. Both models have now been superseded. Zeiss produced their spectrophotometer version that analyzed light transmission across the entire visible spectrum and into the UV range down to approximately 320 nanometers. The Zeiss system used a monochromator, which recorded each wavelength individually and plotted them as a transmission curve. This comprehensive spectral analysis represented a major advancement.
AI already operates extensively in light performance analysis, assessing the return of light from diamonds. In early days, graders performed this manually with tools such as "Hearts and Arrows" viewers, but only to a visual level. Technology has advanced significantly beyond these simple assessments.
AI systems now calculate grades based on equipment-supplied information. The Sarine Diamension, in use since 1992, provides detailed angles and dimensions. Systems then process this data through ray-trace style algorithms to calculate light return levels and assign grades. Systems such as D-Imaging from Dynagem employ this technology to provide cut, clarity, and color grades from a single unit, with digital images reviewable later to confirm clarity grading decisions.
Machine learning predominates in modern grading practices. Unlike ChatGPT and other AI models that scour the internet, diamond grading machine learning relies on iterative processes, with systems learning through repetition. It operates the same way average human graders do: a database of results compiles, providing experience that the system calls upon when grading the next diamond.
Machine learning offers a crucial advantage over human training:
Automated systems can process thousands or millions of iterations, improving their recognition and analysis ability continuously.
DeBeers operates proprietary machines for color and clarity that work with human graders to ensure reliability.
The Falcon System for Color:
The Falcon currently serves as an assistive tool but could evolve to remove human intervention entirely.
The Eagle System for Clarity:
As with the Falcon, two trained graders visually assess each diamond before all results combine to produce the final grade.
From their website and research materials, Sarine appears at the forefront of AI development in the diamond industry. They cover all fields from mapping rough diamonds and suggesting cutting options through to final grading. Sarine developed their automated color and clarity system around 2016, with GIA working with IBM following a year later in 2017. These organizations possessed the extensive datasets required to initiate machine learning research programs.
Color Capabilities:
Clarity-II Features:
Sarine acknowledges the difficulties in clarity grading, as it requires a very large sample set to cover the range of possibilities given each diamond's uniqueness.
Sarine also introduced their "Light" system, which analyzes how diamonds interact with light through reflection, refraction, and dispersion. They claim to operate the only fully automated symmetry and cut grading scanner in the form of their DiaMension Axiom. One possible issue has emerged: one Sarine representative center stated they "no longer produce a color machine," raising questions about their current focus areas.
GIA's version of AI clarity grading involves imaging the diamond, then sending digital images to the cloud. The system compares them to an expansive database of diamond grading information, and the AI computes a clarity grade.
GIA states this approach will:
The implication suggests the system will rapidly reduce or eliminate human involvement in routine grading decisions.
Gubelin Gem Lab employs AI as part of its Gemtelligence system, predominantly on colored stones rather than diamonds. This system examines various analysis results for patterns and markers that may elude human laboratory workers. In testing, Gubelin claims their Gemtelligence system has surpassed human gemologists in detecting origin or heat treatment in stones.
As with any mathematical computation, there is always a confidence interval, which can be thought of as an error margin. Gubelin indicates that any Gemtelligence result with less than 98 percent confidence would automatically undergo additional tests by at least two senior gemologists. The main drawback is that the system cannot yet detect new origins or treatment techniques, as machine learning is always based on past experience.
Although larger laboratories maintain secrecy about their AI development work, smaller laboratories sometimes speak more openly. Conversations with several smaller laboratories offering diamond grading services revealed a common concern.
All smaller labs expressed worry that:
These concerns reflect the industry's changing economics as technology costs drop and efficiency improves for larger organizations while smaller operations struggle to keep pace.
Whether gemologists like it or not, AI is here to stay and will gain further footholds into gem analysis worldwide. Diamond grading was an obvious starting point given the quantity and value of diamonds graded worldwide each year. Once laboratories perfect the technology, it will readily lend itself to analyzing any area of gemology.
However, gemologists retain important advantages:
The key lies in adaptation. Gemologists who embrace AI as a tool rather than viewing it as a threat will find themselves better positioned in the evolving industry. The combination of human insight, intuition, and experience with AI's speed, consistency, and analytical power represents the most powerful approach to gemological analysis.
Traditional skills remain valuable, particularly in areas where AI has not yet been trained or where new phenomena emerge. The gemologist's ability to recognize when something does not fit established patterns, to question results, and to investigate anomalies ensures their continued relevance. AI excels at what it knows; humans excel at discovering what remains unknown.
