Korea's AI kingdom: The debate South Korea needs – and is not having

There is an invisible revolution taking place throughout Korea and the world that we do not see — it is called artificial intelligence (AI) agents, small programs designed with the use of AI that do small things on our behalf. They can sort your data, go through your shopping list, search and even buy items for you, run your schedules, draft and answer your emails and many more. What is unique is that they are created by tinkering by average persons, like you and me.

At first, many of these agents are limited and unreliable. But users test them, fail and adjust their instructions and tools, and gradually make them more capable. Some have already become reliable enough for use in businesses. Employers are now asking whether agents can run routine administrative tasks and if humans are actually needed to do them.

The consequences are huge. For now, most agents appear modest: a research assistant here, an automated scheduler there. Yet they are likely to become standard features of computers, phones, messaging platforms and workplaces far sooner than most governments are prepared for. In other words, imagine those agents as minions — some will have more of them, some fewer and many may have none at all. And that is the main topic of my pitch today — the AI divide, or, more accurately, a series of overlapping divides.

After several years of following AI development and its growing use in public administration, I see three big areas screaming for an urgent debate: who benefits from the technology, who understands it and who controls its deployment.

The first divide is already more or less apparent — the economic one. As semiconductor companies are raking in huge profits, their employees are seeing multidigit increases in income. Want it or not, the impact of the money is visible everywhere starting from the apartment and stock prices all the way to consumer products and delivery services. The market is bound to react to this group of nouveau riche, as they will be hunting for the best, while feeling that they do deserve it. On the other hand, if you are not affiliated with the industry in any way, you are stuck living with a salary that is way behind by almost a decade.

Then there is a know-how divide with those who understand and are able to tap into the new opportunities AI is providing, and those who can't. This one is very much a reminiscent one mirroring the digital divide of the 1990s. It may lock out young generations that are still growing and have yet to learn about it. In fact, contemporary education is only able to offer knowledge and credentials to the way of living of the past. And this way is going away. Then there are the older citizens who are still grappling with the rise of the internet in their lives and now AI.

The third divide — the most political one — is the one of control of AI. Companies are deploying their own AI, without asking serious questions. For example, Kakao has added ChatGPT service to its ecosystem, and universities and public agencies are eager to do the same. However, they are not asking whether people want them or not. They are making decisions for us, claiming that AI’s efficiency is a public good. Suddenly, we find ourselves surrounded by AI in various forms.

AI competition is already geopolitical. The Korean government placed a huge investment drive, $650 billion dollars to be exact, to create infrastructure that will pave the way to sovereign AI. On the one hand, the government is pushing AI into our lives by giving us an army of AI minions. And on the other, it realizes that AI will eventually cause a lot of disruption, to the point that even Labor Minister Kim Young-hoon acknowledges that we need “a new social contract.”

The debates about AI and its place in our lives are superficial is the least to say. We need to place philosophy, ethics, law and social science inside the institutions that design, procure, test and govern AI systems. Just hoping that AI will be nice to us is dangerous. We simply do not know exactly how AI “thinks” per se, how it comes to certain decisions. We know in general terms, but not what is “under its hood,” or in other words, we don't know how millions or billions of internal interactions produce a particular answer, strategy or failure.

The goal of AI companies, or frontier labs, is to reach a point when their models would design, create and train other models without any human supervision. And that is a scary idea. Here is a little bit of food for your thought. Just a few days ago one of those labs was testing one of its newest AI models on another company’s infrastructure. The model hacked the system, got out of the box and found out all the questions it was going to be asked and the best answers to them, then tried to change the environment to execute the test of itself when it was caught red-handed, luckily without access to the internet.

I am oversimplifying the incident, but it did really happen. If AI is already capable of doing that, how far are we from the point of AI pervasively controlling all dimensions of our lives and shaping our ideas? What happens if they find their way to banking, public administration, education, health care and national security? Where and how should the government stand in this situation? What are the guarantees of safety and security? These are the hard questions we should be trying to answer and act upon before AI systems become so embedded in public and commercial life that a true democratic choice is no longer possible.

Eugene Lee (mreulee@gmail.com) is a lecturing professor at the Graduate School of Governance at Sungkyunkwan University in Seoul. Specializing in international relations and governance, his research and teaching focus is on national and regional security, international development, government policies and Northeast and Central Asia. The views expressed in this article are his own.