Is Korea ready for the regulation of AI?

LEE MOON-TAE
The author is head of the LG Advanced Machine Learning Lab.

On October 30, U.S. President Joe Biden signed the Executive Order “Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence.” The executive order contains comprehensive guidelines and regulations for the establishment AI governance in response to the side effects and potential threats of generative AI.

AI governance reflects a global leadership battle. The executive order contains content that not only AI researchers but also the general public should carefully review.

First, AI governance must reflect changing technology trends. The executive order defines the standard for “AI foundation model” as a model “that is trained on broad data” and “contains at least tens of billions of parameters.” Papers published this year claim that algorithms with just 1 billion parameters can approach the performance of existing large models. Efficiency is the major trend of recent studies. Therefore, AI governance should not be limited by specific numbers and should consider the scope or influence of AI capabilities.

Second, we should be wary of blind faith in concepts. The word “bias” appears nine times in the executive order. The word itself sounds negative, but humans are not free from bias in any processes. Of course, bias against identities — such as specific genders or occupations — should be removed. However, if you do not carefully classify which information belongs where and just hold unconditional wariness about bias, AI may end up injecting “a bias to eliminate biases.”

Third, a philosophy on hallucinations is needed. As AI advances, the amount of generated data increases, and it becomes increasingly difficult to determine authenticity. Not long ago, a lawyer who worked for 30 years in the United States submitted a precedent written by generative AI to a court and found himself sanctioned. To secure trust in the AI era, a process to verify how certain information was created is needed.

In the wake of Biden’s executive order, discussions and movements around AI governance are heating up in the international community. Korea has also established the National Data Policy Committee and the AI Ethics Policy Forum to actively analyze such topics. In order to properly realize AI governance, it is necessary to understand technology trends and generate specific ideas for innovations that humans and technology can develop together.