As technology advances, the role of gemologists will likely shift from performing routine grading tasks to overseeing AI systems, investigating unusual cases, training new AI models, and exploring new frontiers in gemology. This evolution mirrors what has occurred in many other fields where automation has transformed but not eliminated the need for human expertise.
When you join the IGS community, you get trusted diamond & gemstone information when you need it.
Diamond grading stands at a technological crossroads where artificial intelligence meets centuries-old gemological expertise. Since GIA introduced the first official grading system in 1953, the industry has transformed from subjective visual assessments to sophisticated automated analysis. Today's laboratories employ spectrophotometers, 3D mapping systems, and machine learning algorithms that can analyze thousands of diamonds with remarkable precision.
This article explores:
Before 1953, diamond grading lacked global consistency. No officially recognized system existed, and jewellers applied different standards across organizations. This inconsistency created confusion in the trade and complicated transactions between dealers. GIA's introduction of a standardized grading system marked a pivotal moment, establishing clear parameters that laboratories could replicate worldwide and giving consumers confidence in quality assessments.
Gemologists have long envisioned a single "black box" capable of performing every analysis task, from stone identification to synthetic detection, origin determination, and treatment indication. Complete color and clarity grading would round out its capabilities. Unfortunately, this unified device has not materialized. Instead, the industry has developed numerous specialized devices focusing on different analytical areas.
Modern technology enables unprecedented diamond analysis:
The substantial investment in diamond research has pushed this field ahead of all others in automation and AI adoption.
Since the 1950s, GIA employed colorimeters to assign color grades. These early tools operated within limited ranges, typically working only on diamonds with yellow tints. Laboratories never relied on these devices for final grades; human graders always maintained priority. As technology improved, this fundamental reliance on human vision persisted, though automated systems gradually gained more influence.
Master stones represented one of the first steps toward standardized grading. Every laboratory maintains a set of diamond masters spanning color grades from D through L or beyond. Graders carefully select these diamonds, usually at 0.50 carat or one full carat, within 0.10 carat of each other.
Ideal master stones meet specific criteria:
These master stones technically form the basis for all grading developments, extending right up to modern AI-enhanced systems.
Clarity grading occupies a gray area in standardization. Producing useful master sets for each clarity grade proves impossible due to the extensive range of inclusion combinations falling into each grade. Instead, the industry has always relied on training by experienced graders and comprehensive textbooks. Publications like "Diamond Grading ABC" by Pagel-Theisen and "Photo-Masters For Diamond Grading" by Roskin convey possibilities through multiple images, providing the visual education that master stones cannot deliver for clarity assessment.
Proportion assessment was the first area benefiting from technological assistance. Anyone who has examined a diamond understands the difficulty in accurately assessing angles or sizes. For example, crown and pavilion angles for a well-cut round brilliant should measure 30-35 degrees and 41-42 degrees respectively. Accomplishing this by eye proves extremely difficult.
The proportionscope, introduced in 1967, made the task considerably easier. GIA and Gem Instruments Corporation developed this back-projection tool for determining angles and sizes. Graders simply positioned the diamond in the holder, aligned the stone's silhouette with guide lines, and read off measurements.
In 1992, proportion grading became fully automated through the Sarine DiaMension, followed by the OGI Systems Scanox S. These machines analyze every facet, angle, and proportion of a diamond in seconds, employing lasers and reflections to measure to tenths of a degree and hundredths of a millimeter. Though not true AI, these systems rely on algorithms to determine deviations from specified tolerances and assign overall grades.
Color grading represents a prime area for AI and machine learning utilization. Graders initially performed this assessment by eye, using a white grading tray as background and comparing the diamond to this shade or to master stones.
This traditional approach suffered from several issues:
Automated color grading offers consistency, repeatability, and profitability advantages. Computers never take vacations, can operate continuously 24/7, and never complain.
Initial automation attempts employed colorimeters, either spectrophotometers or tristimulus machines. Spectrophotometers measure absorption at selected wavelengths or across the entire spectrum. Research and development has focused mainly on spectrophotometer techniques due to their relative simplicity.
Early models included the Eickhorst Diamond-Photometer, which analyzed absorption of yellow and blue light as it totally internally reflected through the diamond. The system converted the ratio between these absorption levels into a color grade. The drawback was that analysis limited to blue and yellow light meant only Cape series and Canary yellow diamonds could receive grades.
Robert Shipley of GIA developed the Shipley Electronic colorimeter in the 1940s, working along similar lines. Both models have now been superseded. Zeiss produced their spectrophotometer version that analyzed light transmission across the entire visible spectrum and into the UV range down to approximately 320 nanometers. The Zeiss system used a monochromator, which recorded each wavelength individually and plotted them as a transmission curve. This comprehensive spectral analysis represented a major advancement.
