Japan has a strong history in artificial intelligence (AI), dating back to the 1960s. As far back as 1967, Shun’ichi Amari proposed a way to enable neural networks to self-adjust the way that they categorise patterns through repeated training examples. This was one element that led to backpropogation, a key development in AI. Geoffrey Hinton, Nobel laureate, acknowledged the contribution made by Amari and others when he was awarded the Nobel Prize in Physics in 2024. The first multilayer convolutional neural network, the foundation of deep learning, was called the Neocognitron, and was created by Japanese researcher Kunihiko Fukushima in 1979. His research was fundamental to visual pattern recognition, used in facial recognition and in the technology that allows self-driving cars to avoid hazards.
The Japanese Fifth Generation Computer Project in the 1980s was a major investment in AI research, though research efforts dwindled in the 1990s in the “second AI winter”, just as they did in Western countries. Deep learning and natural language research continued, with the emphasis switching to generative AI, in particular, in the development of large language models in Japanese. Japanese is a complex language, with three different scripts: kanji, hiragana and katakana. Sentences are often written in a combination of these, and there are a great many characters. Compared to the 26 letters of the English alphabet, there are over 50,000 characters, though only around 2,000 are used in most everyday communication. This linguistic difference means that there is much less training data available for large language models than English (just 4% of internet content is in Japanese), but that also represents an opportunity. Companies developing Japanese foundational language models include TDAI Lab, ELYZA and rinna.
The Japanese government launched the AI Promotion Act in June 2025, aimed at encouraging AI innovation. This sets out principles and a framework for the government to promote AI, working with business and academia. It calls for transparency and ethics in AI research, but has a light-touch regulatory approach, encouraging voluntary alignment with government AI strategies. There are several interesting Japanese AI start-up companies. Sakana AI is exploring alternative architectures to build foundation models. Preferred Networks focuses on applying AI to manufacturing and, partnering with Toyota and others for industrial applications of AI. Jitera has built an AI development platform to automate the software development process, while ABEJA has machine learning solutions for industries including logistics and manufacturing.
Japan has long been a powerhouse in robotics, with 38% of the world’s industrial robots in 2024 being built in Japan. It is also the second-largest market for robots, behind China. These robots do precision welding, factory automation and even act as surgical assistants. Woven by Toyota is an AI unit that works on driverless vehicles and manages the software for Toyota’s vehicle operating system. One challenge that particularly resonates in Japan is the ageing population, with 17% of Japanese now aged over 75, and with an acute shortage of care workers. The government has been investing in research into AI-driven robots that may, in time, be able to help. For example, AIREC is a humanoid robot that can turn patients in bed and even put on socks or fold laundry. Many challenges remain, but the prototype AIREC robot is already being trialled and may be able to enter service in a few years. A less ambitious robot called Moxi is already used in hospitals to deliver supplies and lab samples. There are already robots in use that can carry out limited tasks, such as assisting hospital patients with standing and walking. Such tools are likely to supplement and assist human caregivers rather than replace them. Monitoring robots are already used by 63% of nursing homes in Japan.
Japan has a rich history in AI, and has not always been credited in Western media with its genuine and influential contributions to AI research. It has great depth in robotics, and some unique challenges, such as a rapidly ageing population that is driving innovation in the use of robots for elderly care. Its linguistic uniqueness causes issues in terms of limited training data, but may also represent some opportunities for domain-specific applications optimised for Japanese language processing. Japan has historically often forged its own path, such as the Edo period (1603-1868), when it largely cut itself off from the rest of the world. Although there were obvious drawbacks, that period led to some innovations such as advanced woodblock printing and rice futures trading exchanges, driven through necessity. Modern Japan is an open society, but as anyone who has spent any time there will attest, it has many unique cultural aspects. It remains to be seen whether these will foster any unique developments in AI. Japanese culture has long been a distinctive blend of tradition and innovation, and this unusual history and background may represent an opportunity for developments in robotics and AI that differ from those in Western cultures.
