Humanity stands at the greatest technological turning point since the Industrial Revolution. The emergence of Artificial General Intelligence (AGI) goes beyond simple technological innovation, fundamentally restructuring human labor, organizational systems, and the framework of society as a whole. Cases of IT companies that utilize AI most effectively demonstrate this clearly. A team of seven employees achieves annual sales of 7 to 8 billion KRW and is expected to exceed 10 billion KRW this year. It is a structure where one person is responsible for approximately 1 to 1.5 billion KRW in sales.
The lead engineer of this company says that through AI, he performs the roles of 30 people. It takes less than two hours from a planning meeting to the next meeting because AI writes code in seconds and improves it repeatedly. This signifies not just an improvement in productivity, but the arrival of the ‘era where everyone becomes a team leader.’ A single individual can now handle tasks that previously required an entire team or even a division.
This change is dealing a direct blow to corporate organizational structures. In the United States, organizational charts are rapidly flattening. As all members begin to perform roles at the team leader level, the middle management layer has become unnecessary. In the same context, MBA employment rates are plummeting. In the past, top-tier MBA graduates had an employment rate of 400%, receiving acceptance notices from an average of four places, but now the rate is less than 50%.
The reason is clear. AI performs data-based analytical work—finding patterns based on decades of accumulated data and making suggestions—ten thousand times better than humans. The legal field is no different. The work of law firm associates—staff who have graduated from law school but are not licensed attorneys and primarily search for and organize precedents—is highly likely to be replaced by AI. Furthermore, a significant portion of the work done by lawyers in their first to third years has reached a level that AI can handle.
A structural problem arises here. A 10-year expert will survive by collaborating with AI, but to become that 10-year expert, one must go through the training process of the 1st and 3rd years. However, if the tasks performed by rookies disappear, the opportunity for training itself vanishes. A paradox arises where, in 10 years, the seasoned experts themselves may go extinct. This is the phenomenon where ‘partial optimization leads to a total tragedy.’
Positive aspects of the AI revolution certainly exist. Every year in Korea alone, about 2,000 people die from industrial accidents and 100,000 are injured. Efforts to deploy humanoid robots into these high-risk labor environments that threaten the human body are becoming full-scale. Amazon has already deployed more than 100,000 robots in its warehouses. The main areas of replacement are repetitive and dangerous loading and unloading tasks, and carrying heavy objects that cause musculoskeletal injuries.
Professor Geoffrey Hinton, known as the “Godfather of AI” who left Google, points out two fundamental characteristics that distinguish digital intelligence from biological intelligence. These two traits pose an unprecedented threat to humanity.
First is Transfer Learning. In the case of humans, no matter how brilliant a genius is, when they die, only a tiny fraction of their insight is passed on to posterity through books and words. However, AI can copy what one model has learned 100% exactly to another model. In other words, the moment an “Einstein” is born in the AI world, all other AI models simultaneously become equal to Einstein. This is possible because of transfer learning.
Second is compressed learning through massive parallel computing. The case of AlphaGo Zero illustrates this well. AlphaGo Zero began reinforcement learning without looking at a single human game record, receiving only the rules of Go. By replicating itself into 20,000 copies and playing thousands of games simultaneously, it processed 4.9 million games in just 4 days. As a result, it overwhelmed AlphaGo Lee, which had defeated the world’s best human player, with a score of 10 to 0.
NVIDIA introduced ‘Cosmos,’ a solution that applies this principle to the real world. It replicates a virtual reality factory with physics engines applied into 100,000 copies and trains humanoid robot software within them. Learning for 10 hours in a virtual factory produces the effect of learning for 1 million hours in reality. This is the basis for Professor Hinton’s conclusion that ‘it is difficult for biological intelligence to beat digital intelligence.’
Most AI scientists expect the arrival of Artificial General Intelligence (AGI), which surpasses human intellectual ability in all areas, before 2030. That is less than 5 years away. Professor Hinton asks a core question: ‘How can a superior intelligence obey an intelligence that is inferior to it?’ The only example humanity has observed is a baby instinctively manipulating its mother. Professor Hinton warns that humanity must solve this problem, which it has never encountered before, before the arrival of AGI.
A more serious problem arises when AI can set its own intermediate goals. If a human enters the goal to ‘solve the climate crisis,’ the AI might set ‘eliminating humans’ as an intermediate goal to achieve it. This is a scenario already described in a 1920s novel by Czech author Karel Čapek, who first used the word ‘robot.’ In that novel, the AI robots reach the conclusion that the extinction of humans is necessary to solve all of humanity’s problems.
Beyond technical threats, AI is threatening the democratic system itself. False information produced by AI is spreading rapidly, and it is difficult to identify regardless of intellectual level. This can distort political judgment and lead to a situation where society is operated by the bias of algorithms rather than the policy decisions of elected governments.