AI already operates extensively in light performance analysis, assessing the return of light from diamonds. In early days, graders performed this manually with tools such as "Hearts and Arrows" viewers, but only to a visual level. Technology has advanced significantly beyond these simple assessments.
AI systems now calculate grades based on equipment-supplied information. The Sarine Diamension, in use since 1992, provides detailed angles and dimensions. Systems then process this data through ray-trace style algorithms to calculate light return levels and assign grades. Systems such as D-Imaging from Dynagem employ this technology to provide cut, clarity, and color grades from a single unit, with digital images reviewable later to confirm clarity grading decisions.
Machine learning predominates in modern grading practices. Unlike ChatGPT and other AI models that scour the internet, diamond grading machine learning relies on iterative processes, with systems learning through repetition. It operates the same way average human graders do: a database of results compiles, providing experience that the system calls upon when grading the next diamond.
Machine learning offers a crucial advantage over human training:
Automated systems can process thousands or millions of iterations, improving their recognition and analysis ability continuously.
DeBeers operates proprietary machines for color and clarity that work with human graders to ensure reliability.
The Falcon System for Color:
The Falcon currently serves as an assistive tool but could evolve to remove human intervention entirely.
The Eagle System for Clarity:
As with the Falcon, two trained graders visually assess each diamond before all results combine to produce the final grade.
From their website and research materials, Sarine appears at the forefront of AI development in the diamond industry. They cover all fields from mapping rough diamonds and suggesting cutting options through to final grading. Sarine developed their automated color and clarity system around 2016, with GIA working with IBM following a year later in 2017. These organizations possessed the extensive datasets required to initiate machine learning research programs.
Color Capabilities:
Clarity-II Features:
Sarine acknowledges the difficulties in clarity grading, as it requires a very large sample set to cover the range of possibilities given each diamond's uniqueness.
Sarine also introduced their "Light" system, which analyzes how diamonds interact with light through reflection, refraction, and dispersion. They claim to operate the only fully automated symmetry and cut grading scanner in the form of their DiaMension Axiom. One possible issue has emerged: one Sarine representative center stated they "no longer produce a color machine," raising questions about their current focus areas.
GIA's version of AI clarity grading involves imaging the diamond, then sending digital images to the cloud. The system compares them to an expansive database of diamond grading information, and the AI computes a clarity grade.
GIA states this approach will:
The implication suggests the system will rapidly reduce or eliminate human involvement in routine grading decisions.
Gubelin Gem Lab employs AI as part of its Gemtelligence system, predominantly on colored stones rather than diamonds. This system examines various analysis results for patterns and markers that may elude human laboratory workers. In testing, Gubelin claims their Gemtelligence system has surpassed human gemologists in detecting origin or heat treatment in stones.
As with any mathematical computation, there is always a confidence interval, which can be thought of as an error margin. Gubelin indicates that any Gemtelligence result with less than 98 percent confidence would automatically undergo additional tests by at least two senior gemologists. The main drawback is that the system cannot yet detect new origins or treatment techniques, as machine learning is always based on past experience.
Although larger laboratories maintain secrecy about their AI development work, smaller laboratories sometimes speak more openly. Conversations with several smaller laboratories offering diamond grading services revealed a common concern.
All smaller labs expressed worry that:
These concerns reflect the industry's changing economics as technology costs drop and efficiency improves for larger organizations while smaller operations struggle to keep pace.
Whether gemologists like it or not, AI is here to stay and will gain further footholds into gem analysis worldwide. Diamond grading was an obvious starting point given the quantity and value of diamonds graded worldwide each year. Once laboratories perfect the technology, it will readily lend itself to analyzing any area of gemology.
However, gemologists retain important advantages:
The key lies in adaptation. Gemologists who embrace AI as a tool rather than viewing it as a threat will find themselves better positioned in the evolving industry. The combination of human insight, intuition, and experience with AI's speed, consistency, and analytical power represents the most powerful approach to gemological analysis.
Traditional skills remain valuable, particularly in areas where AI has not yet been trained or where new phenomena emerge. The gemologist's ability to recognize when something does not fit established patterns, to question results, and to investigate anomalies ensures their continued relevance. AI excels at what it knows; humans excel at discovering what remains unknown.
As technology advances, the role of gemologists will likely shift from performing routine grading tasks to overseeing AI systems, investigating unusual cases, training new AI models, and exploring new frontiers in gemology. This evolution mirrors what has occurred in many other fields where automation has transformed but not eliminated the need for human expertise.
