[Initial Insight] Media Articles Like 'Service Gunmandu' Even in the AI Era

Developing Korean-Style Generative AI Is Important,
But Preparing for What Comes Next Matters Too
Generative AI and News Copyright Infringement
Preemptive Measures Needed to Prevent Excessive Legal Disputes

Those who get their news solely from YouTube might not be aware, but quite a few well-known YouTubers copy article headlines directly onto thumbnails (preview images) and recite the article content in their broadcasts. Expecting them to cite sources at the end of the video is a mistake. For Koreans, articles are a service freely accessible on portals and social networking services (SNS). As a result, it is taken for granted that they are free, like complimentary fried dumplings at a Chinese restaurant.

Regrettably, generative artificial intelligence (AI) has inherited this worldview from YouTube. When asked questions in Korean, OpenAI's ChatGPT or Google's Gemini produce answers that seem to be learned from domestic news articles. Of course, they do not disclose sources like YouTubers do. Naver’s generative AI HyperClover initially responded to users in this manner as well. However, after Korean news organizations raised issues about unauthorized use of articles, it stepped back. It switched to learning from Naver’s community posts, blogs, and comments instead of articles. But if you look closely, most of the various types of writing people upload are either copied wholesale from articles or written after reading articles.

China’s DeepSeek has shocked and simultaneously inspired the domestic AI industry. Korean AI companies are now filled with expectations that they can create 'low-cost, high-performance' generative AI on their own. "We will allocate supplementary budgets to purchase thousands of graphics processing units (GPUs)" (National Assembly), "We will accelerate the establishment of a National AI Computing Center" (Ministry of Science and ICT), "We will open-source models at the level of DeepSeek" (LG AI Research Institute). All of these developments occurred within the past week.

To enhance Korea’s AI competitiveness, developing generative AI models is important, but preparing for the future is equally crucial. The key is to train AI with trustworthy content through legitimate methods. Last year, OpenAI signed a contract with Time magazine and trained ChatGPT on 100 years’ worth of articles. Mistral AI, known as the European version of OpenAI based in France, also signed a news provision contract with AFP. No matter how much information is available on the internet, training AI with verified and refined information is proof that it can improve the accuracy of responses.

Before reaching such agreements, there were intense legal battles worldwide between big tech companies and news organizations over news usage rights. Korea should take this as a lesson and consider preemptively organizing copyright issues through legislation. If lawsuits continue to proliferate in Korea, where AI competitiveness must be strengthened, it would be like tripping a marathon runner who is already behind. Incidentally, the Ministry of Culture, Sports and Tourism is reportedly preparing to propose an amendment to the AI Basic Act that would require generative AI operators to disclose the sources of AI training data. If this amendment passes, policies on AI training and news copyright will naturally become a topic of public discussion.

I opened the Gemini application and typed, "Critically evaluate Korea’s low birthrate policy based on articles." The response was fluent but did not specify exact sources. When I asked again, "Which articles did you base your answer on?" it finally showed the names of five or six Korean news organizations and article titles. Then I asked, "Is it okay to use Korean news articles for free?" Gemini excused itself by saying, "I synthesized multiple articles and did not copy any specific article."

Finally, I asked for opinions on "How should copyright issues be resolved?" Gemini cited "contracts and licensing" as examples. So far, I have never heard of any big tech company paying Korean news organizations for news usage fees. However, their AI already knows what should be done going forward.