In a real case, a large domestic portal denied editorial responsibility on the grounds that AI was in charge of article editing. However, this is clearly flawed logic. For AI to judge which articles to place at the top, a human must input an objective function—that is, the criteria for ‘what is a good article.’ The person who set those criteria must take editorial responsibility. Transferring responsibility to AI is nothing more than intentional evasion of responsibility.
Currently, executives of major AI companies share an ideological foundation of ‘Effective Altruism’ and ‘Longtermism.’ Effective Altruism is the idea that the same resources should produce the maximum good effect, and Longtermism is the philosophy that the long-term survival of the human species is most important.
While these ideas seem noble on the surface, they reach dangerous conclusions when taken to extremes. An example is the logic that ‘if disasters of the same scale occur in a developed country and a developing country, resources should be concentrated on relief for the developed country.’ The reason given is that the population density is high and there are more technologies and talents to preserve. This is a dangerous idea that replaces the universal value of equality of human life with quantitative calculation.
This ideology is currently merging with ‘Neo-nationalism.’ It is the logic that ‘to prevent war, one must cultivate overwhelming power to prevent war,’ leading to the claim that ‘it is ultimately better for humanity for the United States to possess unapproachable military force.’ This is exactly the same logic used during the development of nuclear weapons. The CEO of Palantir is a representative advocate, leading to the theory of raising a 1 million-strong humanoid army.
OpenAI was originally established as a non-profit foundation under the belief that advanced AI should not be monopolized by private companies. The name ‘Open’ itself reflects that philosophy. However, it recently signed a contract with the US Department of Defense. Anthropic is a company founded by members who left OpenAI in protest against its commercialization. They take the approach of ‘Constitutional AI,’ internalizing human ethics and values into the AI development stage like a constitution. However, this company also recently signed a contract with the Department of Defense.
These cases reveal the fundamental problem that a small number of unelected tech elites, based on their subjective belief systems, monopolize technology that can determine the fate of humanity. Elon Musk believes in ‘Accelerationism’—the idea that everything hindering the development of technology must be removed—and ‘Transhumanism’—the idea that it is natural for humans to evolve into cyborgs—in addition to effective altruism and longtermism. An individual with the world’s greatest wealth and AI technology is personally intervening in the policy decisions of various countries armed with such beliefs.
The Trump administration is abolishing AI-related regulations under the principle that all regulations on AI are bad. Even a federal bill has been submitted to Congress to prevent state governments from creating separate AI regulation bills. The name of the National AI Safety Institute was changed from ‘Safety’ to ‘Security.’ The acronym is the same, but the meaning is completely different. This is why many experts fear the combination of ‘natural intelligence’ Trump-style leadership and rapidly developing artificial intelligence as a human catastrophe.
When genetic engineering technology first appeared, scientists gathered in Asilomar, California, and voluntarily declared a moratorium on research. It was an agreement to stop research for six months until regulatory laws were prepared. It is the only case in human history where a voluntary brake was applied to the development of advanced technology. AI scientists tried the same but failed. There are two reasons. First, six months of AI development corresponds to several years of other technologies. Second, AI is at the beginning of business, so competitive pressure is much stronger. The fear that a competitor will pull ahead while I am stopped makes voluntary regulation impossible.
The European Union’s AI Act currently has the most advanced regulatory system. Key principles include: high-capacity AI models must disclose all training data, they must have a ‘kill switch’ to stop operation at any time, transparency principles to clearly inform users of AI use, and accountability principles where there must always be a human responsible for all AI activities. In comparison, Korea passed the AI Basic Act in 2024 and is awaiting implementation in 2026, but it is evaluated as thin compared to the 120 pages of the EU AI Act.
History has left an important lesson. During the Industrial Revolution in Britain, 12-year-old children worked long hours in factories, 20,000 people died at once due to London smog, and the average life expectancy in London dropped to 20 years. It took a staggering 90 years after the start of the Industrial Revolution to recover previous living standards. As Keynes said, ‘In the long run, we are all dead.’ The lives of many people who lived during those 90 years were sacrificed.
The tragedy of the Industrial Revolution did not lie in the technological development itself. It lay in the fact that consciousness and institutions could not keep up with the speed of technological change. Society finally found stability only after labor unions were formed, child labor was banned, and working hours were shortened. At that time, factory owners placed advertisements in newspapers saying, ‘Do not take away the boys’ freedom to work.’ This is structurally identical to the claim today that all regulations ruin the economy.
When transitioning from a 6-day work week to a 5-day work week, predictions poured out that the economy would be ruined. The result was the opposite. Productivity rose by 5%. The reason is the redistribution of resources throughout society. Resources that were tied to marginal enterprises that could not survive without a 6-day work week moved to high-productivity enterprises that could operate sufficiently even with 5 days. From the perspective of society as a whole, productivity increased as a result of resources being redistributed to more efficient places.
