Workers' Daily Tasks Train Robots via Camera Data
Listen to Course
0:00
0:00
#1
In Chennai, southeastern India, a 25-year-old homemaker named Nagireddy Sriramiyachandra attaches a smartphone to her head every morning.
在印度東南部的清奈，一位名叫納吉雷迪·斯里拉米亞錢德拉的 25 歲家庭主婦每天早上都會將智慧型手機固定在頭上。
#2
The camera captures her hands while she slices a mango, folds laundry, and washes dishes.
當她切芒果、摺衣服和洗碗時，相機會捕捉她的雙手。
#3
She earns about 250 rupees per hour of video, which is roughly four dollars.
她每小時的影片大約賺取 250 盧比，這大約相當於四美元。
#4
This type of work has been called "data labor" because the camera data is used for robot training.
這種類型的工作被稱為「數據勞動」，因為相機數據是被用於機器人訓練。
#5
Thousands of Indian homemakers and factory workers have started filming their daily routines.
數以千計的印度家庭主婦和工廠工人已經開始拍攝他們的日常生活。
#6
These first-person videos show how humans grasp objects, move their fingers, and complete tasks step by step.
這些第一人稱影片展示了人類如何一步一步地抓取物體、移動手指並完成任務。
#7
The recordings become a kind of "behavior textbook" that teaches robots how to perform actions like slicing fruit or folding clothes.
這些錄影成了一種「行為教科書」，教導機器人如何執行像切水果或摺衣服這樣的動作。
#8
Agence France-Presse reported that this trend is growing quickly across India.
法新社報導，這一趨勢正在印度各地迅速增長。
#9
This new form of data labor is different from the work that was done for language-based AI systems.
這種新型的數據勞動與過去為以語言為基礎的人工智慧系統所做的工作不同。
#10
Large language models like ChatGPT were trained on text that had already been organized and categorized.
像 ChatGPT 這樣的大型語言模型，是利用已經被整理和分類過的文本進行訓練的。
#11
However, physical AI and humanoid robots need real-world movement data.
然而，實體人工智慧和人形機器人需要真實世界的運動數據。
#12
Technology workers in the field of AI development say that robots must learn from actual human actions and sensory experiences.
人工智慧開發領域的技術人員表示，機器人必須從實際的人類動作和感官體驗中學習。
#13
Automation experts believe that this kind of data collection will become more common in the future.
自動化專家認為，這種數據收集在未來將會變得更加普遍。
#14
Humans naturally adjust their grip when a cup starts to slip, and robots need to learn these small but important reactions.
當杯子開始滑落時，人類自然會調整握力，而機器人需要學習這些微小但重要的反應。
#15
As a result, everyday activities performed by ordinary people have become valuable for AI development.
因此，由一般人所進行的日常活動，已變得對人工智慧的發展極具價值。
#16
The growing demand for camera data has created new job opportunities for workers in developing countries.
對鏡頭數據日益增長的需求，為開發中國家的勞工創造了新的就業機會。