[10th EAI Academy] ⑧ Artificial Intelligence and South Korea's AI Policy

YouTube Link: https://www.youtube.com/watch?v=Woxx63bfJ64

■ Paek Seo-in_Professor, Department of Global Culture and Commerce, Hanyang University.

Video Transcript

Good evening. Thank you for being here so late on a weekday. My name is Seo-in Baek from Hanyang University. You might have been surprised by all the talk about technology during the introductions, but my undergraduate major was in engineering. I believe we have one student today who also came from an engineering background. Are you here? Ah, yes. Welcome. I only studied engineering for my bachelor's degree; my master's and doctorate were in technology management. After graduating, I've continuously worked in technology policy. So, my focus has been on technology-centered management, then policy, and now, as AI itself has become paramount, my research subject has shifted slightly toward AI. I believe we shared some reading materials before this class. Have you had a chance to look at them?

Last year, after the change in government, South Korea created something called the AI Action Plan. We developed the Republic of Korea AI Action Plan because the United States had created one first, followed by China, and we felt we couldn't afford to fall behind. We created 100 action items that we need to implement immediately to become one of the top three global AI powers, or 'AI G3.' I was responsible for creating item number 98. It deals with how to strategically formulate global AI diplomacy. So today, I plan to share some of my practical experiences in this area. I imagine AI has been mentioned at least briefly in almost all of your lectures.

The Impact and Geopolitical Implications of AI

I'm sure it was mentioned in today's defense lecture, and Professor Hyung-ja Bae probably discussed it extensively as well. I expect it will also be a major topic in Professor Seung-joo Lee's upcoming lecture. You can basically think of it as a default topic now. In social sciences, especially for this audience, where most of you are majoring in political science and international relations, almost every paper seems to be about AI. This is true for nearly every field. Engineering is about AI, business administration is about AI, social sciences are about AI, policy studies are about AI—everyone is focused on AI. The key question, particularly in political science and international relations, is how should we view AI? There are many potential comparisons: nuclear power, electricity, and so on.

In some fields, like business administration, there's a tendency to be very optimistic, viewing AI as the new electricity that makes the previously impossible possible. In political science and international relations, however, there seems to be a greater focus on the significant risks. Personally, while it may still take some time to fully unfold, I believe its impact is akin to all of those things combined. I think its ripple effect will be comparable to that of electricity, nuclear power, and oil all put together. So, while it holds infinite potential, it also carries enormous risks. It carries a great many risks, and perhaps more importantly, as I'll discuss later, it has a very, very high probability of entrenching polarization.

The US-China AI Competition and the Securitization Perspective

That point seems to be a crucial one. When did AI become so important? If you look into the history of AI, you'll find that debates have been going on for a very long time. However, the AI we are seeing now—the post-Transformer era of AI—seems to have gained prominence in non-engineering fields around 2016. There was a growing sense that this technology could no longer be left unchecked. More importantly, in the context of US-China relations, there was a realization that China was catching up much faster and more threateningly than expected. This seems to have created a sense of crisis. As a result, if you look at US AI policy today, it is very much a security policy. It heavily focuses on checking its rival, implementing export controls, and prohibiting investment in Chinese companies. It even restricts sales to the US if Chinese components are used. These are very strong policies. If you want to see the origin of these measures, you should look at the 2021 report from the National Security Commission on Artificial Intelligence (NSCAI).

This new organization was formed, and it produced a report of about 700 pages. In truth, almost every tool the US could possibly use is outlined in those 700 pages. The question is how those tools have been developed since. While some recent measures may not be in there, nearly every instrument we can think of is included. They are just being rolled out with a time lag. This marked the point when the US perspective on AI shifted. It was no longer seen as a general internet technology but as a security asset. It necessitated dividing partners into trustworthy and untrustworthy camps, ensuring the US remained number one, and preventing a competitor that threatens its system from surpassing it.

It cannot allow that to happen. This is when this securitization perspective was fully embraced. The commission itself has 'Security' in its name, which clearly shows the strong tendency to view AI through a security lens. In parallel with this, the two forces affecting nearly every sector and social domain across the globe are, first, geopolitics, and second, AI itself. The development of AI is happening much faster than we think. It's incredibly fast, across many sectors. It's been just over three years since ChatGPT was released, but in that time, all sorts of things have become possible.

All sorts of things have become possible, the technology has evolved significantly, and many competitors have emerged. And surprisingly, Claude recently surpassed ChatGPT. It started generating revenue, and its corporate value became much higher. Ironically, ChatGPT, the original and leading company, has postponed its IPO, while Claude is scheduled to go public this year. We are seeing strange phenomena like this. In the world of science, the most surprising development was undoubtedly related to the Nobel Prize. The winners of the 2024 Nobel Prizes in both Physics and Chemistry were all deep learning researchers.

Deep learning researchers. This was, in fact, much faster than we thought. Typically, a Nobel Prize is awarded to individuals who have made a tremendous impact in their academic field over several decades. Of course, someone like Professor Geoffrey Hinton has been working on this for decades, even when no one else was. But someone like Demis Hassabis, the CEO of Google DeepMind, while he has also been researching for a long time, has only been applying AI to this specific area for a short while. His first major achievement was AlphaGo in 2016, followed by AlphaFold, which is used for synthesizing proteins in chemistry. Now, everyone in that field uses it by default, just as we use ChatGPT. A world without it is now unimaginable.

So, this achievement seems to have come from a research career of less than 10 years. On one hand, you could say this is astonishingly and incredibly fast. But on the other hand, we need to recognize that the timeline for AI is different from what we are used to. In AI, one year can be like ten years, and sometimes a month is considered a year. That cycle is accelerating. So, in reality, the organization led by Dr. Demis Hassabis has accomplished an enormous amount in just six or seven years.

The Pace of AI Development and Changes in the Scientific Community

In fact, it's not just about AI; there are about five or six Nobel laureates among the researchers affiliated with Google. That's more than the entire country of South Korea. So, this landscape is also changing very rapidly. And not too long ago—though personally it feels like a very long time ago—the emergence of Mithos has intensified things even further. With the advent of AI that develops other AI, like Mithos, the pace has accelerated dramatically. Initially, there were people developing AI. They would nurture it well and then have it do their work for them, transcending time and space, day and night.

The reason new, powerful models are emerging at an ever-faster rate is that AI itself is now doing the development. Data tracking this trend shows that, on average, more than one new, meaningful model is released every day. And then, AI started breaking through decades-old barriers in fields like security and defense. This has been a huge shock, and the shockwaves are still being felt.