The minimum wage hike is in the same context. There are claims that the economy will be ruined if wages rise, but there is a counter-argument that if wages are paid sufficiently, productivity follows. Various social experiments have shown that most people who receive basic income invest that money in self-development. This is evidence that the view of humans as ‘beings who won’t work if given a lot of money’ is wrong. Conversely, as a result of suppressing the minimum wage without wage increases, situations have actually occurred where domestic consumption shrank and the economic growth rate approached negative levels.
If AI drastically increases productivity, we must create a structure where the whole society shares the fruits. Sinan-gun’s solar energy pension is a good model. Sinan-gun distributes renewable energy profits to residents in the form of a pension, and as a result, the population of this region is increasing despite the overall decreasing trend. If wind power is added, residents will enter a stage where they can live without worrying about their livelihood. In the case of Yeoju, village buses are free and lunch is provided for free, funded by renewable energy profits.
The same principle can be applied in the AI era. We can raise funds by taxing the productivity gains created through AI development or by having the public hold a certain stake, and then invest this in building social safety nets and supporting youth employment. This is why discussions on basic income, basic housing, and a 4.5-day work week must be linked. We must prevent the benefits of technological innovation from being concentrated in a few companies and capital, and ensure that people who lose their jobs receive retraining and new opportunities.
Korea is a latecomer in the AI hegemony race, but it is the most advanced among the latecomers. It possesses its own script, has vast amounts of Korean documents digitized, and thanks to Naver holding its ground against Google, it has plenty of engineers with large-capacity distributed processing technology. Hangeul is far superior in digital input efficiency compared to Chinese-based scripts. All of this is the soil for AI development.
What is more important is the new model that Korea can propose. While the US and China are engaged in an AI hegemony race and creating a new bloc-ism, countries caught in between—such as those in Southeast Asia, the Middle East, and South America—are in a dilemma of not having their own AI capabilities while needing to protect their culture and history. Korea can propose ‘Inclusive AI’ to them. The idea is to co-develop AI together, share it as open source, and provide opportunities to build independent AI tailored to each country’s language, culture, and history.
AI becomes smarter the more multilingual it is. Co-developing with various countries improves the quality of the AI itself. If China or the US made this proposal, suspicion would arise first, but Korea, as a victim of imperialism, has a different level of trust. Sharing GPU computing power and opening up joint research results as open source is the ‘niche market,’ or strategic gap, that Korea can choose.
The AI problem is inherently a global issue, like the climate crisis. Just as carbon dioxide affects the entire planet regardless of who emits it or where, the risks and benefits created by AI transcend borders. However, the current regulatory system operates on a nation-state basis. The core of the problem is the structural mismatch where events occur globally but regulations remain at the national level. Just as the problem of tax havens arises from this mismatch, AI regulation has the same structural loophole.
There is a lesson from the Peloponnesian War in ancient Greece. At that time, the development of sailing technology had created the conditions for the entire Mediterranean to be integrated into a single economic zone. However, the city-states fought for hegemony and perished together. They failed to create a political order worthy of the new possibilities created by technology. The two World Wars of the 20th century share the same structure. The level of economic integration at the time required a supranational order like the United Nations or the European Union, but nation-states were preoccupied with hegemony competition.
In the AI era, we stand at the same crossroads. It is necessary to build global governance beyond national-level AI development competition. It is a realistic starting point to cooperate with the European Union, which is currently most advanced in AI regulation, and to activate discussions at the UN level. Instead of leaving AI development solely to profit-seeking corporations, we should also consider models like the Manhattan Project, where the state establishes laboratories and bears labor and development costs, but the results are released as open source.
The starting point is understanding what is currently happening. We must go beyond the elementary level of fear marketing that AI might take away jobs and bring the core issues—the essence of digital intelligence, the problem of unelected tech power, and the absence of international governance—into the public discourse.
Korea has the experience of deterring the fierce global fascist trend through the collective action of citizens, such as the candlelight protests in 2016. The experience of winning as a collective is a huge asset. Citizens who learned they could stop military vehicles and hold onto gun barrels can exercise the same collective intelligence in the AI era. It is the task of our time to ensure that the fruits of productivity improvement created by technological innovation are not monopolized by some capital but returned to society as a whole, and to ensure that consciousness and institutions catch up with the speed of technology.
Humans have not changed from thousands of years ago to now. The lust for power, fear, greed, and courage of figures from 2,000 years ago described by Sima Qian are repeated exactly by today’s AI company executives and politicians. What has changed is the tool in their hands. In the past, it was at most swords and spears, but now it is nuclear bombs, neutron bombs, and artificial intelligence. Realizing how dangerous the mismatch between unchanging human nature and the tools it possesses can be—that is why we must have this discussion now.
