内容来源：https://www.wired.com/story/i-tried-doordashs-tasks-app-and-saw-the-bleak-future-of-ai-gig-work/

内容总结：

近日，外卖平台DoorDash推出一款名为“Tasks”的全新应用，旨在通过众包形式收集人类日常活动的视频数据，用于训练人工智能模型及人形机器人。用户只需将手机固定在胸前，录制自己完成各类任务的视频即可获得报酬，标志着零工经济正悄然向AI训练数据领域延伸。

该应用目前提供家务整理、手工制作、烹饪、场景导航及外语对话等五大类任务。例如，用户需拍摄自己折叠衣物、煎鸡蛋或探索公园等过程，视频要求双手全程清晰入镜。平台根据任务复杂度预付报酬，时薪约为15美元，但单次任务收入普遍较低——完成一次洗衣录制仅获约0.37美元。

然而，这种模式面临隐私与实操挑战。平台虽明确规定不得拍摄未成年人、私人场所及未经许可的他人，但用户在公园等公共场所录制时仍难免触及伦理边界。有体验者表示，为避免拍到路人，不得不中途放弃任务。

目前该服务已在美国部分州上线，但明确禁止加州、纽约市等地居民使用。业界分析指出，此类以AI训练为目的的微任务或将成为零工经济新形态，但其低廉报酬与数据收集的敏感性仍引发广泛讨论。若机器人时代终将到来，人们或许期待它们能开出更体面的“工资”。

中文翻译：

iPhone摄像头的闪光灯照亮了我脏兮兮的袜子和内裤——我正将每件衣物举到镜头前，确保视频能清晰记录。当我把散发着汗味的衣服塞进洗衣机时，手机每次发出刺耳的"哔"声都让我心头一颤，这提示我的手可能超出了拍摄范围。"必须拍到手指！"别误会，我并非为了生计转型拍摄特殊癖好内容，而是在体验外卖平台DoorDash最新推出的零工应用"Tasks"。

这款与送餐毫无关联的新应用，核心是采集人类行为数据来训练生成式AI和人形机器人。"这些数据能帮助AI和机器人系统理解物理世界，"DoorDash在新闻稿中宣称，"报酬将根据任务难度和复杂度预先显示。"多数任务要求将手机固定在胸前，录制双手完成特定操作的视频。

此类视频资料可助力AI模型与机器人开发者优化系统性能。例如，成千上万段清晰记录双手叠衣过程的视频，能通过计算机视觉技术教会机器人完成相同家务。

DoorDash计划未来拓展更丰富的任务类型和用户群体。目前该应用在美国的具体开放范围尚不明确——加州、纽约市、西雅图和科罗拉多州的居民明确被禁止使用Tasks服务（我本人在堪萨斯州成功使用并完成了任务）。

出于好奇，我注册成为"任务员"并下载了Tasks应用。登录后的首个引导任务是拍摄自己将三件物品移过桌面的视频。简单！我打开摄像头，将咖啡杯、钢笔和笔记本电脑从书桌一端推到另一端。完成后的奖励并非现金，而是DoorDash寄来的免费手机胸戴支架，以便我完成更多任务。

通过新手引导后，我看到了完整的任务清单。当前应用内的零工主要分为五大类：家务劳动、手工项目、烹饪料理、地点导览和外语对话。

每类任务涵盖范围颇广：家务清单包含铺床、使用洗碗机、移植盆栽到倒垃圾等；手工项目从换灯泡到浇筑混凝土难度不一；烹饪任务基本围绕鸡蛋展开——煎蛋、水波蛋、炒蛋；导览任务包括探索博物馆和绕行公寓小区；语言类任务则要求用俄语、汉语普通话等语言进行"自然对话"。

每项任务都有具体要求和DoorDash制定的通用准则：禁止录制未成年人、个人隐私或非法内容，拍摄他人前必须征得同意。平台还列出了医院、学校、监狱、机场和军事基地等禁止拍摄的场所。

我尝试的首个任务是拍摄将待洗衣物放入洗衣机。这堆攒了一周的脏衣服着实不少。该任务需要胸戴支架（当时尚未寄达），我只能横握手机，用空着的手抓取衣物。每件衣服都需单独拎起在镜头前展示，再投入洗衣机。

该任务时薪15美元，最长耗时20分钟。操作虽简单，但每当衣物遮挡手指触发手机提示音时总令人烦躁。即使放慢动作，我也仅用一分半就完成了10件衣物的拍摄。应用预估这段视频价值0.37美元。

见识过琳琅满目的鸡蛋任务后，我决定尝试其中一项。这类任务报酬与洗衣任务相同，要求全程清晰展示双手和鸡蛋，且从打蛋到烹制必须连续拍摄。"最后需保持成品稳定展示，"说明中写道。即便拖到鸡蛋烧焦，最高收益也不过5美元。完成后我三口吞掉了这个实验品。

时至午后，在屋里为AI"老板"打了一天工后，我决定呼吸新鲜空气，于是选择了时薪15美元的"探索公园"导览任务（同样限时20分钟）。

离家两个街区外有片绿草如茵的公园，常有人打网球或遛狗。我握着手机信步而至，踏上步道便开始录制，将手机滑入衬衫胸袋。按指示拍摄地标景物，又在岔路口驻足停顿。

尽管公园空旷，但避免拍到享受阳光的游人仍让我显得形迹可疑。当一位母亲推着婴儿车慢跑靠近时，我仅录制五分钟便中止了任务。在近乎无人的公园尚且难以遵守拍摄规范，若在酒店大堂或博物馆等拥挤场所完成类似导览任务，用户几乎不可能遵循平台规则。

