DeepSeek: The AI Revolution for Japan at a Fraction of ChatGPT’s Cost

“Is DeepSeek smarter than humans?”

When asked this question, DeepSeek thought for six seconds before responding:

“DeepSeek provides data-driven answers and can efficiently process information, but it does not possess elements such as human creativity, ethics, or intuition based on experience. It can be said that we complement each other.”

In January 2025, the China-developed generative AI DeepSeek-R1 burst onto the scene like a comet, shocking the world. Despite having development costs of less than one-tenth of GPT-4—developed by U.S.-based OpenAI for $78 million (approximately 11.6 billion yen)—it boasted nearly the same performance.

U.S. President Trump, who has been promoting the adoption of generative AI as a national strategy, immediately reacted, stating, “This should serve as a warning that we need to focus all our efforts on competition in order to win.” Thus, an “AI war” between the U.S. and China erupted.

AI researcher Shota Imai analyzes that the emergence of DeepSeek will bring significant changes to Japan as well.

“The reasons why DeepSeek is attracting attention are its overwhelmingly low cost and the fact that its development methods have been made public. This means that anyone with a few hundred million yen can create a generative AI on par with DeepSeek.

Developing a generative AI typically requires an initial investment of hundreds of billions of yen—comparable to building a Tokyo Skytree tower. Ultimately, trillions of yen would be needed to produce something of high quality. As a result, Japanese companies weren’t even at the starting line.

However, thanks to DeepSeek making its development methods public, if properly replicated, a generative AI of the same level can be created at a very low cost. Japan has suddenly been catapulted to the starting line. The idea that Japan can now create its own original generative AI is extremely exciting for us researchers.”

One key difference between DeepSeek and previous generative AIs is that it possesses thinking ability.

“The reason DeepSeek and ChatGPT can have thinking ability is that they have incorporated a method called reinforcement learning.

Traditional generative AI used a learning method where it memorized answers that humans had already derived. In contrast, reinforcement learning repeatedly solves problems and derives answers on its own, allowing the AI to think about how to solve them.

Traditional AI, having been taught a vast amount of knowledge by humans, could pass exams that test memorization, such as the bar exam or the national medical licensing exam. However, it stood no chance against University of Tokyo entrance exam math problems, which require critical thinking.

On the other hand, generative AI trained with reinforcement learning can solve these high-level math problems. DeepSeek has achieved this at a low cost.” (Imai)

With the evolution of generative AI, how will the way Japanese people work change?

Highly competent AI employees

Satoshi Endo, CEO of the AI venture company LifePrompt, compares the latest generative AI to a “fresh graduate from the University of Tokyo.”

“A new graduate from the University of Tokyo possesses a wealth of foundational knowledge and can communicate effectively, but lacks company-specific expertise and knowledge of how tasks are carried out. However, with proper training, they can absorb industry-specific and company-specific knowledge, eventually becoming valuable contributors. This trained graduate is what generative AI represents.

For instance, in a consulting firm, research tasks already follow a set methodology and sources. If these processes are clearly defined and taught to a new employee, they can quickly adapt. The same principle applies to generative AI—by feeding it proprietary company knowledge, it can be transformed into a highly competent AI employee tailored to the business.”

In fact, more companies are incorporating generative AI into their operations. In April 2024, Sumitomo Corporation fully adopted Copilot for Microsoft 365, an AI tool developed by Microsoft, using it extensively for tasks such as drafting emails, creating documents, and summarizing meetings. Other global enterprises, including Seven & i Holdings, Kirin Holdings, and Fujifilm Holdings, are also moving towards full-scale AI integration.

With the emergence of DeepSeek, AI is becoming more accessible and cost-effective, potentially allowing even small and medium-sized businesses to adopt it.

Still, how will this impact our daily lives?

Restaurant reservations are already being handled by AI, but the technology now goes beyond simple bookings. AI like DeepSeek can perfectly accommodate complex requests, such as “Suggest a menu for four people with a budget of 10,000 yen.” Additionally, AI can provide tailored responses for personal matters like “How should I ask someone I just met out on a date?” or “Can you help me write a birthday letter to my parents?”

Could AI ever replace humans entirely?

“For sales professionals, AI can create proposal documents, but it cannot visit clients, build trust, or close deals. The ability to read subtle emotional cues through tone and facial expressions, and to instill confidence in a customer—these are uniquely human qualities that AI cannot replicate,” Endo explains.

Reflecting once more on the initial question—Is DeepSeek smarter than humans?—it seems DeepSeek gave the correct answer after all.

From the March 14-21, 2025 issue of “FRIDAY”