ETRI Establishes International Standard for 'AI Testing'... "Achievement After Five Years"

An international standard for verifying the safety and reliability of artificial intelligence (AI) systems has been established under the leadership of a Korean research team.

The Electronics and Telecommunications Research Institute (ETRI) announced on November 19 that the "Overview of AI System Testing" standard, which defines the procedures and methodologies for testing AI systems, has been officially established through the International Organization for Standardization (ISO/IEC JTC1).

Jeon Jonghong, Senior Researcher at ETRI, who led the establishment of international standards for AI testing (right), and Dr. Stuart Reed, co-editor and chair (left). Courtesy of Electronics and Telecommunications Research Institute (ETRI)

This achievement is significant as it represents the first core international standard for AI testing initiated by Korea within the ISO/IEC AI Technical Committee (SC 42). It demonstrates that ETRI has positioned itself as a "first mover," rather than a "fast follower," in the global competition for AI technological leadership.

The "Overview of AI System Testing" standard is the first to define end-to-end testing methodologies for aspects such as data quality, model performance, and bias in AI. It is expected to be utilized in future internationally accredited testing and international conformity assessments.

In detail, the standard expands existing software testing standards to suit AI systems, introducing new testing stages tailored to AI characteristics, such as data quality testing and model testing. This is recognized as establishing a foundation for comprehensively verifying all core elements of AI systems, from data quality to model performance.

ETRI has also introduced the concept of "risk-based testing" to proactively assess potential risks associated with AI. Through this, AI-specific testing procedures have been concretized, including: ▲ "bias testing" to verify AI bias, ▲ "adversarial testing" using changes in input values, and ▲ "drift testing" to detect performance degradation during operation.

This standard is expected to serve as a "general framework" for subsequent standards, such as AI red team testing and generative AI testing, that will be established in the future.

ETRI President Bang Seungchan stated, "Ensuring the safety and reliability of AI is a core challenge in the AI era," adding, "The establishment of an international standard for AI system testing will serve as a foundation for Korea to lead not only AI technology but also the norms for AI testing and evaluation."

This content was produced with the assistance of AI translation services.