When ChatGPT exploded onto the scene, it was Silicon Valley’s latest crown jewel – an AI marvel woven from the English language. But across the Pacific, Japan had its own ambitions. Rather than passively adopt this Western innovation, Japan aimed to seize the AI high ground on its own terms, in its own tongue.
The government, researchers, and tech titans like NEC and SoftBank are pooling resources to develop “wa kotoba AI” – AI that comprehends the intricacies of the Japanese language and culture. They are investing hundreds of millions to birth LLMs – large language models – nurtured on Japanese data, not just translations of English counterparts.
This undertaking faces profound technical challenges. As Keisuke Sakaguchi of Tohoku University explains, Japanese sentence structure and grammar differ radically from English. Japanese also utilizes 48 basic characters, over 2,000 everyday kanji, and 50,000 rare kanji, each with multiple pronunciations. This enormous complexity trips up AI systems steeped in English, like ChatGPT.
When using ChatGPT to generate Japanese text, “it often produces extremely uncommon kanji characters that most native speakers have never encountered, resulting in disjointed words and phrases,” Sakaguchi notes. To power innovation domestically, Japan’s AI must achieve true fluency, not just clumsy translations.
NEC is already reaping rewards from its new Japanese-based AI system, cutting report writing time by 50% and accelerating software coding by 80%. Meanwhile, SoftBank is investing ¥20 billion in a Japanese-trained LLM launching next year to turbocharge business productivity.
But the most ambitious efforts are still to come. In 2023, researchers will release an open-source LLM with over 30 billion parameters trained on Japanese corpora. By 2031, scientists aim to develop a boundary-pushing scientific LLM starting at 100 billion parameters and expandable over time. This AI scholar could catalyze discoveries by analyzing decades of Japanese research.
With scientific rigor and cultural fluency combined, Japan’s AI has the potential to unlock innovations that ripple through every facet of society. By colonizing the algorithmic frontier on its own terms, Japan is determined to shape the future in its own image and transform the nation into an AI powerhouse.
References
- Zheng, L. et al. Preprint at https://arxiv.org/abs/2306.05685 (2023).