There is a vast hidden workforce behind AI Will they become redundant as the technology develops? Apr 10th 2025| SUMMARY When DeepSeek, a Chinese AI firm, released a cheap and efficient large language model in 2024, it challenged long-held beliefs about the resources needed to build AI—hardware, energy, and especially human labor. Behind every AI model lies a workforce, largely unseen, responsible for annotating data such as labeling images and transcribing audio to help machines "understand" the world. 2024 年，中國人工智慧公司 DeepSeek 發布了廉價高效的大型語言模型，挑戰了人們長期以來對建立人工智慧所需資源（硬體、能源，尤其是人力）的看法。每個人工智慧模型背後都有一支很大程度上看不見的勞動力，負責註釋數據，例如標記圖像和轉錄音頻，以幫助機器「理解」世界。 Though often dismissed as “unsexy,” this data work is foundational. Since Fei Fei Li’s ImageNet project in the 2000s, firms have outsourced** annotation** to low-cost labor markets, like India and rural China. While firms like Scale AI and iMerit professionalized the field, many annotators still earn low wages and face tight monitoring. 儘管這些數據工作常常被認為是“不夠吸引人”，但卻具有基礎性。自 21 世紀李飛飛( ai教母--李飛飛，史丹佛大學首位紅杉講席教授，美國國家工程院院士，美國國家醫學院院士，美國文理科學院院士。)的 ImageNet 專案以來，企業已將註釋工作外包給印度和中國農村等低成本勞動市場。儘管 Scale AI 和 iMerit 等公司使該領域專業化，但許多註釋員的工資仍然很低，並且面臨嚴格的監控。 However, the nature of annotation is changing. Basic labeling tasks are declining as AI matures and begins to label its own data or use pre-labelled datasets. Advanced models require higher-skilled annotators—often with PhDs or coding expertise—to handle nuanced tasks like evaluating chatbot responses or checking AI-generated synthetic data. 然而，註釋的性質正在改變。隨著人工智慧的成熟並開始標記自己的數據或使用預先標記的數據集，基本標記任務正在減少。高階模型需要更高技能的註釋者（通常具有博士學位或編碼專業知識）來處理細微的任務，例如評估聊天機器人的回應或檢查人工智慧產生的合成資料。 This human input remains crucial. AI’s complexity, especially in language models, demands human judgment to ensure accuracy, ethical alignment, and cultural nuance. For example, ChatGPT’s frequent use of the word “delve” reflected the linguistic habits of its African annotators, illustrating how human influence shapes AI output. 這種人力投入仍然至關重要。人工智慧的複雜性，尤其是在語言模型方面，需要人類的判斷來確保準確性、道德一致性和文化細微差別。例如，ChatGPT 頻繁使用「delve」一詞，反映了其非洲註釋者的語言習慣，說明了人類的影響如何塑造人工智慧的輸出。 Although the industry aims for fully self-training AI, models still rely heavily on humans—especially for feedback,** edge cases**, and *ethical oversight*. As AI advances, the demand may shift but not disappear. Annotators, once likened to parents guiding AI's early steps, are evolving into advisors helping AI navigate a complex world. 儘管該行業的目標是實現完全自我訓練的人工智慧，但模型仍然嚴重依賴人類——尤其是在反饋、邊緣情況和道德監督方面。隨著人工智慧的進步，需求可能會改變但不會消失。註釋者曾經被比喻為引導人工智慧早期發展的父母，現在正在演變成幫助人工智慧探索複雜世界的顧問。 In short, far from being obsolete, human annotators remain essential to AI’s development—just in new and more sophisticated roles. 簡而言之，人類註釋者遠未過時，他們對於人工智慧的發展仍然至關重要——只是扮演著新的、更複雜的角色。 -- Hosting provided by SoundOn

Ask this episode

“What did the council decide on item 42?” — answered with the speaker and the moment.

Transcript

Related