Ministry of Statistics to build AI-powered early-warning system for consumer prices

South Korea's Ministry of Statistics plans to build an AI-powered early-warning system that continuously monitors price movements for 21 everyday items, including eggs and processed foods.

The system will track near-real-time price changes in key goods separately from the monthly consumer price index, feeding the data to policy ministries. The ministry also plans to lay a national statistical foundation to reduce AI "hallucinations" — instances where AI models cite incorrect statistical figures.

Minister Ahn Hyeong-jun unveiled the AI-based national data innovation plan Wednesday at the second "People's Policy Briefing" session chaired by President Lee Jae-myung at Cheong Wa Dae's banquet hall.

The ministry will select 21 items from the 458 goods that make up the consumer price index — focusing on processed foods and manufactured goods with high price volatility and strong public impact — and subject them to AI-based price monitoring. The analysis will be conducted independently of the regular CPI compilation process and shared with relevant policy ministries on an ongoing basis.

"In the past, we compiled and published a monthly price index. Going forward, we plan to use AI to analyze whether prices for key items are stable, whether they have risen sharply, and how they are likely to move — and provide that analysis to policy ministries," Ahn said. "It is a kind of early-warning system that checks information in advance to support price policy."

He added that the initiative was not about creating a new consumer price index, but about continuously monitoring price movements in key items to signal when a policy response may be needed. "For now, we are primarily considering use by policy ministries, and we plan to decide whether to make the data public after the system is built," he said.

The methodology for compiling the consumer price index will also be partially revised.

Jeonse and monthly rent prices, which surveyors previously collected through on-site visits, will now be calculated using administrative data such as housing lease registration records. "As administrative data has accumulated steadily, we plan to use jeonse and monthly rent records to compile a housing rent index," Ahn said. "When we compared the figures against historical data, there was no significant difference from the existing survey-based method."

The ministry will also expand the infrastructure for AI use in national statistics. It plans to build a pan-government data management framework so that AI can draw on policy data from across ministries, and will develop a statistical "ontology" — a structured map of relationships among datasets — to help AI accurately interpret national statistical data.

"Current LLMs frequently present incorrect statistical figures," Ahn said. "If all national statistics are structured in ontology form, AI will be able to read the Ministry of Statistics database directly, and we expect it can be used in the statistics field with almost no hallucinations." He added that a pilot test confirmed the approach works in practice, and that while international organizations are exploring similar directions, none has fully implemented such a system yet. "We will push ahead as a leader in this area," he said.

Five statistical surveys have been selected for the pilot metadata build: the Economically Active Population Survey, the Regional Employment Survey, the Internal Migration Survey, the Cause of Death Survey, and the Mining and Manufacturing Survey.

Alongside this, the Ministry of Statistics will establish a national data framework using a "hub-and-spoke" model that connects data scattered across ministries via a dedicated network rather than centralizing it in one place. The ministry plans to apply homomorphic encryption so that data can be linked and processed even in an encrypted state, securing both usability and data protection in the AI era.