Technology

Why Japan is lagging behind in generative A.I. — and how it can create its own large language models

Japan is known for its futuristic technology. But the nation is lagging behind in the generative AI race and is trying to create its own large language models.
Mr.cole_photographer | Moment | Getty Images

Countries are racing to develop their own generative artificial intelligence algorithms, but high tech Japan is already behind.

Generative AI has been the trendiest topic in tech since OpenAI made waves with its chatbot ChatGPT. Breakthroughs in generative AI possess the potential to fuel a 7% increase in global GDP, or almost $7 trillion, over the next decade, according to Goldman Sachs research.

Key to generative AI development are large language models which underpin the likes of ChatGPT and Baidu’s Ernie Bot, capable of processing vast data sets to generate text and other content. But Japan is currently trailing behind the U.S., China and the EU in developing these algorithms, said Noriyuki Kojima, co-founder of Japanese LLM startup Kotoba Technology.

Chinese organizations, including tech giants Alibaba and Tencent, have launched at least 79 LLMs domestically over the past three years, Reuters reported in May citing research from a consortium of state-run institutes. U.S. corporate powerhouses such as OpenAI, Microsoft, Google and Meta play a significant role in propelling the country’s LLM advancements, said Kojima.

Japan lagging behind in generative AI

Japan, however, lags behind the U.S., China and Europe in the scale and speed of its LLM development.

“Japan’s trailing position in the field of generative AI largely stems from its comparative shortcomings in deep learning and more extensive software development,” said Kojima.

Deep learning requires a “robust community of software engineers” to develop necessary infrastructure and applications, Kojima added. Japan, however, will face a deficit of 789,000 software engineers by 2030, according to the Ministry of Economy Trade and Industry. The nation is now ranked 28th out of 63 countries in terms of technological knowledge, according to the IMD World Digital Competitiveness Ranking.

Japan also faces hardware challenges as LLMs need to be trained using AI supercomputers like IBM’s Vela and Microsoft’s Azure-hosted system. But no private company in Japan possesses its own “world-class machine” with those capabilities, Nikkei Asia reported.

Government-controlled supercomputers like Fugaku therefore “hold the key” to Japan’s pursuit of LLMs, Kojima explained.

“Access to such large-scale supercomputers forms the backbone of LLM development, as it has traditionally been the most significant bottleneck in the process,” he said.

How Japan’s supercomputers can help

Tokyo Institute of Technology and Tohoku University plan to use Fugaku to develop LLMs based primarily on Japanese data in collaboration with the supercomputer’s developers Fujitsu and Riken, Fujitsu announced in May.

The organizations plan to publish their research results in 2024 to help other Japanese researchers and engineers develop LLMs, Fujitsu added.

The Japanese government will also invest 6.8 billion yen ($48.2 million), about half the total cost, to build a new supercomputer in Hokkaido that will begin service as early as next year, Nikkei Asia reported. The supercomputer will specialize in LLM training to promote Japan’s development of generative AI, said Nikkei Asia.

In April, Japanese Prime Minister Fumio Kishida said the country supports the industrial use of generative AI technology. Kishida’s remarks followed his meeting with OpenAI CEO Sam Altman, who said the company is looking to open an office in Japan.

Japanese companies pursuing generative AI

Big Tech players have also joined the fray to boost Japan’s standing in generative AI. In June, SoftBank’s mobile arm said it plans to develop its own generative AI platform, reported local media. This was underscored by SoftBank CEO Masayoshi Son’s announcement that the investment firm plans to shift from “defense mode” to “offense mode” and intensify its focus on AI.

“We would like to be [in] the leading position for the AI revolution,” Son said during a shareholders’ annual general meeting.

SoftBank Group sold its 85% stake in SB Energy to Toyota Tsusho in April and recently agreed to sell its 90% stake in U.S. investment manager Fortress Investment Group, Nikkei Asia reported. Trimming these other investments helps SoftBank free up cash, allowing it to focus largely on AI through its Vision Fund venture capital investment unit.

SoftBank-owned chip design company Arm is also set to pursue a U.S. IPO listing later in the year. “It will be by far the biggest IPO that’s hit the world,” said Amir Anvarzadeh, Japan equity market strategist at Asymmetric Advisors.

The IPO will provide a hefty sum to boost funds at SoftBank, which reported a record 4.3 trillion yen loss at Vision Fund for its fiscal year ending March 31.

Arm originally sought to raise between $8 billion and $10 billion. But with demand for semiconductor chips “through the roof,” Anvarzadeh suggested Arm could raise as much as $50 billion to $60 billion — or “85% of SoftBank’s market cap.”

He said SoftBank’s share price will likely rise, although this does not guarantee the success of its AI efforts.

“Fundamentally, I don’t think SoftBank is going to change Japan’s landscape … they are no savior of Japan’s AI,” he said.

Japanese telecommunications company NTT also announced plans to develop its own LLM this fiscal year, aiming to create a “lightweight and efficient” service for corporations. NTT said it will funnel 8 trillion yen over the next five years into growth areas like data centers and AI, a 50% increase from its previous level of investment.

Local media reported that digital ad company CyberAgent released an LLM in May that enables companies to create AI chatbot tools. The company said it is one of few “models specialized in the Japanese language and culture.”

While it has yet to catch up in the generative AI space, Japan is making its first stride with these private sector efforts. Once a “robust infrastructure” is established, the remaining technical challenges are likely to be “significantly mitigated” by using open-sourced software and data from previous pioneers, Kojima said. Bloom, Falcon and RedPajama are all open-sourced LLMs trained on vast amounts of data that can be downloaded and studied.

However, companies venturing into this field should anticipate competition spanning a “relatively longer timeframe,” Kojima said. Developing LLMs requires substantial capital investment and a workforce highly skilled in natural language processing and high-performance computing, he explained.

“SoftBank and NTT, joining this competition, will not change the AI landscape in the short-term.”

AI regulation in Japan

Japanese tech companies’ increased participation in generative AI development coincides with a positive stance on AI adoption in other sectors. Over 60% of companies in Japan have a positive attitude toward using generative AI in their operations, while 9.1% are already doing so, a survey by Teikoku Databank found.

Hitachi has established a generative AI center to promote employee’s safe and effective use of the technology, it said in May. With the expertise of data scientists, AI researchers and relevant specialists, the center will formulate guidelines to mitigate the risks of generative AI, the conglomerate said.

Japan will even consider government adoption of AI technology like ChatGPT, provided that cybersecurity and privacy concerns are resolved, said Chief Cabinet Secretary Hirokazu Matsuno.

As Japan becomes more open to the use of generative AI, the government should formulate and facilitate soft guidelines regarding its use, while assessing the need for hard regulation based on specific risks, said Hiroki Habuka, research professor at Kyoto University’s Graduate School of Law.

“Without clearer guidance on what actions companies should take when using generative AI, practices may become fragmented,” the professor said.