旧金山的许多开发者将这类代表AI或机器人完成的低薪临时工作视为零工经济的新形态。今年初，RentAHuman平台因宣称用AI代理雇佣人类完成实体任务而爆红——不过经我实测，该网站雷声大雨点小。

尽管生成式AI和机器人行业仍获得数十亿美元投资，我分到的蛋糕碎屑却少得可怜。完成三项任务后预估报酬不足10美元，刚够在回家路上买些鸡蛋和零食。如果未来机器人成为人类的新主宰，但愿它们能开出更体面的酬劳。

英文来源：

The flash from my iPhone camera illuminates my dirty socks and underwear as I hold each item up for the video recording to capture clearly. As I load my smelly clothes into the washer, I tremble a bit each time the phone loudly beeps, detecting that my hands may be out of frame. Gotta see those fingers! No, I haven’t pivoted to filming some kind of fetish content to make ends meet—I’m trying the latest gig work app from DoorDash, called Tasks.
The new Tasks app from food delivery app DoorDash has nothing to do with delivering food—it’s all about gathering training data from humans, that’s you, for improving generative AI models and humanoid robots. “This data helps AI and robotic systems understand the physical world,” reads DoorDash’s press release. “Pay is shown upfront and determined based on effort and complexity of the activity.” Most of the gigs involve strapping a smartphone to your chest and recording your hands performing specific tasks.
This kind of video data can be used by developers of AI models and robotics to improve performance. For example, thousands of videos of people folding laundry, with their hands clearly visible, could help teach a robot how to do the same task using computer vision.
DoorDash plans to expand this service to include an even wider range of tasks and users in the future. It’s unclear where exactly the app is available for users at launch in the US—residents of California, New York City, Seattle, and Colorado are explicitly blocked from using Tasks. (I was able to use the Tasks app and complete gigs while residing in Kansas.)
Curious about what kinds of tasks DoorDash is offering right now, I signed up to be a “dasher” and downloaded the Tasks app. After logging in, the onboarding quest was to film yourself moving three objects across a table. Easy! I turned the camera on and shifted my coffee cup, pen, and laptop from one side of my desk to the other. My reward for this wasn’t cash—DoorDash shipped a free body-mount for my smartphone camera afterward, so I could complete more gigs in the app.
After that quick onboarding session, I could see the full list of potential jobs and start making some cash. The gigs currently available in the Tasks app mainly fall into five major categories: household chores, handiwork projects, cooking food, location navigation, and foreign language conversations.
The tasks within these categories are fairly broad. The chore list includes everything from making a bed and loading a dishwasher to repotting plants and taking out the trash. The handiwork projects range from simple tasks, like changing a lightbulb, to more complex ones, like pouring cement. The cooking gigs mostly revolve around eggs: frying them, poaching them, scrambling them. Navigation gigs include exploring a museum and walking around an apartment complex. For the language-based tasks, the app requests “natural conversations” in Russian and Mandarin Chinese, as well as other languages.
Each task has its own requirements and an overarching set of rules laid out by DoorDash. These include things like not recording minors, personal data, or anything illegal, and always asking for consent before filming anyone else. DoorDash also has a list of prohibited filming locations, like hospitals, schools, prisons, airports, and military bases.
The first task I attempted was loading my laundry into the washer. I’ve been putting it off all week anyway, so there was a hefty pile of clothes to pick up. This task required a body mount, which hadn’t arrived in the mail yet, so I just held my phone in landscape mode and grabbed the laundry with my free hand. Each article of clothing needed to be individually picked up, held in front of the camera, and then dropped into the wash.
This task paid $15 an hour, with a max time of 20 minutes. It was simple, but I was annoyed each time the phone made a beeping sound, indicating that my hands weren’t visible enough in the frame. That happened sometimes when a piece of laundry was covering my fingers. Even moving slowly, I was able to load the washer with 10 articles of clothing in about a minute and a half. The Tasks app estimated that I would make $0.37 for this video.
After seeing all the different tasks regarding eggs, I needed to try one of those next. The egg tasks paid the same rate as the laundry ones. I had to make sure my hands and the eggs were completely visible the whole time, as well as record from cracked to cooked without stopping. “Hold final egg state steady,” read the instructions. Even if I stretched out this task to the point of burning the egg, the most I could earn was $5. I gobbled it down when finished.
By this time, it was the early afternoon. After being cooped up in the house all day, performing tasks for my AI gig boss, it was time to get some fresh air. So, I picked a navigation task, "exploring a park,” which also paid $15 an hour for a max of 20 minutes.
There’s a grassy park just a couple of blocks from the house, where people play tennis and walk their dogs, so I jaunted on down, phone in hand. As I reached the paved trail, I started recording and slid my phone into the breast pocket of my shirt. I pointed the camera at landmarks, as instructed, and paused at forks in the path.
Even though the park was mostly empty, I felt like a total creep as I tried to avoid recording anyone just enjoying the sunshine. When I saw a mom jogging toward me with a stroller, I abandoned the task about five minutes in and stopped recording. Since I had issues not filming people without consent in an almost empty park, it seems nearly impossible for users to follow DoorDash’s rules for similar navigation tasks located in more crowded settings, like a hotel lobby or a museum.
Many developers working in San Francisco see these kinds of low-paying, temporary jobs performed on behalf of AI or robots as the next evolution of the gig economy. Earlier this year, the RentAHuman platform went viral for allegedly having AI agents hire humans to complete physical tasks—though in my tests, the site was all hype, no execution.
While the generative AI and robotics industries are still receiving billions of dollars in investment, my slice of the pie felt like chump change. After completing three tasks in DoorDash’s app, my estimated pay was less than $10. That’s just enough to buy some more eggs and a snack on the walk back home. If the robots are our new overlords one day, I hope they pay a bit more than that.