難以言喻的映射可能：陽春麵硏究舍的潛空間探索

The Possibility of Ineffable Mapping: Simple Noodle Art's Exploration of the Latent Space

在基於深度學習所建構的人工智慧系統中，文本、音訊、影像等非結構化資料，被濃縮、轉換成向量形式，映射在一個高維度的座標系統中，也就是所謂的潛空間（Latent Space）。在這個空間裡，每筆資料的向量或是座標即構成其表徵（Representation），而這些向量之間的距離、方向與分布則反映了它們在語意或特徵上的相互關係。

In Artificial Intelligence, the ''Latent Space'' is not a neutral, idealized realm; rather, it is a microcosm of information that reflects real-world power dynamics and cultural boundaries, shaped by the uneven distribution of training data. Many works by Simple Noodle Art explore these reflections on latent space through their artistic practice.
In their earlier work, the team discovered that Al image restoration technology treated the Earth in space photography as noise and removed it. In response, they intentionally removed the moon-which had been identified by algorithms-in their work, revealing the disparity between algorithmic perception and human-ascribed meaning. The team further explored the significance of ''noise'' in their recent works. By introducing real-world Gaussian noise through physical devices and pitting hand-drawn errors against computer precision, they contemplated abstract meanings within the juxtaposition of ''Artificial Intelligence'' and the ''Artificial'' (human-made). The high dimensionality and stochasticity of Al are difficult to articulate using human a priori knowledge entirely; art, as a similarly ineffable context, offers a potential mapping to peek inside this black box.