ETRI develops explainable AI decision-support technology, wins international standard approval

The Electronics and Telecommunications Research Institute (ETRI) said Wednesday it has developed core technology for an "explainable AI decision-support" system that can not only present reasoning results based on expert knowledge but also explain the grounds and rationale behind its judgments and independently verify the factual accuracy of its outputs. The International Telecommunication Union (ITU) has also approved an international standard covering the system's architecture and requirements.

The standard provides a common framework that companies and public institutions can use when designing and building explainable AI services.

It defines the basic architecture and requirements for a system that goes beyond simply stating "this is the answer" — enabling AI to explain that "this result was derived based on the following grounds and reasoning, and this portion requires further verification."

The standard encompasses a range of capabilities: providing results and explanations in response to user queries, expert knowledge-based reasoning, delivering reports containing the basis for judgments and analytical content, identifying information that requires additional verification, incorporating user feedback, continuously updating evolving expert knowledge, and handling information in multiple formats — including tables, images and diagrams, not just text.

The standard is not tied to any specific AI model or product. Rather, it serves as a common framework for companies and public institutions to draw on when designing explainable AI services going forward.

With this technology, AI can move beyond simply generating fluent sentences — understanding the context of a question, reasoning on the basis of expert knowledge, and then providing the grounds and rationale for its judgment along with verification results.

Through explainable classification, self-verifying retrieval-augmented generation (RAG), output factuality verification and interactive report generation, ETRI has established a foundation for applying the technology in fields where reliability and accountability are paramount, including medicine, law, finance and public administration. The technology is also expected to support compliance with explanation obligations for high-impact AI under Korea's AI Basic Act.

"This international standard is among the first internationally recognized criteria to define the common architecture and requirements that explainable AI must meet," said Lee Yong-ju, head of ETRI's Visual Intelligence Research Laboratory. "We expect it to serve as an important benchmark for companies and public institutions as they build trustworthy AI services and fulfill the explanation obligations required under the AI Basic Act."