[논문 리뷰] Smell with Genji: Rediscovering Human Perception through an Olfactory Game with AI

Awu Chen, Vera Yu Wu외 8명|arXiv (Cornell University)|2026. 02. 02.

Olfactory and Sensory Function Studies인용 수 0

한 줄 요약

Genji-kō를 sensing AI co-smelling partner로 보강하는 AI 매개 후각 상호작용 시스템으로, 협력적 향 비교 및 성찰 대화를 통해 인간과 기계의 후각 인식을 탐구한다.

ABSTRACT

Olfaction plays an important role in human perception, yet its subjective and ephemeral nature makes it difficult to articulate, compare, and share across individuals. Traditional practices like the Japanese incense game Genji-ko offer one way to structure olfactory experience through shared interpretation. In this work, we present Smell with Genji, an AI-mediated olfactory interaction system that reinterprets Genji-ko as a collaborative human-AI sensory experience. By integrating a game setup, a mobile application, and an LLM-powered co-smelling partner equipped with olfactory sensing and LLM-based conversation, the system invites participants to compare scents and construct Genji-mon patterns, fostering reflection through a dialogue that highlights the alignment and discrepancies between human and machine perception. This work illustrates how sensing-enabled AI can participate in olfactory experience alongside users, pointing toward new possibilities for AI-supported sensory interaction and reflection in HCI.

연구 동기 및 목표

• Address the olfactory–verbal gap by reframing scent perception as a collaborative, reflective activity.
• Reinterpret Genji-kō as an AI-supported social interaction rather than a digitized solo task.
• Demonstrate how an AI co-smelling partner can perceive temporal scent dynamics and engage in meaningful dialogue with humans.
• Provide a system that combines olfactory sensing, LLM-based reasoning, and mobile visualization to foster sensory discovery and well-being.

제안 방법

• Implement a guided Genji-kō inspired game with a mobile app and an AI co-smelling partner.
• Use time-series olfactory sensing (MOS sensors) to capture scent dynamics without burning incense.
• Develop a temporal scent classification model with a Transformer encoder and MLP head for 5-class incense categorization.
• Employ retrieval-augmented generation (RAG) with static and dynamic databases to support context-aware dialogue.
• Visualize Genji-mon patterns in real time and provide a final debrief comparing human and AI interpretations.

Figure 1. Diagram showing the interactive experience of Smell with Genji

실험 결과

연구 질문

• RQ1 How can AI participate as a co-smelling partner to augment olfactory perception and reflection?
• RQ2 Can temporal sensor data and an LLM dialogue facilitate deeper articulation and comparison of scents than traditional Genji-kō?
• RQ3 What is the nature of alignment or discrepancy between human judgments and machine interpretations in olfactory experience?
• RQ4 Does grounding olfactory dialogue in sensor data and prior interactions foster richer reflective practices?

주요 결과

• The system enables five rounds of smelling, comparison, and dialogue that progressively build a Genji-mon pattern.
• The AI perceives temporal sensor dynamics and engages in dialogue grounded in both sensor data and prior sessions.
• AI interpretations are designed to be supportive rather than competitive due to limited classification accuracy (~40% under controlled conditions).
• Participants experience moments of alignment and discrepancy with the AI, prompting deeper reflection on scent similarity judgments.
• A final reveal juxtaposes the participant’s Genji-mon with the AI-informed pattern, followed by a debrief on how sensor trends informed judgments.

Figure 2. Overview of the Smell with Genji setup. (a) Mobile application for guided gameplay and Genji-mon visualization. (b) AI co-smelling partner embodied in a 3D-printed enclosure with sensing and display. (c) Incense holder for smelling.