随着人工智能的不断发展，越来越多的人担心自己的饭碗会不会被机器人抢走。从翻译到秘书，许多职业面临到人工智能的威胁。在人工智能时代，我们人类以后真的就成为多余的存在了吗？
“人力云”：人工智能将为我们创造新的工作机会
译者：倪凌晖
校 对：刘 璠
策划：倪凌晖
The human cumulus
人力云
本文选自The Economist | 取经号原创翻译
关注 取经号，
回复关键词“外刊”
获取《经济学人》等原版外刊获得方法
When the first printed books with illustrations started to appear in the 1470s in the German city of Augsburg, wood engravers rose up in protest. Worried about their jobs, they literally stopped the presses. In fact, their skills turned out to be in higher demand than before: somebody had to illustrate the growing number of books.
15世纪70年代，在德国城市奥格斯堡（Augsburg）人类出版了有史以来第一批印有插图的书，而在那里的木雕艺人害怕自己的技艺在这类印刷技术问世后不再为人所需要，因此起身抗议。为了保住饭碗，他们叫停了这批印刷机器。但事实却与他们想象的刚好相反。随着图书数量的不断增加，总要有人绘制书中的插图，这时他们的技艺就派上用场了。（译者注：15世纪70年代之前的插图只有木刻的，后来发展到金属腐蚀版。）
Fears about the impact of technology on jobs have resurfaced periodically ever since. The latest bout of anxiety concerns the arrival of artificial intelligence (AI). Once again, however, technology is creating demand for work. To take one example, more and more people are supplying digital services online via what is sometimes dubbed the “human cloud”. Counter-intuitively, many are doing so in response to AI.
自那以后，人们对于科技进步会造成失业的担忧呈周期性地不断出现。最近一次就在人工智能出世之际。但是，我们再一次发现，科技进步非但不会造成失业，甚至会创造新的工作机会。例如，通过所谓的“人力云”（human cloud），越来越多的人正开始从事线上数字信息服务。许多数字信息工作都是由于人工智能才应运而生的。这与我们认为科技进步会造成失业的直觉恰恰相反。
According to the World Bank, more than 5m people already offer to work remotely on online marketplaces such as Freelancer.com and UpWork. Jobs range from designing websites to writing legal briefs, and typically bring in at least a few dollars an hour. In 2016 such firms earned about $6bn in revenue, according to Staffing Industry Analysts, a market researcher. Those who prefer work in smaller bites can use “micro-work” sites such as Mechanical Turk, a service operated by Amazon. About 500,000 “Turkers” perform tasks such as transcribing bits of audio, often earning no more than a few cents for each “human-intelligence task”.
根据世界银行提供的数据显示，已有超过五百万人通过“Freelancer.com”和“UpWork”这类线上自由职业平台从事远程工作。自由职业包括网页设计和撰写诉讼案情摘要这类的工作，大部分的工资最少也有一小时几美元。据市场调查公司Staffing Industry Analysts 透露，这类公司2016年的年收入为大约60亿美元。那些倾向于从事小额收入工作的人会用“微工作”网站，例如由亚马逊运营的Mechanical Turk。大约有五百万的Mechanical Turk使用者在从事类似转录音频这样的工作，每份任务的工资则基本不超过几美分。
Many big tech companies employ, mostly through outsourcing firms, thousands of people who police the firms’ own services and control quality. Google is said to have an army of 10,000 “raters” who, among other things, look at YouTube videos or test new services. Microsoft operates something called a Universal Human Relevance System, which handles millions of micro-tasks each month, such as checking the results of its search algorithms.
许多大型科技公司雇佣成千上万人来监督自己公司的服务，从而进行质量监控。它们大多通过外包公司来雇佣这些人。据说谷歌就拥有一支一万人的质量检测部队，专门负责浏览YouTube和测试谷歌的新服务。微软（Microsoft）则运营着一个名为“Universal Human Relevance System”的平台，每月上面都会有成千上万的“微工作”，例如测试搜索程序的结果等等。
These numbers are likely to rise. One reason is increasing demand for “content moderation”. A new law in Germany will require social media to remove any content that is illegal in the country, such as Holocaust denial, within 24 hours or face hefty fines. Facebook has announced that it will increase the number of its moderators globally, from 4,500 to 7,500.
上面提到的这些数字很可能会在未来有所上升。首先是由于网上内容审核的需求正在不断增加。德国颁布了一项新的法条：社交媒体上不得出现不合法的内容，例如大屠杀。社交媒体需要在24小时内清除非法内容，否则将会面临巨额罚金。脸书（Facebook）宣布，公司将会增加线上审核人员，将从目前的4500人增加到7500人。
AI will eliminate some forms of this digital labour—software, for instance, has got better at transcribing audio. Yet AI will also create demand for other types of digital work. The technology may use a lot of computing power and fancy mathematics, but it also relies on data distilled by humans. For autonomous cars to recognise road signs and pedestrians, algorithms must be trained by feeding them lots of video showing both. That footage needs to be manually “tagged”, meaning that road signs and pedestrians have to be marked as such. This labelling already keeps thousands busy. Once an algorithm is put to work, humans must check whether it does a good job and give feedback to improve it.