I'll talk more about this later, but this came out of the United States. Until just before this, the US was concerned about China's rapid rise. The thinking was, 'We are ahead now, but China is right on our heels. The way we currently foster and regulate technology won't allow us to compete with China. We can't widen the gap.' So the approach was to push technology development aggressively, removing all obstacles. In short, the idea was to act more like China. That was the prevailing mode. But after Mithos appeared, the US was shocked, as was the rest of the world. The sentiment shifted to, 'Maybe we should pause for a moment. Things we can't even imagine might happen.' So that's the current situation.

Other countries are in the same boat. When Mithos was first released, it was available for a few days, then Fable-5 was available, and then it was suddenly cut off. The cutoff was based on nationality. US citizens could access it, but everyone else was blocked. For countries like ours, which rely heavily on the US for defense—and this would be true for many middle powers—this created a realization that the best defense-related technology is controlled. This led to a kind of trauma.

A trauma that made us think, 'Ah, we need to develop this kind of technology ourselves, and quickly.' Of course, access was later relaxed a bit. So, the situation is quite chaotic right now, and that's true for the United States as well. The US is not a monolith, after all. I'm sure there's some internal confusion at Anthropic too. The point is that technological development has become far faster than we thought. It seems to have surpassed even the imagination of the developers themselves. This is similar to what I tell my students: trying to predict when AI will arrive or how it will change the world by reading academic papers is too slow, given the rapid pace of AI development I've just described. Reading papers means you're already behind.

My favorite way to keep up is by following the long-form interviews with tech leaders that are often released. You can find interviews that are three hours long, and they are posted almost immediately. I think reading these interviews is the best approach. For instance, if a major US tech media outlet interviews someone this morning, the transcript might be up the same day, or even in real-time. So, if you want to know how AI will change things, the best way is to follow the key figures. You could probably add Sam Altman to this group. Follow their keynote speeches at real-time conferences or their interviews. The interviews these days are very in-depth. I think using YouTube for this is a great feature, and it's the best way to understand the current landscape.

Multidimensional Competition in AI and the Reorganization of Academic Fields

Even these leaders' thoughts evolve slightly over time. But one thing they all consistently say is that this technology will arrive sooner than we think and its impact will be far greater than we imagine. There has been a constant stream of countless new services. For a long time, they were coming out of the US, then China started catching up, and after DeepMind's breakthroughs, we saw a flood of various models. It's a chaotic race. Currently, the competition in AI involves model competition, but from my personal perspective, there are two main focal points. One seems to be in Physical AI, focusing on how it can be applied in the real world. The other is Agentic AI. And in connection with science and technology, AI for Science is a major battleground. There are many papers like this one. You might wonder why I included this. It's a paper evaluating the use of Large Language Models (LLMs) for scientific discovery, and it has a huge number of authors. My point isn't about the number of authors. If you look at their affiliations, you'll see that every academic discipline is represented, from law to the humanities. So...

Most universities, including those in Korea, are already changing. AI has become a fundamental literacy, a basic knowledge that is required in all fields. People in linguistics need to be able to use and evaluate AI. Lately, it seems a lot of our education is focused on how to use these tools quickly and effectively. Using them well is important, very important. But it's what social scientists can do well to add AI to their existing knowledge, to critically examine AI in depth. This is what we often refer to as 'AI + X'—using AI to transform existing industries.

But conversely, 'X + AI' is just as important: viewing AI from the perspective of one's own domain knowledge. If I am a philosopher or a political scientist, how should I view the emergence of AI from that standpoint? You shouldn't just be creating an AI agent to diligently collect articles related to AI and political science. The really important discussion is about what this emergence means and what changes it will bring to the existing order. The tool training part is something you can do on your own, through a boot camp, or a short-term course. Most importantly, the best way to learn is to just use it a lot yourself, in a brute-force way.

The perspective of viewing AI from within one's own domain knowledge is equally important. For example, the question is how a philosopher or a political scientist should view the emergence of AI from their own disciplinary standpoint. It is not enough to simply create an AI agent to collect relevant articles. The discussion about what the emergence of AI signifies and what changes it will bring is crucial. This kind of discussion can be pursued independently, or through boot camps and short-term courses. What is most important is to use the technology extensively yourself.

AI and Innovation in Scientific Research: The US-China Rivalry

Therefore, I want to emphasize that you don't need to focus solely on that. The country that excels most at scientific research and innovation is, of course, the United States. Recently, the US announced something called the Genesis Mission. The US is the country that has produced the most Nobel laureates and world-changing innovations. Given that it is already so far ahead, why is it placing even more importance on AI? This is related to the US-China competition. It seems that even the US, the leader, feels threatened that if it doesn't further enhance its scientific capabilities using AI, China could catch up.

Furthermore, the US likely determined that scientific research is an area where it can overwhelmingly reinforce its current lead and further widen the gap with China. What is most important in science? While many scientists are needed, exceptionally talented individuals are even more crucial. You need a lot of outstanding people, and even more importantly, you need a vast accumulation of scientific data. In both of these aspects, the United States is the frontrunner. The US has accumulated modernized scientific data for the longest time, and the amount of data it holds without making it public is enormous.

Over 95% of it has not been released. There is also data that is freely available. In fields like space exploration, a lot of data is public. Where are the world's most brilliant scientists concentrated? In the United States. What about China? It's a bit behind the US, but it has a very large number of scientists and is diligently accumulating data. However, there is still a significant gap with the US. The one thing the US lacks is a sufficient number of junior researchers or fellow researchers to work alongside its top scientists. AI can supplement this. It's like a single professor forming a research group with multiple PhD students or postdoctoral researchers.

I suspect they are pursuing this with the idea that it can become a massive laboratory. This is an example from the Korea Institute of Science and Technology (KIST)'s autonomous lab. Concepts like this are already emerging frequently. It varies by academic field, but especially in experimental fields, a problem at night can be a disaster. For biological experiments, someone probably needs to be on watch. PhD students might take turns, but soon, this may no longer require a human. Physical labor and simple repetitive tasks have already been largely replaced. Beyond that, AI is already heavily involved in formulating new hypotheses and undertaking creative and challenging endeavors. It is producing results superior to those of many human scientists.

Job Shock and Deepening Polarization in the AI Era

This field is already quite advanced. While AI shows enormous potential, especially in science, it also carries immense risks. The most frequently mentioned issue is jobs. According to our surveys, there is a lot of talk about misinformation and military applications. These are significant issues, especially among the younger generation, but the overwhelming concern is about jobs. The biggest fear is that jobs will be lost because of AI. And in fact, a significant number of jobs are already disappearing in the United States.

