This chapter examines Shitsukan recognition as a key challenge for computer vision and multimodal AI. Unlike object recognition, Shitsukan involves subtle properties—such as glossiness, transparency, and softness—as well as holistic impressions like warmth or atmosphere. We review classical approaches that frame Shitsukan recognition as material classification, multi-label attribute prediction, or ranking-based estimation, noting their limitations in scope and robustness. Recent advances in web-scale image—text mining and contextual captioning broaden recognition beyond predefined categories. With the rise of generative AI, multimodal large language models (MLLMs) trained on massive image—text corpora now achieve remarkable image understanding, often capturing Shitsukan at near-human levels. Yet challenges remain, including generalization beyond training distributions and the absence of embodied experience. We argue that Shitsukan recognition highlights both the potential and the limits of current AI, offering a path toward deeper alignment with human perception.