人工智能会淘汰掉一批本来由人类提供的数字信息服务。例如，在转录音频方面，软件就要比人工更加出色。但是，人工智能也能创造出其他数字信息服务。人工智能可能要用到庞大的数据处理能力和高超的数学运算，但它也需要依赖于人类为它提炼数据。例如，自动驾驶汽车之所以能够辨别路标和行人，是因为事先有人将大量的路标和行人录像输入给它。这些含有路标和行人的镜头需要人为地将它们标志出来。就单单这一项镜头挑选的工作就已经创造了数千个工作岗位。一旦一项程序投入运行，那就需要有专门的岗位来检查它是不是正常运作，并对它进行反馈从而提升程序性能。
A service offered by Crowd Flower, a micro-task startup, is an example of what is called “human in the loop”. Digital workers classify e-mail queries from consumers, for instance, by content, sentiment and other criteria. These data are fed through an algorithm, which can handle most of the queries. But questions with no simple answer are again routed through humans.
一家名为Crowd Flower的“微工作”的新企业推出了一项服务。例如，“数字民工”将客户发来的咨询邮件以内容、情绪等标准进行分类。然后，这些经过分类的邮件会由电脑程序进行处理。有部分电脑处理不了的，没有统一简单回复的咨询邮件则会反工交由人工进行回复。这可以看成是“人力云”的典型案例。
You might expect humans to be taken out of the loop as algorithms improve. But this is unlikely to happen soon, if ever, says Mary Gray, who works for Microsoft’s research arm. Algorithms may eventually become clever enough to handle some tasks on their own and to learn by themselves. But consumers and companies will also expect ever-smarter AI services: digital assistants such as Amazon’s Alexa and Microsoft’s Cortana will have to answer more complex questions. Humans will still be needed to train algorithms and handle exceptions.
你可能会认为，一旦程序得到升级，人工最终还是会被淘汰。对于这一点，在微软研究院工作的Mary Gray表示，如果真有那么一天，那也是很久以后的事，近期人工并不会被淘汰。程序可能最终会变得非常聪明，可以独立处理一些任务，并进行无人工帮助的“机器学习”。但那时，消费者和企业也会对它们提出更高的智能化要求。例如，人类会向亚马逊Alexa和微软Cortana这类的语音助手提出更复杂的问题。将来，程序仍然需要由人工来训练，并处理那些非常规的任务。
Accordingly, Ms Gray and Siddharth Suri, her collaborator at Microsoft Research, see services such as UpWork and Mechanical Turk as early signs of things to come. They expect much human labour to be split up into distinct tasks which can be delivered online and combined with AI offerings. A travel agency, for instance, might use AI to deal with routine tasks (such as booking a flight), but direct the more complicated ones (a request to create a customised city tour, say) to humans.
因此，Mary Gray和她的同事Siddharth Suri将UpWork和Mechanical Turk这类线上自由职业平台看作是一种前兆。他们认为，在未来大量的人工将分配到一个个任务制的在线岗位上，并与人工智能产品相结合。例如，一家旅行社可以使用人工智能来处理常规工作，如订机票。而人工可以用来处理更加复杂的工作，如预定个性化的城市观光服务。
Michael Bernstein and Melissa Valentine of Stanford University see things going even further. They anticipate the rise of temporary “firms” whose staff are hired online and configured with the help of AI. To test the idea, the researchers developed a program to assemble such virtual companies for specific projects—for instance, recruiting workers and assigning them tasks in order to design a smartphone app to report injuries from an ambulance racing to a hospital.
斯坦福大学的Michael Bernstein和Melissa Valentine甚至比其他人想得更长远。他们预测将来会有一种“临时性公司”出现。这类公司在网上雇佣员工，并利用人工智能来分配岗位。为了测试他们的预测，他们开发了一个程序，组建了这种虚拟的临时公司，并赋予它们特定的任务。如雇佣员工，并向他们分配工作：开发一款手机软件，用于在救护车运送伤员到医院的途中向医院汇报伤情。
Working in such “flash organisations” could well be fun. But many fear that the human cloud will create a global digital
proletariat
. Sarah Roberts of the University of California, Los Angeles, found that content moderators often suffer from
burnout
after checking
dodgy
social-media content for extended periods. Mark Graham of the University of Oxford concludes that platforms for online work do indeed offer new sources of income for many, particularly in poor countries, but that these services also drive down wages. So governments need to be careful when designing big digital-labour programmes—as Kenya has done, hoping to train more than 1m people for online jobs.
在这样昙花一现的公司工作可能会相当有趣。但也有许多人担心这种“人力云”公司会在全球产生一批收入低下的“数字工人阶级”。加利福尼亚大学的Sarah Roberts发现，那些网络审查员在长时间审查社交媒体上的违规信息之后常常感到精疲力竭。牛津大学的Mark Graham推断：线上自由职业平台确实可以为许多人带去新的收入来源，特别是在那些贫困国家。但是这些平台也拉低了工资水平。所以政府应该谨慎制定大型的数字人力项目。肯尼亚政府（Kenya）在这方面就做得很好。该国的数字培训项目有望为社会培训超过一百万位线上工作人员。
the proletariat