There are concerns that all jobs except those of the top 0.01% could disappear. Another important point is that it's the jobs of the younger generation—the juniors—that are vanishing. For the current establishment, the mid-level and senior-level professionals, AI is an excellent tool. They can leverage it. But for those who need to find a job after graduation, AI is a competitor. It will be difficult to get hired unless you can prove you are significantly better than an AI. Running a few AI agents might be sufficient for the company.

This is a very serious dilemma. We joke about it, but some say AGI (Artificial General Intelligence) will arrive in a few years, with many expecting it within five. We talk about how to survive in that era and what changes will occur. While getting a simple job has become harder, the possibility of becoming a business owner right away has increased. This is because you can do things immediately if you use AI well. Although there's a lot of negative news, there are also many cases where AI makes previously impossible things possible. For example, there are many instances of people who don't speak a word of Arabic and have never been to an Arab country successfully selling products there with the help of an AI agent. People are using AI and agents very effectively to solve problems that were unimaginable in the past.

That's why we're seeing the emergence of many 'one-person unicorn' companies. A unicorn is a privately held company with a valuation of over 1 trillion won. While revenue isn't guaranteed, it has to show some solid results. Many one-person unicorns are appearing. There are many cases where two people generate 100 billion won in revenue, or where a company founded by two people reaches a valuation of 2 trillion or even 10 trillion won. A common feature of these companies is that their number of employees is small relative to their valuation or impact.

In the case of DeepMind, although the person who first developed the model was Chinese, it was developed by 150 people. 150 people is less than half the size of a typical national research institute in Korea. Many national research institutes have thousands of employees. A single small lab created this. If it were to go public, its valuation would be in the tens oftrillions of won. The world is changing this way. The problem, however, is that this is only possible for those who have a certain basic foundation. The opportunity comes to those who can afford a subscription or have the basic knowledge to recognize the potential in AI.

You can't do much with AI if you're in a place without electricity. The IMF conducted a study with a report on this a few years ago. As I mentioned, AI changes month by month, so the situation might be different now, but the general trend is as follows. First, high-income or highly specialized professions have a higher risk of exposure. Due to AGI (Artificial General Intelligence), their exposure risk is bound to be high. Doctors, lawyers, and others are highly exposed. Why is that?

From an operator's perspective, the desire to replace high-cost personnel is strongest. At the same time, if these professionals use AI effectively, they also have a high potential to secure new opportunities. A doctor who uses AI well or a lawyer who uses AI well is likely to earn much more money or create greater value. For developing countries or low-skilled blue-collar occupations, the likelihood of being replaced is low, and the chance of seizing new opportunities is also low. This trend applies to both developed and developing countries, and it's similar globally in terms of gender. Women have lower exposure and also fewer opportunities. The same pattern holds for education levels. The biggest issue is that existing gaps—between developed and developing nations, between the highly educated and the less educated—are likely to widen or become entrenched. The impact also varies greatly by job category. This is US data shared by a professor with a researcher on GitHub. They linked all US employment data to create a dashboard. Red indicates higher exposure. AI exposure differs for each job category. I will share this material, so I encourage you to go in and look at it yourself; it's very interesting. Please click the link to visit the original site. This is also a very serious issue. Currently, physical or blue-collar jobs are less affected, but all repetitive labor will be replaced. Robots can also replace all repetitive labor.

People say it's still a long way off, but in countries that utilize AI well, like China, much of this has already happened. Repetitive labor, very precise labor, is almost all being replaced. High-end labor like surgery is a different matter, but you can assume that easy, repetitive tasks will almost all be replaced. Anthropic has also released a report based on its own data analysis. The most important finding here is the job shock for juniors. Junior positions have decreased. While many other things are still uncertain, the decline in junior jobs is clear. LinkedIn also announces annually which job sectors are growing and which are shrinking. OpenAI also releases usage data on what people do with AI, and some researchers write papers based on this. The reason I'm showing you this series of data is that in addition to jobs, the change in tasks is also significant. If I break down my job into tasks, there would be education, research, administration, external activities, and so on. Research can be broken down further: reading papers, conducting a literature review, collecting and analyzing data, and then deriving results and submitting them. If you break it down even more, some of these tasks can be automated with AI. The submission process can be fully automated, and final proofreading for typos is already largely automated. Literature reviews can be further assisted by AI, and as I mentioned earlier, hypothesis generation can also be done through AI. Therefore, the idea that tasks are being reconfigured is the most widely accepted view these days. The role of a doctor is the same.

The role of a doctor can be to collaborate with AI to see far more patients and make more accurate diagnoses. If you do this reconfiguration well, productivity increases significantly. However, this is also a somewhat idealistic story. The ideal hypothesis we've held is that if AI alone has an accuracy of 90% and a human alone has an accuracy of 95%, then a human and AI working together will achieve 100%. Therefore, the long-dominant ideal has been that we must coexist and thrive with AI. But this isn't true for all tasks.

In many cases, accuracy improves when the human steps aside. There are cases where AI alone scores 150, a human alone scores 100, and a human and AI together score around 95. A lot of research shows that clumsy human intervention can actually create noise. We need to see that it depends on the task. From a corporate perspective, since most of you are preparing for employment, the best approach is to analyze job postings. You need to understand what kind of people companies are looking for.

Practical Challenges of AI Adoption and Organizational Design

There's AI. These days, you'll probably see many job postings looking for people with an understanding of AI or proficiency in using it. In fact, last year, MIT conducted a survey on the business use of AI and concluded that adoption is not going as well as expected. If that's the case in the US, the situation in Korea is likely more severe. The reason is obvious. Installing AI isn't like buying a washing machine and having your chores magically taken care of. You have to plug it in and know how to use it well. In the case of AI, the 'electricity' would be data. In many cases, the data itself hasn't even been collected. Therefore, adopting AI often yields no effect. To significantly increase productivity, as I mentioned earlier, tasks need to be broken down effectively. The organizational design must already be tailored for AI, allowing for the recombination of these tasks to decide 'this will be done by AI, and this part will be done by humans.' Simply introducing AI into an existing organizational structure with the same people is virtually impossible. These days, top overseas universities, especially business schools, teach subjects like 'Agent Management.' This is analogous to how traditional business administration teaches human resources and organizational management, learning how to manage people to achieve results.

This has all been switched to AI, to agents. Because agents are our colleagues and competitors, and we must utilize them as a theory. It's about dividing tasks: how will we collaborate, which tasks will be handled by AI, and which are unreliable and must be verified by humans? This kind of preliminary preparation takes quite a long time. That's why these initiatives often don't work well. The basic story, as of last year, is that it has not been successful. The order of discussion was individuals, companies, and then universities.

