Korea considers watermarking AI-generated content to mitigate side effects

[Image source: Gettyimagesbank]

In line with the White House and AI giants like OpenAI and Google agreeing to watermark AI-generated content for enhanced safety, South Korea’s IT authorities are considering a similar policy. According to a key official from the Ministry of Science and ICT on Friday, the ministry is actively engaging in discussions with the AI sector regarding the implementation of watermarks on AI-generated audios, videos, images or texts.

This effort mirrors a recent U.S. initiative where major AI firms collaborate with the government to employ watermarks on AI-created content. This step aims to mitigate potential negative consequences of generative AI like human rights issues and the proliferation of deepfake-driven misinformation, all while upholding copyright protection for the original training content.

Technical aspects of watermark implementation for AI content remain under deliberation between the Korean government and the industry.

Prominently displaying “AI Generated” either at the start or end of a video, image, or text is the favored approach. However, its effectiveness is hindered when content creators omit intentional watermarking on AI-produced material.

To address this issue, researchers both at home and abroad are exploring the use of “non-visual watermarks,” which entails automatic recognition of AI-generated content without relying on human visual verification. This technology mechanically identifies AI-generated content by watermarking datasets used for AI learning as they pass through a large language model (LLM).

“Some may raise concerns that attempts to watermark AI-generated content will shrink the AI industry, but in a situation where there is no way to stop the utilization of AI technology, it may be more positive for the development of the AI industry to give guidelines to legitimately identify and widely utilize AI products,” said an official from the ministry.

Andrew Ng, one of the world’s leading AI experts and a professor at Stanford University, also commented on the U.S. AI companies’ decision to use watermarks, saying, “Most voluntary commitments on AI ethics are vague, but this is a concrete and actionable commitment.” “Society would be better off if we could know what content is AI-generated and what is not,” he added.

The Korean government, in the meantime, has revealed its plan to introduce technical complementary measures to mitigate AI side effects next month, even before the pending “AI Basic Bill” is passed by the National Assembly.

“We believe that we need to create technical measures against AI side effects such as fake news and intentional attempts to misuse AI,” said Park Yun-Kyu, vice minister of Science and ICT, at a recent press meeting.