[Hotterview] Rep. Lee Haemin, who proposed 'High-Impact AI', says "Regulation must be practical"

Google Alum Lee Hae-min, Lawmaker of the Joguk Innovation Party
"'Affected Parties' Must Also Be Protected for Users"
"Public Data and Renewable Energy Needed for Promotion"

The 'Basic Act on the Development of Artificial Intelligence and the Establishment of a Trust-Based System' (AI Basic Act), which is highly demanded by the industry, is set to pass the National Assembly this year. The AI Basic Act has been under discussion since 2021 and last week, it passed the full meeting of the Science, Technology, Information and Communications Committee, the standing committee in charge, through bipartisan agreement. After the Legislative and Judiciary Committee’s textual review and the plenary session’s approval, the law will be finalized.

The AI Basic Act is legislation aligned with the global trend of organizing AI-related legal frameworks and regulations. Regarding this, Representative Lee Hae-min of the Innovation Party held a Q&A session with reporters for over two hours at the National Assembly Members’ Office Building at 6 p.m. on the 28th of last month. He is the lawmaker who proposed the term 'High-Impact AI,' which is the core of Korea’s AI Basic Act. Representative Lee explained the legislative background and his own specific industrial promotion measures, even skipping dinner.

Lee Hae-min, member of the Joguk Innovation Party.

What was important when creating the AI Basic Act?

We strived to create the most advanced law that is globally applicable. Europe has introduced many regulatory laws starting with GDPR (General Data Protection Regulation), and the United States basically started from industrial promotion. However, the U.S. eventually imposed regulations due to ethical issues. Europe is concerned about 'why big tech companies do not emerge here.' Since Korea is about three years behind, I think an AI Basic Act positioned between Europe and the U.S. is necessary.

Why did you propose the more inclusive concept of 'High-Impact AI' rather than the term 'High-Risk AI'?

Using AI models, an enormous number of services can be created. Obligations can be imposed on the service creators or direct users. However, people who are actually affected might be excluded. For example, a celebrity whose voice was used in training a voice model is an 'affected party.' These people need to be protected.

The industrial classifications included in the two terms are not significantly different. However, there is a philosophical difference. It can be considered a more developed version after more deliberation. Instead of regulating 'Risk,' it regulates 'Impact,' aiming to protect not only users but also those affected. We included detailed provisions for technologies that can be classified as High-Impact AI. Also, we left room to add things we cannot yet imagine in the future.

Assemblyman Lee Hae-min of the Party for National Innovation was invited to the AI Basic Act study meeting held at the National Assembly Members' Office Building at 6 p.m. on the 28th of last month, where he engaged in a Q&A session for over two hours. Photo by Assemblyman Lee Hae-min's Office

There is civil society backlash over the absence of 'Prohibited AI' in the law.

Regulation must be practical. You cannot regulate imagination. The Basic Act is like creating a framework. We made sure the Basic Act contains sufficiently flexible and expandable content. That way, specific laws can be created in various fields. Although the purpose of the Basic Act is regulation, I believe the greatest purpose of regulation should be promotion. It is like how the automobile industry developed under traffic rules such as traffic lights. The AI Basic Act is just the beginning.

What specific measures are there for promoting the AI industry?

I have experience at Google where AI functions were integrated into LMM (Large Multimodal Model) services. The AI industry can be broadly divided into areas that operate data centers using chips, areas that develop models through software, and areas that create services like ChatGPT. The support needed varies by company depending on the area.

The most important thing is to unlock public data that machines can read. The industry demands that all text readable by humans should be machine-readable. The reason for proposing the Judgment Disclosure Act was with AI industry promotion in mind. It is best for companies to use high-quality, refined public domain data as training data. Doing so increases the amount of training data and can stimulate startup activities.

Also, large companies capable of operating data centers should be supplied with electricity that meets the globally recognized 'RE100 (Renewable Energy 100%)' standard. The government should enable the use of renewable energy supported by public assistance.