From Detecting AI Errors to Redressing Harm... National Assembly Futures Institute: "Response Systems Must Be Linked"

Concerns over accumulating harm from AI errors... limits of service-recovery responses
Proposes joint agency action and users' right to request reviews

The National Assembly Futures Institute proposed establishing an integrated response system that covers everything from detecting errors in AI decisions to correcting them and providing redress for resulting harm. The institute said the current response framework, which focuses on restoring services after disruptions, should be expanded because harm can accumulate when AI repeatedly makes incorrect decisions even while operating normally.

On November 11, the National Assembly Futures Institute published a brief titled "AI Integration and Digital Service Resilience," outlining three policy directions.

As AI adoption in real-world work continues to spread—including AI-assisted medical imaging analysis at hospitals, welfare counseling by local governments, and conversational bank transfers—the risk of malfunctions is also growing. Errors may not be detected until harm has accumulated from repeated incorrect decisions in areas such as loan or welfare benefit assessments, even when the systems are operating normally.

The institute noted that South Korea has relevant laws, including the AI Basic Act and the Personal Information Protection Act, but said that their scopes and procedures are fragmented and responsibility for responses is divided among multiple agencies. As a result, it said, there is no adequate system that extends from correcting errors to providing redress for harm.

The institute proposed establishing criteria for restricting the use of AI or switching to human review when errors exceed a certain threshold, as well as setting up joint response procedures for the lead agency and other relevant agencies. It also stressed that when AI decisions affect people's rights and safety, users should be guaranteed the right to request a review regardless of whether the decision was automated. In addition, other users harmed by a confirmed error should be identified and their cases corrected.

Yoo Hee-soo, a research fellow who wrote the report, said, "We need to connect into a single system the measures for detecting errors, correcting them, and providing redress for harm, starting with services that affect people's rights and safety."

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