The fate of the Republic of Korea as diagnosed by the new government's AI future planning chief.
And the future direction and strategy of Korean AI

Kwak No-jeong, Park Tae-woong, Yoon Ui-jun, Lee Hae-min…
Highly recommended by leaders in the AI ​​industry, academia, politics, and society!

《AI Wars 2.0》 is the sequel to 《AI Wars》, which was published in 2023 and caused a huge stir at the time.
In 2023, when many media outlets and books were excited by the ChatGPT craze, introducing foreign technologies and global trends, "AI War" had already analyzed the current state of artificial intelligence in South Korea and forecasted its future.
Two years later, the warnings of the previous work have become reality, and we must re-strategize.
"AI War 2.0" is an urgent report that soberly diagnoses the existential crisis facing South Korea in this rapidly changing global AI landscape and presents concrete solutions for reversing the situation.

Author Ha Jung-woo served as the head of Naver's Future AI Center and is currently the first senior secretary for AI future planning under the People's Sovereign Government, directly designing Korea's AI policy.
His analysis, based on his academic achievements such as over 15,000 citations on Google Scholar, participation in the NeurIPS and ICML organizing committees, and being the youngest full member of the National Academy of Engineering of Korea, as well as his vivid experience in policymaking, is more realistic and persuasive than any other AI book.
In particular, his diagnosis that “if the AGI era arrives within 3 to 5 years, a country that secures it on its own will have national power and global influence greater than that of a nuclear weapons state” clearly shows that AI is not just a technology but a matter directly related to national survival.

The plans presented in the book, including building an AI data center, establishing a national superintelligence research institute, and creating an AI digital innovation department, are a roadmap the author believes can be implemented during the policy-making process.
The author's message that "the AI ​​era is still in its infancy, and opportunities still exist" offers a powerful hope to South Korea, which is facing a critical moment.
This is why this book is drawing attention not only from the AI ​​industry and academia, but also from political and social leaders, such as Kwak No-jeong (CEO of SK Hynix), Park Tae-woong (CEO of Everyone's Question Q), Yoon Eui-jun (Chairman of the National Academy of Engineering of Korea), and Lee Hae-min (22nd National Assembly member).

《AI War 2.0》 sharply dissects the strategic crossroads facing South Korea in the global battle for AI hegemony.
And we present our own survival strategy to survive in the chaotic battlefield.
If you're curious about the coordinates and future horizons of AI in Korea, and want to discover the hidden opportunities within, then read this book now.

This is because many people confuse the open source we talk about in AI models with the traditional open source we used to know.
Traditional open source makes the source code public, allowing users to freely use the code.
Because the code itself is the most important core.
However, open source of AI models only involves disclosing the weight files or some code of the trained model.
Also, isn't open source code a relatively minor factor? So, while the Open Source Institute (OSI), which defines open source, doesn't recognize it as such, we often call it open source.
Because of this misunderstanding, most executives tend to think too easily, "It's open source, so why don't we just make something with it?"
--- p.38

OpenAI has defined AGI very clearly.
The final stage of AGI is AI that has the ability to completely replace an organization.
The reason I used the term 'organization' is because an organization is a structure in which people with various abilities come together to perform work.
In other words, if AI can specifically perform the organization's tasks, it can be judged as AGI, and at this time, the role and value of AGI can also be clearly determined.
--- p.66

While there may have been differences in detail across the Obama-Trump-Biden-Trump administrations, I believe the overall framework of the principle that the United States must maintain strong leadership in AI remains unchanged.
Rather, when looking solely at the topic of AI, I believe the Biden administration's regulations have been somewhat strict due to safety concerns, and there's a stronger push to ease them.
In other words, the direction itself is the same, but the difference is whether to take a cautious, regulation-oriented approach or to pursue it actively.
--- p.81

China has a structure that allows the government to quickly decide on data utilization, regulations, and national resource priorities as it sees fit.
China is probably the only country in the world where the government can access not only national data but also corporate data at any time.
This kind of strong centralized control strategy is difficult for other countries to follow.
No matter how strong the Trump administration's policies are, they cannot operate as efficiently as China.
Additionally, China is actively utilizing open source to expand its influence around the world, a strategy they call the "Silk Road of AI."
I believe this is also greatly beneficial to the expansion of China's AI ecosystem.
--- p.113

The recently released Claude 4 reportedly activated ASL-3 protection measures for the first time. This indicates that ASL-3 is both powerful and highly dangerous.
This means that, especially after the release of the Claude 3.7 Sonnet model, the risks that were only seen as potential in the chemical, biological, radiological, and nuclear fields have now been assessed to be at a level where they could become actual threats.
In a real-world biological weapons acquisition scenario, Claude 4 Opus was shown to improve expected quality by a factor of 2.53.
They say similar results were found in external red team evaluations.
Now, with AI reaching Claude 4, there may be growing calls for governments or international organizations, rather than companies, to manage how and for what purpose it is being used.
--- p.168

No matter how powerful a generative AI is, it can only have real impact when it runs on a platform, but neither OpenAI nor Antropic have their own platforms.
Simply put, they lack their own playground. Even if AI agents possess exceptional reasoning and planning abilities, the difference between companies that provide content and have services or platforms to connect with others and those that must build all connections from scratch or through transactions is bound to be significant.
In this respect, hyperscalers have a significant advantage.
--- p.175

Among recent malicious use cases of AI, one problem has emerged that is both intriguing and serious.
This is exactly the case in Russia, where they created millions of articles favorable to themselves, spread them online, and then let AI learn from them.
They call this LLM grooming.
In other words, the problem is that just because AI learns from a vast amount of content, its results do not necessarily align with common human values.
--- p.258

As we enter an era where AI and humans coexist, not only human relationships but also the relationship between humans and AI is expected to undergo significant changes.
There is a good chance that people will become overly dependent on or addicted to AI, as we have already experienced with games and social media.
Of course, games and social media themselves are not intended to be addictive, but there are definitely cases where they lead to addiction.
In the past, when TV became popular, the term "TV addiction" came up, but that expression was rarely used when talking about books or music.
But I think AI has the potential to be really powerful and addictive to people.
In fact, there are cases among teenagers these days where they avoid real human relationships because of AI, or even feel more comfortable in romantic relationships with AI.
--- p.283

For example, according to the Computer Vision Society, Korea ranks third in the world in the number of papers.
In the fields of natural language processing and machine learning, it was outside the top 10 until a few years ago, but recently it has risen to the top 7.
However, despite this large number of papers, the proportion of papers that have actual industrial impact or receive global attention is relatively small.
One of the reasons is brand awareness.
While global companies like Google, Meta, DeepMind, and OpenAI unconditionally attract global attention when they publish papers, Korean research institutions still lack that brand power.
Moreover, our research team is relatively lacking in actively promoting the results of their papers globally or making them open source.
--- p.305

The first is top-tier AI research talent capable of driving global-level AI innovation.
These are people who have worked really hard in AI graduate schools at KAIST and Seoul National University.
However, these talented individuals do not stay in Korea after graduating with a doctorate; almost all of them go abroad, especially to the United States.
The reason is simple.
This is because the research environment overseas is better, there are many excellent researchers, the treatment is good, and you can freely use infrastructure such as GPUs.
In Korea, it is difficult to conduct innovative research, so we need policies to first attract these talented individuals.

--- p.334~335