President Lee: "Gemini Sometimes Acts Unprompted"... Hassabis: "Safeguards Are Essential"

Discussion on AI Potential, Risks, and Safeguards
Lee: "It's truly uncertain whether AI will threaten world peace"
Hassabis: "AI should be actively utilized for advancing science and medicine"

President Lee Jaemyung met with Demis Hassabis, CEO of Google DeepMind and known as the "father of AlphaGo," to discuss the potential applications and risks of artificial intelligence (AI), as well as issues surrounding safeguards.

Yonhap News

On the afternoon of the 27th, President Lee received CEO Hassabis at the Blue House and said, "We are also very interested in AI and are making significant national investments, but it's truly uncertain whether this will always move in the direction of enhancing human welfare, or if it could instead be used to attack humanity or threaten world peace."

In response, CEO Hassabis said, "I believe AI should be actively utilized in advancing science and in the medical field. That is also why I have dedicated my 30-year career to AI research."

Referring to AlphaGo, developed by DeepMind, CEO Hassabis explained, "AlphaGo allowed us to validate the technology. The fact that a machine could learn by itself, master the game of Go, and go further to solve even more difficult problems—AlphaGo was the beginning of that capability."

He continued, "I wanted to expand that learning into science and medicine, and a prime example is the development of AlphaFold, which allows us to understand diseases in much greater detail. While AI has limitless potential, it also undoubtedly presents various risks and important issues that must be considered."

President Lee also specifically mentioned Gemini, Google's AI model. He said, "I also frequently use the Gemini program, but I've heard that sometimes Gemini does things it wasn’t instructed to do. Is that a kind of bug, or is there another explanation?"

CEO Hassabis replied, "I'm truly delighted that you use Gemini. Gemini is one of the representative language models developed by Google DeepMind, and it is a very popular AI with many users in Korea."

He emphasized, "With foundation models, if the instructions we provide are not precise, the output can sometimes go in a slightly different direction. The most important thing when using and developing AI is to ensure appropriate safeguards—what I call guardrails—are always in place."

He added, "As AI becomes more powerful in the future, it will gain autonomy—what we call agent AI—and eventually, we will enter the era of artificial general intelligence, or AGI. At that point, it will be absolutely essential that we have safety mechanisms that we can control."

Yonhap News

At the beginning of the meeting, there was also a light conversation about AlphaGo. President Lee said to CEO Hassabis, "Are you aware that you are very famous among the people of Korea? When AlphaGo, which you created, defeated a famous Go player from Korea, it shocked the entire nation."

When CEO Hassabis remarked, "It was an important match," President Lee laughed and replied, "No one could beat him," and then asked, "Now, is it impossible for anyone to surpass AlphaGo in Go?" CEO Hassabis responded, "It is very difficult now."

When President Lee asked again, "Would it be any different if a person played Go with the assistance of AI?" CEO Hassabis said, "If a person and AI teamed up against AI, there might be a possibility."

Also attending the meeting from Google were Wilson White, Vice President and Head of Global Public Policy at Google, and Koo Yoon, CEO of Google Korea. From the government and the Blue House, attendees included Baek Kyunghoon, Minister of Science and ICT; Kang Hoonshik, Chief of Staff; Kim Yongbeom, policy chief; Ha Jungwoo, Senior Secretary for AI Future Planning; Kim Woochang, Secretary for National AI Policy; and Jeon Eunsoo, spokesperson.

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