Japanese AI is Amazing: An Interview with AI Researcher Hiroaki Kano

The discussion centers on different types of AI, including deep learning AI used for image recognition, and generative AI like ChatGPT that responds to user queries based on internet knowledge.

Kano highlights a Japanese-developed AI framework called FKDP, which learns through trial and error, similar to how humans develop intuition and experience. Unlike traditional AIs that only analyze data or text, FKDP can mimic the decision-making of craftspeople—for example, adjusting salt levels when making pickled plums, or learning how to carve wood like a skilled carpenter.

This AI is already in practical use, such as in chemical plants, where it has operated continuously for two years, performing control tasks more effectively than human operators. It’s also being tested for creating complex recipes, like beer brewing, by learning from temperature changes and experimenting on its own.

Kano stresses that this kind of AI is not meant to replace humans, but rather to assist and enhance human skills. He draws comparisons to AlphaGo, the AI that defeated the Go world champion, who then used the experience to further improve his own skills. He believes AI can inspire new possibilities rather than make humans obsolete.

The conversation also touches on the current limits of AI. For instance, while generative AIs have read most of the content on the internet, they are reaching a saturation point where there isn’t much new data left to learn from. AI systems don’t truly “understand” what they process—they manipulate language patterns, like solving word puzzles, rather than having real comprehension or emotions.

Kano believes that true human qualities, such as empathy and the ability to physically experience the world, remain beyond AI’s reach for now. While AI might seem to have personality or feelings, it's really a reflection of clever design and our own projections. Whether AI will ever gain emotions or consciousness remains an open question even among experts.

The segment ends with a mention of Kano’s new book on “AI-First” practices using strong AI, encouraging deeper exploration of the subject.

Source: 上泉雄一のええなぁ!