The government will bring forward by 1 year the opening of its 'AI and high-value public data Top 100' datasets, which are in high demand for AI development. It will also open 37 types of public data and works previously provided to proprietary AI foundation model development teams so that all AI companies can use them.

Prime Minister Han Sung-sook (한성숙) chaired the second meeting of related ministers on data at the Government Complex Seoul on Aug. 27 and discussed measures to expand the opening of public data.

The government decided to complete by 2027 the 'AI and high-value public data Top 100' project, which selects and opens 100 types of public data with strong demand for AI development and large social and economic ripple effects. The completion date is 1 year earlier than originally planned.

After opening 10 datasets last year, it opened 25 this year. It will open the remaining 65 next year. The original plan was to additionally open 30 in 2027 and 35 in 2028.

The government said it opened a total of 267 high-quality public datasets from 2015 through this year. It said 3,625 private-sector services were created as a result.

As an example, synthetic data on transit card usage from the Korea Transportation Safety Authority was used to analyse the effects of operating demand-responsive transport. Case examples of statutory interpretation from the Ministry of Government Legislation and rulings from special administrative appeals were used by legal tech companies to train legal AI and as a basis for answers.

The government will also expand support for public data and works to advance proprietary AI foundation models.

It has provided elite teams participating in proprietary AI foundation model projects with about 23.93 million items across 50 types of public data and works managed by 33 ministries and agencies. The Ministry of Science and ICT and the Ministry of Culture, Sports and Tourism reviewed an additional 102 types of public data and works held by relevant ministries and agencies, based on further requests from the elite teams.

As a result, it decided to convert 37 types with no copyright or personal information issues so that all AI companies can use them, including policy materials and publications, resolutions and decisions and casebooks from various committees, and some past questions from qualification exams. The conversion will include the 'Public Nuri AI type'.

The Public Nuri AI type was created in January to allow public works to be freely used for AI training. The review confirmed 64 types of data can be provided to the elite teams developing proprietary AI foundation models. Of those, 37 types will be opened to general AI companies as well.

Each ministry plans to additionally check the status of Public Nuri labelling for public works under its jurisdiction and identify materials that can be applied to the AI type. The culture ministry will also pursue revisions to the Copyright Act to expand the opening of public works.

Han said public data are an irreplaceable national asset accumulated over a long period across all areas of the economy and society. She said global AI competitiveness hinges on how actively high-quality public data are opened and how well they are used.

Han also said public data should be opened as a matter of principle because they were built with taxpayers' money. She said the government will support the private sector so it can make good use of AI and AX data.