Government-level standards for ‘open data on the Web’ utilized in the event of generative artificial intelligence (AI) models have been released. Nonetheless, they don’t provide specific examples, but relatively provide a basic framework for future discussions.
The Personal Information Protection Commission (Chairman Koh Hak-soo) announced on the seventeenth that it has prepared a ‘Guide to the Processing of Open Personal Information for AI Development and Services’ to be sure that open data essential to AI development could be processed legally and safely throughout the current personal information regulation system.
Public data refers to data that anyone on the Web can legally access, reminiscent of Common Crawl, Wikipedia, blogs, and web sites.
The Personal Information Protection Commission announced that it has prepared a guide that clarifies the legal standards for collecting and using disclosed personal information and presents the minimum standards that corporations can discuss with regarding what safety measures are appropriate to take throughout the AI development and repair stages. It also explained that it focused on establishing internationally interoperable standards, taking into consideration situations within the European Union (EU) and the US.
A very powerful a part of the announcement that day is that “personal information disclosed under the ‘legitimate interest’ clause of Article 15 of the Protection Act could be used for AI learning and repair development.”
Article 15, Paragraph 1, Item 6 of the Personal Information Protection Act stipulates that “in cases where it’s mandatory to attain the legitimate interests of the private information processor and where it clearly takes precedence over the rights of the information subject. On this case, it is proscribed to cases where it’s significantly related to the legitimate interests of the private information processor and doesn’t exceed an affordable scope.”
It also stated that to ensure that the ‘legitimate interest’ clause to be applied, three requirements have to be met: legitimacy of the aim of AI development, necessity of processing disclosed personal information, and specific profit balance.
Nonetheless, this can’t be regarded as a final standard, as there isn’t a specific explanation for the legitimacy part and no detailed guidelines for individual cases.
For instance, the EU is currently in a trend of prohibiting AI model training without consent from users’ posts on social media. Accordingly, the default setting is ready to ‘consent to make use of’, and the ‘opt-in method’, which requires users to search out the refusal items and express their refusal, is prohibited.
Meanwhile, within the US, Metana X (Twitter) is using SNS data for learning by obtaining user consent through an opt-out method.
On this case, detailed criteria for specific cases are needed. Nonetheless, the cases presented on this guidebook are mainly principled cases reminiscent of prohibiting illegal use or not including sensitive personal information.
As well as, this guidebook only covers the private information aspect amongst the varied elements of information learning, and doesn’t address copyright issues. This can also be something that the relevant departments might want to improve in the long run.
The Personal Information Protection Commission also announced that the guidebook can be constantly updated in consideration of future enactment and revision of private information-related laws, trends in AI technology development, and overseas regulatory development trends.
He added that he would further specify the premise and standards for the legal processing of users’ personal information after gathering opinions from academia, industry, and civic groups.
As well as, this guidebook only presents standards for legal interpretation and has no legal binding force.
Professor Kim Byeong-pil of the Korea Advanced Institute of Science and Technology, who participated within the discussion regarding this guide, said, “I consider that the guide will function an excellent reference material for developing and using reliable AI.”
LG AI Research Institute Director Kyung-Hoon Bae, who can also be the co-chair of the public-private policy council, said, “The publication of this guide is a vital step forward and a primary step toward achieving each AI technology advancement and private data protection.”
Personal Information Protection Commission Chairman Koh Hak-soo said, “I hope that through this guide, corporations will give you the option to create AI and data processing practices that the general public can trust, and that the accrued best practices can be constantly reflected within the guide.”
Reporter Park Soo-bin sbin08@aitimes.com
