AI partners help humans describe scents through a Japanese game

A new interactive experience called Smell with Genji uses artificial intelligence to help people articulate their olfactory perceptions.

Researchers developed a system that adapts Genji-ko, a traditional Japanese scent-matching game, to bridge the gap between smelling something and finding the right words to describe it. The system integrates multichannel olfactory sensors with a Transformer-based classification model and a large language model. These components work together to translate chemical signals into descriptive dialogue.

Instead of simply identifying a smell, the AI acts as a co-smelling partner that shares its own sensory impressions with the user. This setup encourages participants to reflect on their own intuition by comparing it to the machine's perspective. By providing visual patterns of comparative judgments, the tool helps users navigate the complexities of scent. The project aims to foster reflective communication while ensuring that human judgment remains the primary focus of the experience.