Inter net - Labeling me — 陳姿尹 Chen Zi Yin

Inter net - Labeling me
Print
Dimensions variable
2023

Inter net - Labeling me
輸出品
尺寸依場定而定
2023

（圖片提供：忠泰美術館 Jut Art Museum）

In large image datasets used to train AI models for image recognition and generation, the “baseline” (image classification, description, etc.) is “labeled” through “crowdsourcing” done by Amazon Mechanical Turk. Labeling through crowdsourcing is theoretically the “collective opinion of the public” and matches the goal of maintaining as objective as possible requested of many AI systems.

Inter net – Labeling me is extended from the art series, Inter net. The work comprises two walls, one of which is covered by photos put into the category of “self” by AI in cellphones. The descriptions below the photos are “labels” obtained via crowdsourcing by Amazon Mechanical Turk. The other wall is covered by all of the artists’ unselected search entries since they began using Google over a decade ago: these include worries about budding romantic relationships, hesitation about choosing a major in college, anxieties over political situations, or a simple desire to find a deli to eat. The Google search entries form a personal profile unconsciously created and accumulated over more than ten years. Audiences are welcome to tear down the Google search entries and paste them below the photos, and conduct an alternative “crowdsourced labeling” using the artists’ self-writing of “I.”

用於訓練影像辨識、生成等等的AI模型使用的「大型影像資料集」中，訓練AI時的「基準」（影像的分類、描述等等）的一大來源是使用像是Amazon Mechanical Turk這類的「群眾外包」平台，透過發包給大眾的方式進行「標註」。眾包的標註是理論上是常態分布的，也就是說這樣的標註會是一種「大眾的集體意見」，亦符合多數AI盡可能客觀的目標。

《Inter net - Labeling me》為《Inter net》系列的延伸作品。作品包含兩個牆面，其中一個牆面貼滿了手機裡被AI分類為「自己」的照片，照片下方的描述我透過Amazon Mechanical Turk發包所得到的「標註」；另一個牆面貼滿了我開始使用Google至今的十幾年間，未經挑選的所有搜尋紀錄，裡面有著情竇初開時的煩惱、選擇科系的徬徨、對政治情勢的焦慮......，又或者只是想找一個盜版網站、一個英文單字的翻譯，或一間小吃店......。搜尋紀錄就像是累積了十幾年的無意識自我人格側寫。觀者可以將搜尋紀錄撕下來，貼到照片下方，試著用我的自我書寫，進行另一種意義上的「眾包標註」。

AI正逐漸嵌入我們的生活中，我們也許也將會跟依賴網際網路一樣地依賴AI，但AI似乎更需要那些不認識我的大眾所標註的「客觀集體意見」，而不是自我書寫之下的我。