The Failure of University Education and Corporate Talent Selection

So, what role should universities play? Are you familiar with the company Palantir? The Palantir group created the Meritocracy Fellowship. If you visit their website, it's very interesting. The most intriguing phrase is at the top: 'Skip the degree.' What this means is that the glamorous tech and AI companies of Silicon Valley are saying that higher education in the US has failed. In other words, instead of going into student loan debt for a higher education degree, you should get a job right away. This company isn't an AI company, but it's a very innovative software company that increases corporate productivity through ontology. This company focuses on hiring exceptional talent.

It's completely impossible. To significantly increase productivity, as I mentioned, tasks need to be broken down effectively. The organizational design, which recombines these tasks to say 'this will be handled by AI, and this part by humans,' must be well-suited for AI in advance. Introducing AI into an existing organizational structure is virtually impossible. That's why top overseas universities, especially business schools, are now teaching subjects like 'Agent Management.' This is just like how traditional business administration deals with human resources and organizational management to understand how to manage people for performance.

This is all being replaced with AI agents. Because agents are also our colleagues, competitors, and a theory we must utilize. It's about deciding how to divide tasks, how to collaborate, which tasks AI will handle, and which are unreliable and need human verification. This preliminary preparation takes quite a long time. Therefore, these attempts have often been unsuccessful, and the basic assessment as of last year is that they were not as successful as expected. Next, let's look at individuals, companies, and universities.

So, what kind of place should a university be? Are you familiar with the well-known American company Palantir? The Palantir group created the 'Meritocracy Fellowship.' If you visit their website, it's very interesting. The most intriguing phrase at the top is 'Skip the degree.' This is a statement from the glamorous tech and AI companies of Silicon Valley, declaring that higher education in the United States has failed. In other words, instead of going into student loan debt for a higher education degree, you should get a job right away. This company isn't an AI company, but it's a very innovative software company that deals with ontology to enhance corporate productivity. This company focuses on selecting exceptional talent.

Then, they are immediately put to work in the field. It is far more effective to teach them while they are working within the company. This may be a particularly serious problem in Korea, but not exclusively so: university education is rigid, and a four-year university education does not necessarily translate into competence at work. Graduating from a good university or a good department doesn't guarantee you'll be good at your job. Students who get into good universities generally have strong academic abilities, so they tend to do well at whatever they are tasked with. Therefore, if it can be proven that someone is capable of getting into a good university or a good position, it would be much more efficient to hire them directly and have them start practical work early.

Everyone recognized the potential to save time, but they were the ones who made it a reality. More importantly, companies themselves set the selection criteria. South Korea's second problem is that in the age of AI, it still selects future talent through the outdated College Scholastic Ability Test (CSAT) system. This is like trying to race a car on foot.

Companies are creating their own methods for talent selection, picking out outstanding individuals and having them solve real-world problems. The ultimate goal of a university education is not to stay in university forever but to move on to the next stage. Many students these days want to work for companies. And what do you do at a company? Ultimately, there are customers, and you have to solve their problems. This program starts that process from the age of 19. By identifying who the customers are, what the AI-related problems are, and getting hands-on experience, they can learn to work much faster and more effectively. Cases like this are increasing, and places like Minerva University have attempted similar things. Even those with PhDs in AI are trending toward companies rather than academia, which is a bigger shock than you might think.

More high-quality papers are now coming out of companies. Ideally, Nobel Prizes should come from places like MIT or Stanford, but now they are also coming from Google. Recently, we've seen a paradoxical situation where Stanford professors move to OpenAI, or brilliant professors from Harvard go to OpenAI to conduct research, causing the existing order to collapse. In response, prestigious American universities are also making major adjustments, such as laying off existing professors or merging departments. We are also seeing a trend toward more granular academic programs.

Issues like academic dishonesty are already rampant in schools. There are programs like microdegrees, but there's a growing trend of breaking them down even further by function, allowing students to graduate after completing only specific courses. You pay tuition per semester and take only the courses you need. This enables a lifelong learning model, where people can take specific courses as needed while working, or do both concurrently. As you can see, a major tectonic shift is occurring in all areas, including universities, companies, and among researchers. Research at universities is not being conducted properly, talent is moving to companies, and the likelihood of students bypassing university altogether is increasing. If universities do not change, they could lose their reason for existence in an instant.

AI Academic Competitiveness: The US-China Gap and Geopolitical Conflict

At NeurIPS, the world's top-tier conference on artificial intelligence, China accounts for about half of all published papers. The top institution is Google (4.8%), and the second is Tsinghua University. Many other Chinese universities follow. South Korea, including KAIST, holds a share of about 2-3%.

The US accounts for about 30%, China for about 50%, and Singapore for about 5%, with the rest of the world making up the remainder. This can be seen as an indicator of AI competitiveness. A similar trend is seen at top conferences in other science and technology fields, such as materials engineering. This year at NeurIPS, the organizers announced that, for geopolitical reasons, they would not accept papers from Chinese universities on the US government's sanctions list. In response, China threatened to boycott the submission of papers from all its universities. A few days later, the conference reversed its decision. This shows the escalating geopolitical conflict even within the academic sphere.

Geopolitics has intervened in academia. When the US decided not to accept papers from universities it deemed risky, China retaliated by threatening to withdraw all its universities' participation. This sparked arguments that it violated the fundamental spirit of academia, countered by the claim that since the research is based on open data, there should be no issue. This conflict is intensifying. Nevertheless, when measured by the number of influential papers that have changed the world, the United States is still ahead.

China can be seen as a strong second. In terms of the total number of papers, there are fields where the US is number one, but in the top 0.01% or 0.05% of papers, China is increasingly taking the top spot. In the patent sector, Chinese companies are filing a massive number of generative AI-related patents, occupying 7 to 8 of the top 10 spots. In the past, Chinese companies were passive about filing international patents, but now they are actively seeking to protect their core technologies. This is an effect of the US-China competition, as they now recognize that securing patents is crucial for freedom in research, development, and innovation.

This is also part of my personal research area. A characteristic of Chinese papers is as follows: as various metrics show, the number of Chinese papers is very high. In particular, as you can see from the graph, the proportion of 'China-only publications'—research conducted exclusively by Chinese researchers—has been steadily increasing. While research from other countries often involves collaboration among researchers from various nations, China has a high proportion of domestic-only research. These papers are being published frequently in top international academic journals.

Recently, China's trend toward knowledge self-reliance has become very clear. Many of the top papers in the AI field are written by researchers in China or by Chinese researchers in the United States. If you include Chinese researchers affiliated with American universities, that network expands even further.

