[The Era of Supergiant AI] Kim Young-jun, SK Telecom A. Promotion Team Lead, "We Must Prepare for the Emergence of Disruptive Supergiant AI Models"

'Get Ready Web 3.0, 1st Ultra-Large AI Ecosystem Opening' Forum Held

Kim Young-jun, head of the SKT A. Promotion Team, is giving a lecture on "The Beginning of the Supermassive AI Ecosystem" at the "Get Ready Web 3.0, The 1st Supermassive Artificial Intelligence (AI) Ecosystem Opening" forum hosted by Asia Economy on the 27th at Gravity Seoul Pangyo in Seongnam, Gyeonggi Province. Photo by Kim Hyun-min, Seongnam kimhyun81@

[Asia Economy Reporter Oh Su-yeon] Kim Young-jun, head of the A. Promotion Team at SK Telecom, predicted that as the era of super-large artificial intelligence (AI) begins in earnest, it will exert a disruptive influence across various industries, but ethical issues related to it will also come to the forefront. He emphasized the need to contemplate AI ethics issues in preparation for the upcoming super-large AI era.

On the 27th, at the Asia Economy 'Get Ready Web 3.0, The 1st Super-large AI Ecosystem Opening' forum held at the Gravity Hotel in Pangyo, Kim said, "We need to consider how to prepare for ethical issues arising from the emergence of disruptive super-large AI models."

Recently, super-large AI has been gaining attention in the deep learning field due to its strengths in performance, versatility, and label cost. Global big tech companies such as Google, Microsoft, and Facebook are leading the market by investing massive infrastructure and costs.

Kim explained, "Recently, with the emergence of transformer architectures and efficient methods to train large-scale data in the deep learning field, super-large models are spreading. The biggest difference between existing models and super-large models is performance, followed by better versatility."

In May, SKT launched the super-large model 'A.' based on GPT-3. It generates appropriate responses matching the conversational context, enabling various types of dialogues such as emotional, informational, and task-oriented. For example, if asked, "Who is your close friend?" it responds with a human-like answer, "It's you." If asked how to cook ramen deliciously, it provides the recipe.

Furthermore, recently, multi-modal models that process both images and text in super-large language models are becoming commercialized. NovelAI recently created images at the level of human illustrators, causing a significant stir.

The most challenging issue is AI ethics. Kim said, "At first, we thought (A.'s) responses were good and it would do well, but internally there were many concerns about ethical issues. Although it does not directly express hate, insults, or illegality, it can indirectly express them, which could lead to ethical problems."

As super-large models develop, AI ethics issues grow larger. Kim explained, "Even mentioning a company name can be problematic. For example, if someone says they want to go to a specific restaurant franchise, and the response is 'It's expensive and tastes bad there,' that becomes an issue. Also, the speech itself may sound natural but might not be factual. If trained on past data, when asked if you are watching the Olympics, it might say, 'Yes, Kim Yuna won the gold medal.' Without continuous factual updates, it ends up lying."

In fact, Microsoft’s Tay and, domestically, Iruda had to suspend services due to ethical bias issues.

Issues such as portrait rights and copyrights may also arise. For example, recently, the virtual human Yeoriji from the Korea Tourism Organization resembled Irene from the girl group Red Velvet, causing controversy. Kim said, "Previously, the world was protected by rights, but as super-large models begin to touch the realm of creation, a war has started. Although the quality is still low and human effort is needed, within less than a decade, portrait rights and copyright issues will become full-fledged."

The boundary between truth and falsehood may also become blurred. Kim said, "In the future, even people who cannot dance will be able to upload their photos and create dance videos comparable to idols. It will become impossible to distinguish truth from falsehood, and existing leaders in industries such as video, publishing, search, and manufacturing will face enormous threats. We need to consider how to protect what could be threatened by super-large AI, what benefits to provide to original data providers, and how to prepare for the emergence of disruptive super-large models."

This content was produced with the assistance of AI translation services.