Open Source Strategy and U.S.-China Competition in the Robotics Industry

In the U.S.-China competition, open source is a critical factor in the AI sector. Chinese models are becoming widespread through open-source distribution. Their significance stems from their high performance, low cost, and high accessibility. Some analyses from last year suggest that Chinese models have already surpassed the performance of their American counterparts. Even if U.S. models perform better, they are expensive, and Chinese models can replace nearly all the functions of American-made models. This trend is now extending to the robotics sector. The software and data used in humanoid robots are also being released as open source, making them readily available for researchers.

While AI is a competition of models, in the field of physical AI—that is, humanoid robots—physical embodiment is crucial. There is a need for robots that can perform work or provide entertainment in place of humans. In the U.S.-China AI competition, robotics is a particular area of concern for the United States. As of last year, China accounted for 85% of global humanoid robot shipments, and this figure is expected to reach 90% this year. The output of a single Chinese company is greater than the entire production of South Korea. When a single country captures over 90% of a specific industry, it is as if that country is effectively creating the industry itself. Even in the semiconductor sector, where South Korea has a strong advantage, Samsung and Hynix hold a 60-70% market share. China's dominance in robotics is even greater.

China's Rise in the Competition for AI Talent and Standardization

If this situation becomes entrenched, it could pose a significant problem. Securing talent is also crucial in the AI competition. In the past, many of China's top talents studied in the United States and either returned to China or remained in the U.S. However, recent analysis shows that while China ranked fourth in the world for the number of top-tier AI researchers about a decade ago, it has now risen to first place. By 2025, China is expected to have the largest number of top AI researchers, followed by the United States and South Korea. While this is positive news for Korea, Singapore stands out even more on a per capita basis. Despite having a much smaller population than South Korea, Singapore has a higher number of researchers relative to its population. To compete with China and the U.S., countries ranked third to eighth need to scale up. Furthermore, China is characterized by a high proportion of young researchers, indicating significant growth potential.

A high proportion of young researchers signifies greater potential for career development, which can in turn strengthen AI competitiveness. In terms of standardization, China has also secured considerable influence in major standards organizations. Its voice is growing in international bodies like the UN, which shows that China has long invested effort in shaping the global order not just through AI technology but also through standards and norms. This indicates a multifaceted effort rather than simply a result of capital or manpower investment.

When ChatGPT was first released, some assessed that the gap with the U.S. was so large that the game was over. However, as subsequent metrics have been updated, the gap is narrowing. New models continue to emerge, with the U.S. and China competing and pushing ahead of each other. It is a fact that, apart from these two nations, South Korea is one of the few countries showing prominence.

South Korea's AI Competitiveness: The Importance of Ecosystem Building

However, the critical question for South Korea is whether it can build a service capability or ecosystem that can rival those of the U.S. and China. A model by itself is meaningless; it requires a supporting ecosystem of robots, home appliances, peripheral technologies, and data. Therefore, we will need to watch this competition over ecosystems going forward.

It is relatively accurate to say that AI competitiveness will increase because there is much greater potential for career advancement. The proportion of young people is very high. This is an important factor when assessing the current AI competition and objectively predicting the future. We also look at areas like standardization, where China has already invested considerable effort to dominate large segments of major global standards organizations. This is worth noting. The reason China has a greater voice and seems more influential at the UN is not just because the UN is not based on a one-person-one-vote system and China has many partner countries and friendly forces, but also because there are many entities like the United Nations University (UNU) within China. Thus, China is very active within the UN as well.

All of this is the result of a long-term effort to shape the global order and its rules not only through AI technology but also by embedding it in standards and norms. It is more than just the result of pouring in money or manpower; other indicators are also very aggressive. When ChatGPT first came out, the gap was so large that many thought the game was over. But with each update of subsequent metrics, the competition appears increasingly neck-and-neck, or even when a gap opens, China quickly catches up, and new models are constantly being released. The U.S. and China are in a structure of mutual competition, constantly pushing ahead. It is a fact that among countries other than the U.S. and China, South Korea is one of the few showing prominence.

However, the crucial question for us is whether we can build a service capability or ecosystem strong enough to truly replace the U.S. and China. Simply possessing a model is meaningless. It must be supported by a comprehensive ecosystem—that is, robots, home appliances, and other peripheral technologies or data on which agents can be built. Ultimately, it is a battle of ecosystems, so this is what we need to watch going forward.

Shift in U.S. AI Policy and Corporate Interests

This is something to consider. As I mentioned earlier, U.S. policy has completely shifted from a focus on safety to one of security. One of the first things the Trump administration did was to abolish DEI, or Diversity, Equity, and Inclusion. The rationale was that it hinders AI development and causes delays, so the priority should be innovation now, with discussions to follow later. The focus on safety has also diminished. The current consensus is to concentrate on the proliferation of American AI, and companies like OpenAI and cloud providers are aligning with this.

In the early 2000s, especially after China joined the WTO in 2001, there was a period in the late 2000s when U.S.-China relations were poor. At that time, the U.S. designated China as a currency manipulator and sought to revoke its most-favored-nation status. There were arguments that the U.S. should cease educational exchanges with China, labeling it a problematic country that did not adhere to market economy principles.

The U.S. was reluctant to recognize China as a normal country. At that time, some voices argued for engaging with China, selling to China, and utilizing the Chinese market. These were American corporations. Companies like General Motors and Boeing were generating, and expected to continue generating, enormous revenues in China. General Motors, for instance, was once on the brink of bankruptcy but was revived by entering and securing a share of the Chinese market. Recently, however, announcements from these companies often sound as if they were written by OpenAI or the U.S. State or Defense Departments. They argue that the AI of the free world must win and that authoritarian states must not be allowed to take the lead. Why this sudden hardline stance?

Is it because these companies are burning with a sense of justice? That could be part of it, but I view it from a business perspective. The reality is that these companies cannot generate profits in China anyway. China has developed its own AI models, and there are alternatives like Deepside and Qumi. The markets where these American companies hold a high share are outside of China. The problem is that Chinese models like Alibaba, Deepside, and those from cloud providers are spreading beyond China. They are dominating Southeast Asia and are being used in South Korea and even in the United States.

Decoupling of AI Ecosystems and U.S.-China Interdependence

The same is true in Silicon Valley. Therefore, Chinese models must be confined within China. This can be seen as a case where government interests and corporate interests have aligned. Not only the U.S. but also China has established an AI action plan and announced an AI cooperation organization this year. A competition is unfolding between the two countries that are best at AI and have the most well-developed ecosystems over how to proliferate their technologies and check each other. The U.S. is on the offensive.

As a superpower, the U.S. is using more offensive measures, while China is on the defensive but continues to push for proliferation. Why does China use an open-source strategy? Looking at past IT competitions, such as the Windows operating system rivalry or the Android vs. Apple competition, latecomers often adopt open-source strategies. It is a strategy to attract and lock in many users. China is using open source because it is a latecomer and the number two player. Comparing its national power and circumstances, China is compelled to adopt a defensive strategy. However, as open source spreads, the U.S. strategy is not working as well as expected.

With this level of pressure, one would expect a collapse, but that is often not the case. The AI ecosystem has now become decoupled and entrenched. From an AI-centric perspective, the fundamental requirements are data, electricity, computing power, AI models, AI services, and devices. However, looking at U.S.-China relations, all these links have been severed. The U.S. and China do not share core data with each other, and the same is true for China. The decoupling continues, with restrictions on the export of chips needed for computing power to China and limits on the import of Chinese products into the U.S. As they block each other, there are areas where each side has a deficit in a completely separated state. China's biggest weakness is semiconductors. Therefore, it is trying to achieve self-reliance in the areas it lacks. China's biggest problem is semiconductors, and it is pushing for self-sufficiency in key components like GPUs. The U.S., on the other hand,

has many shortcomings in the hardware sector. The U.S. seeks to address this by collaborating with allies like South Korea and Japan. However, some assess that U.S. alliance cooperation is weaker than in the past, making its policies less than ideal. In contrast, China's semiconductor self-reliance is progressing better than expected, but it also faces clear limitations. Can the U.S. and China reconcile and return to the past?

That will be difficult for the foreseeable future. This is demonstrated by the fact that even under the current administration, the U.S. State Department is proposing various programs. In forums like the 'Pax Sillica' initiative, new consultative bodies are being formed with Indo-Pacific allies in the semiconductor and key technology sectors, excluding China. A prime example of these programs is the provision of funding to countries or companies that can produce and supply trusted smart devices in Southeast Asia. This is because Chinese hardware and software are widely used in the region. China is geographically close, provides significant support, and from the perspective of Southeast Asian countries, there is little reason not to use Chinese products, as they are cheap and effective. However, there are many projects to create alternatives to Chinese products in this region. South Korea could be a leading candidate, as could countries with electric vehicle producers like Vietnam's Vingroup. As this situation becomes entrenched, the conversation will shift from being just about the U.S. and China to include South Korea, and this raises a concern.

South Korea's Strategy for AI Sovereignty and Strengthening Interdependence

Given the U.S.-China tech competition mentioned earlier, must we choose one side or try to maintain good relations with both indefinitely? The natural thought that follows is, 'How long can we live like this?' Shouldn't we have something of our own to some extent to secure negotiating power and leverage? Couldn't we even think, 'I'll fend for myself'? This is because we cannot leverage China against the U.S., and leveraging the U.S. against China is too great a gamble.

The idea is that we too have something of our own. This is what is commonly referred to as 'AI sovereignty' these days. The case of Europe shows a situation similar to ours, but in a worse position. Europe relies on China for all its hardware, such as batteries and electric vehicle components, and on the United States for all its software. Europe has no domestic services comparable to Naver or KakaoTalk; everyone uses American services. Relying on Chinese hardware and American software means it has no leverage of its own. Thus, there is a growing sense that this cannot continue. Being structurally dependent on one side is risky, and they must create something themselves. Relying on a single company or country must be avoided. The most ideal solution would be to create their own 'Made in Europe,' but this will be difficult without an existing ecosystem.

So, what should be done? The answer is not to depend on just one entity. For example, when creating an OS, instead of relying solely on Google, one could use Microsoft's or South Korea's Naver's offerings as well, creating a structure that diversifies risk. This structure is widely used not only in the AI and digital sectors but also in other areas like food security. Singapore is a prime example. Singapore does not produce its own food but diversifies its imports from neighboring countries like Thailand and Malaysia. If a problem arises in one country, it can source from another. It also employs a two-track strategy of investing in its own development, even if it takes time. This is being commonly applied to AI as well. The term 'sovereign AI' can have exclusive connotations in a political and diplomatic sense, so one must be cautious when discussing it with foreigners. It can provoke a negative reaction, such as, 'Are you trying to exclude the world and treat this like territorial sovereignty?' Therefore, explaining that it is an 'open and cooperative sovereign AI' is necessary to enable dialogue. The U.S. has proposed, and especially now,

Michael Kratsios, the White House director leading U.S. AI policy, says, 'AI sovereignty simply means having access to the best model in the world.' Since the U.S. has the best model, possessing the right to access it is better for your sovereignty. What's important in sovereignty is performance, and developing it yourself is inefficient and a waste of national power. So, use America's, and instead, focus on other areas where you can excel. This is a different concept from traditional economic security or diversification. It's not about diversifying investment to avoid concentrating on one thing, but about becoming fully dependent on the best player while making them fully dependent on you. So, what should we do? Instead of pouring money into developing a sovereign AI model, we should focus on semiconductors to maximize mutual dependence. This creates deterrence. Since our collapse would mean their collapse, mutual trust is strengthened. This is the concept proposed by the U.S., and it is a very meaningful one.

Conceptual Differences in AI Sovereignty Strategies and South Korea's Choice

However, choosing which path to take is not easy. The concepts are currently divided in this way. The concept offered by the U.S. has merit, and recently, U.S. think tanks have been criticizing the sovereignty strategies of major middle powers, arguing that they are inefficient. They contend that since the U.S. cannot do everything, it is better for these countries to focus on complementary roles in the U.S. AI innovation ecosystem, such as robotics and hardware. The argument is, why not just strengthen mutual complementarity? What does China have to say?

China's context is similar. 'Why engage in such an inefficient endeavor when we are going to release something as good as the U.S. for free?' From China's perspective, it's like saying, 'Don't try to build Excel from scratch; we'll provide Excel and PowerPoint programs for you to use, so you can focus on what you do well.' The problem, however, is the concern about how long it will remain free. This situation is evolving with these added complexities. Nevertheless, while that concept—especially the one proposed by the U.S.—has merit and alliances are important, the reason we still need to try to do something on our own is as follows.

When choosing not to produce something, you can diversify. There is a diversification method where you spread your sources across multiple countries like Thailand and Malaysia, so if something happens in one country, you can still get it from another. Additionally, there's the approach of investing in your own development, even if it takes time. It's a two-track approach. This is what has been commonly applied to AI. The term we use, 'sovereign AI,' has strong exclusive connotations in a political and diplomatic sense, which can be met with resistance from foreigners. So, it's essential to explain. We must clarify that the sovereign AI we pursue is not exclusive but open and cooperative to move the conversation forward. What the U.S. has proposed, and particularly what Michael Kratsios, the director leading AI policy at the White House, has said, is as follows.

Michael Kratsios, the White House director leading U.S. AI policy, put it this way: 'I understand. I know what you are trying to do with sovereign AI, but what is sovereign AI? What is AI sovereignty? It is enough to have access to the best model in the world. Doesn't the U.S. have the best model in the world? Then, possessing the right to access it at any time and ensuring that access is guaranteed is better for your sovereignty.'

This is because what matters for sovereignty is performance. Performance. Trying to do something you can't do as well, inefficiently pouring money into it, and ultimately creating a subpar model is a waste of national power. So, use ours. Instead, use that money to do what you can do better. This is a different concept from the diversification we traditionally talk about in economic security or sovereignty. That's about diversifying investments because you shouldn't put all your eggs in one basket. This is different; it's about becoming fully dependent on the country that does it best, but in turn, making that country fully dependent on us. So, what should we do? We don't need to spend money on developing a sovereign AI model; we should go all-in on semiconductors. By doing so, we maximize mutual dependence to the fullest extent. What happens then is that it creates deterrence. Deterrence. If we go down, you go down too, so mutual trust is somewhat strengthened. This is the concept proposed by the U.S. And it is, in fact, a very meaningful concept.

However, choosing what to do and how to do it is not easy. It's not easy, but the concepts are currently divided like this. The concept offered by the U.S. has a valid point, and this is where several U.S. think tanks have recently been coolly criticizing the sovereign strategies developed by major middle powers. They say it's impossible. It's inefficient, and since the U.S. is not a country that can do everything, it's better to focus on something complementary that can play an important role in America's AI innovation. If it's robots, then robots; if it's hardware, then hardware. Why not just strengthen mutual complementarity? This is a common opinion. What does China say?

China's context is similar. 'We are developing something no less impressive than the U.S. and will release it all for free, so why are you doing something so inefficient?' This is China's position. 'We are going to create and release Excel for free, and we're going to create and release the internet, so why are you trying to build Excel from scratch now? We'll give you the PowerPoint program and everything, so just take it and use it, and do what you do best.' That's China's context. The problem is our concern about how long it will remain free. So, this is what's being added as things progress. Nevertheless, while that concept has merit, and especially the concept from the U.S. has merit, and alliances are important and all that is good, the reason why we still need to try to do something on our own, in my view, is this.

AI's Contribution to National Security and Industrial Growth, and the Need for Indigenous Development

AI is an essential element for a nation's defense and industrial growth. A country must be able to defend itself, sustain its economy through industrial growth, and build a social safety net and basic infrastructure. AI acts as the common denominator for all these areas. In the future, all infrastructure will be AI-based, and industries that do not utilize AI will struggle to grow. In the defense sector, AI-powered drones, robots, and other systems are already being introduced.

The importance of AI can be found not only in its contribution to national security and industrial growth but also in the logic that those with direct experience can better understand and utilize it. Just as someone who has done something performs better than someone who has only learned it from a book, the experience of developing AI firsthand not only creates a leverage effect but also strengthens capabilities in areas where one is already proficient. For example, when developing semiconductors for AI, a country with such experience would not rely solely on foreign sources for the AI-related aspects.

By deeply understanding and developing AI technology, we can create far superior existing semiconductor technologies. Our understanding and absorption capacity also increase when adopting external technologies. From this perspective, the government's position, with which I tend to agree, is that it is absolutely necessary to exert a certain level of effort in indigenous AI development. Over the past year, major countries have announced numerous AI-related policies.

AI Strategies of Middle Powers and South Korea's Potential

Every nation is striving to develop AI technology, and numerous middle powers are no exception. The AI strategies of middle powers are a topic worthy of research. Currently, AI technology levels can be divided into Tier 1 (the United States, China) and Tier 2 (South Korea, France, Germany, Japan, etc.). Within the Tier 2 group, there are countries like South Korea and France that aim to develop AI technology autonomously, and others like the United Kingdom and Japan that have decided to follow the U.S. strategy. Even these latter countries do not start their AI development entirely from scratch.

Japan is closely aligned with U.S. AI platforms, while South Korea plays the role of a hardware provider. Nevertheless, even Japan undertakes some indigenous AI development projects in areas like public administration and defense. South Korea is attempting to develop AI technology even more proactively. Beyond semiconductors, South Korea has companies with potential in the AI ecosystem. While it may not be the best in every sector, there are areas where it has world-class potential. In particular, South Korea holds a strong position in advanced manufacturing—in home appliances, automobiles, shipbuilding, and batteries—second only to China, and AI can connect these fields.

In the energy sector, it possesses diverse technologies such as nuclear power, solar, and Energy Storage Systems (ESS), and also has its own portal and messenger services. If these elements are well connected, there is a possibility of success in AI technology development. The United States considers South Korea a top partner in AI, especially in the field of physical AI. It would take a long time for the U.S. to build its own factories and foster a home appliance industry, but South Korea already has everything needed, making it an advantageous partner for cooperation.

South Korea's Diplomatic and Industrial Strategy Amid US-China Competition

The United States views South Korea as a very important alliance partner due to the high degree of complementarity in their security and industrial portfolios. In this context, the crucial remaining challenge is how to define the relationship with China.

There is a significant competitive relationship with China, and intense competition is expected due to many overlapping portfolios. However, South Korea's strategy toward China cannot be the same as that of the United States. China is a crucial partner, considering its geographical proximity and issues like North Korea. Furthermore, the appeal of China's vast market and ecosystem cannot be ignored. At a time when even U.S. companies maintain factories in China, South Korea needs to strategically consider how to configure its relationship with China.

Even as the United States and China compete fiercely, there are areas where they cooperate or engage. For example, cooperation is possible on global challenges or in the SDG sector. The direction for South Korea's global strategy could be to cooperate intensively with the U.S. in security-critical areas, while also collaborating with other countries, including China, in areas of low security sensitivity. Cooperation with China must be approached strategically and can be considered in less security-sensitive domains. This is a difficult task for South Korean companies, which are deeply contemplating how to approach and leverage the massive Chinese market and ecosystem, just an hour's flight away.

Strengthening South Korea's AI Competitiveness and Finding Ways for Mutual Coexistence

While generating profits in the Chinese market would be ideal, if that is not feasible, an approach of quickly identifying changes on the ground and using them for innovation is also important. On the first anniversary of the launch of South Korea's National AI Strategy Committee, an announcement of its achievements is scheduled, and it is producing results through various efforts. The basic philosophy is that although South Korea was a latecomer in the AI field, it has potential, and because it lacks infrastructure and support, its competitiveness must be rapidly enhanced.

We must secure GPUs by collaborating with Nvidia and strengthen competitiveness by leveraging the strengths of Korean companies to cooperate with overseas big tech firms. OpenAI, cloud companies, and Google are all seeking to establish branches in South Korea. Simultaneously, we must foster the competitiveness of Korean companies like Naver, Kakao, and SKT on a two-track basis. Supporting their independent foundation model competition is part of this context. The goal is to promote competition and create success stories by combining global cooperation with the strengthening of our own capabilities.

The goal is for manufacturing companies like Samsung Electronics, Hyundai Heavy Industries, and LG Electronics to adopt AI to create success stories in reducing defect rates, improving productivity, and preventing safety accidents, and then to export these models to the world. Additionally, to address the polarization resulting from the AI transition, it is necessary to establish an institutional framework that includes providing education, creating a research environment, and implementing basic income. We are exploring ways to support those who are marginalized or harmed in the AI era and help them join the current.

I believe these efforts could become a unique initiative that only South Korea can undertake. Alongside strengthening industrial competitiveness, we are concurrently making efforts to support marginalized groups and those who may not benefit from or could be harmed by AI, helping them to advance together. These efforts represent an opportunity for South Korea to present a model case to the global community.

The website of the National AI Strategy Committee discloses a vast amount of data, including meeting minutes, annual budget lists, and lists of AI-related national budget projects for 2025 and 2026. By analyzing these materials, one can understand how South Korea is allocating its budget in the AI sector and which ministries are using how much of the budget. This information can be utilized for carrying out projects and other purposes.

After the data was released, we received criticism from civil society that the budget was concentrated on tech innovation companies rather than social protection. After analyzing all the documents, which amounted to 5,000-6,000 pages, we are taking a stance of accepting this criticism and reflecting on it.

Although we are implementing policies while gathering opinions from experts and citizens, shortcomings and limitations still exist. Policies often do not unfold as ideally planned and can have side effects, which is due to a lack of understanding of human factors or the difficulty of prediction. Policies created with good intentions often fall short of expectations, and this fundamental limitation also exists.

The Strategic Choices and Trade-offs in AI Technology Development

As a latecomer in AI compared to the U.S. and China, South Korea aims to build its competitiveness quickly. The two-track strategy of securing GPUs by collaborating with Nvidia while simultaneously strengthening the competitiveness of domestic companies seems reasonable. However, this leads to a dispersion of resources. By choosing Nvidia because there are no restrictions on GPU imports, we have missed the opportunity for domestic GPU development. China, facing GPU import restrictions, has focused on domestic production, which is resulting in a narrowing of the performance gap and the securing of a vast ecosystem.

This suggests not so much a policy failure as the fact that no policy is perfect and that strategic choices inevitably come with trade-offs. Additionally, there is an approach that prioritizes securing competitiveness through innovation first, and then sharing the fruits of that labor. In this process, the most crucial resource is ultimately time. Money and other resources are important, but for people living a finite life, time is the most precious resource.

The Resource of Time and Contingency Planning in the AI Era

Marginalized groups must endure a period of hardship, and a major problem is how to compensate for this time and how they can endure it amidst the AI divide. Furthermore, we face criticism that the U.S. and China will not stand still while South Korea makes its efforts, and questions about whether there is a contingency plan if the gap fails to narrow. This is a very realistic problem.

Some argue that a US-China duopoly in the AI market is undesirable and that middle powers should join forces. However, from Japan's perspective, cooperating directly with the United States may be more advantageous than collaborating with South Korea. Therefore, practical obstacles make solidarity among middle powers difficult to achieve.

Uncertainty and Potential Risks of AI Technological Advancement

Many assess that if you carefully weigh the benefits and costs, there is less for middle powers to gain from each other than one might think. This is the reality of cooperation among them. Without a well-designed framework, a country sees no reason to cooperate, thinking, 'Why should I work with you? I could just deal directly with the US, or if it's something you can do, maybe I can do it alone.' The same applies to countries like France. That concludes my points on AI, and I should move quickly through the remaining slides. We have discussed much about technological advancement, but a fundamental question remains: will the development of AI be without problems?

Will the innovation we are anticipating this time actually materialize? It might not, and we could face another 'AI winter.' The jobs issue, as I mentioned, also remains a concern. Although Kakao has officially denied it, the idea of not hiring for roles that AI can replace is entirely conceivable. Moreover, universities continue to raise the question of how to manage this dangerous technology. There has been a voluntary movement among researchers calling for a pause in development, arguing, 'It's advancing much faster than I thought, so let's stop for a moment to consider how to respond to catastrophic or uncontrollable problems before resuming.' Most recently, there have been increasing calls for immediate action from institutions like Stanford University.

Institutional Preparedness and Regulatory Discussions in the AI Era

In particular, because AI will require a major restructuring of the economy and labor, failing to establish new institutions could lead to a major crisis. If we wait until problems emerge before considering systems for taxation and distribution, the situation could become unmanageable. For instance, if it becomes impossible to even track how money is earned through AI, taxation will be impossible and redistribution will be difficult. This potential for unimaginable problems is another reason why there have been many calls for a pause.

MIT is managing a database of various risks that AI could cause, and there are active discussions about the red lines that AI should not cross. Although many such positive discussions are taking place, they are unfortunately often ignored due to a lack of strong regulation and accountability. AI is currently being used extensively in warfare, and weapons are the primary topic in discussions about AI risk. The call is to refrain from using AI in weapons development.

Using AI for the Public Good and Achieving the SDGs

Furthermore, longstanding agreements to refrain from using AI to spread disinformation and from developing self-improving AI are not being observed. The reality is that AI is being misused for weapons development, its own self-development, and spreading disinformation. This is a critical point. Another area that receives less attention is the SDGs. Achieving the SDGs by 2030 now seems nearly impossible. About six or seven years ago, an analysis was conducted on the positive and negative impacts of AI on each of the 17 SDG goals. AI can have a major positive impact, and collections of projects using AI to achieve the SDGs already exist.

There are nearly 200,000 such projects. With so many problems still to be solved in the world, we need to consider how to use AI for the public good. AI can be applied in diverse areas like gender, education, and inequality. In particular, developing countries face different types of challenges than we might imagine, and AI can achieve a great deal there.

AI and Environmental Issues: Power Consumption and Carbon Emissions

It would also be worthwhile to examine AI in the context of the future. Reports like uncertainty indices and global risk assessments indicate that environmental issues will become increasingly critical, and AI is closely linked to the environment. AI consumes a tremendous amount of electricity, and in the United States, data centers are already being treated as NIMBY (Not In My Back Yard) facilities due to public opposition. In Texas alone, 1,000 AI data centers are awaiting approval.

The Future of AI Technological Advancement and Comprehensive Considerations

This is because they lead to problems like rising electricity costs and increased carbon dioxide emissions. These issues are also very important, and I hope you will consider them comprehensively. You may also find this material useful for reference. This concludes my presentation. Thank you.