[Annotated] Artificial intelligence will create new kinds of work

Humans will supply digital services to complement^[각주:1] AI

When the first printed books with illustrations started to appear in the 1470s in the German city of Augsburg, wood engravers^[각주:2] rose up in protest^[각주:3]. 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.

Fears about the impact of technology on jobs have resurfaced^[각주:4] periodically^[각주:5] ever since^[각주:6]. The latest bout of^[각주:7] 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^[각주:8] the "human cloud". Counter-intuitively^[각주:9], many are doing so in response to^[각주:10] AI.

According to the World Bank, more than 5m people already offer to work remotely^[각주:11] on online marketplaces^[각주:12] such as Freelancer.com and UpWork. Jobs range from designing websites to^[각주:13] writing legal briefs^[각주:14], 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^[각주:15] such as transcribing bits of audio^[각주:16], often earning no more than a few cents for each "human-intelligence task".

Many big tech companies employ, mostly through outsourcing firms, thousands of people who police^[각주:17] the firms' own services and control quality. Google is said to have an army of 10,000 "raters^[각주:18]" 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^[각주:19]. 

These numbers are likely to rise. One reason is increasing demand for "content moderation^[각주:20]". 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^[각주:21]. Facebook has announced that it will increase the number of its moderators^[각주:22] globally, from 4,500 to 7,500. 

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^[각주:23] and fancy mathematics, but it also relies on data distilled by^[각주:24] humans. For autonomous cars to recognize road signs and pedestrians, algorithms must be trained by feeding them lots of video showing both. That footage^[각주:25] needs to be manually "tagged", meaning that road signs^[각주:26] and pedestrians have to be marked as such. This labeling 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.

A service offered by CrowdFlower, a micro-task startup, is an example of what is called "human in the loop^[각주:27]". Digital workers classify^[각주:28] e-mail queries from consumers, for instance, by content, sentiment^[각주:29] 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^[각주:30] humans.

You might expect humans to be taken out of the loop as algorithms improve. But this is unlikely to happen soon, if ever^[각주:31], 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. 

Accordingly^[각주:32], Ms Gray and Siddharth Suri, her collaborator at^[각주:33] 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^[각주:34] 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^[각주:35] the more complicated ones (a request to create a customized^[각주:36] city tour, say) to humans.

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^[각주:37] AI. To test the idea, the researchers developed a program to assemble^[각주:38] such virtual companies for specific projects - for instance, recruiting workers and assigning them tasks^[각주:39] in order to design a smartphone app to report injuries from an ambulance racing to a hospital. 

Working in such "flash organizations" could well be fun. But many fear that the human cloud will create a global digital proletariat^[각주:40]. Sarah Roberts of the University of California, Los Angeles, found that content moderators often suffer from burnout^[각주:41] after checking dodgy^[각주:42] social-media content for extended periods^[각주:43]. 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^[각주:44]. 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. 

Technology is rarely an unalloyed^[각주:45] bane^[각주:46] or blessing. The printing press^[각주:47] created new work for the wood engravers in Augsburg, but they quickly discovered that it had become much more repetitive^[각주:48]. Similar trade-offs^[각주:49] are likely in future.

1. complement ; [타동사][VN] 보완하다, (금상첨화 격으로) 덧붙이다 [본문으로]
2. engraver ; [명사] (나무・돌・쇠붙이 등에 글자나 무늬를) 새기는[파는] 사람, 판화가 [본문으로]
3. rise up in protest ; 투덜거리며 일어나다. [본문으로]
4. resurface ; 1. [자동사][V] (수면 아래나 보이지 않게 잠재되어 있던 것이) 다시 떠오르다[드러나다] [본문으로]
5. periodically ; [부사] 정기[주기]적으로 [본문으로]
6. ever since ; ~이후로 줄곧[계속] [본문으로]
7. bout ; 1. ~ (of sth/of doing sth) 한바탕, 한차례 [본문으로]
8. be dubbed ; ~로 불리다 [본문으로]
9. counter-intuitive ; [형용사] 반(反)직관적인, 직관에 어긋나는. [본문으로]
10. in response to ; (~에) 응하여, 대응하여 [본문으로]
11. remotely ; 2. 멀리서, 원격으로 [본문으로]
12. marketplace ; 1. [sing.] the marketplace (상품・서비스 등의 경쟁이 벌어지는) 시장 [본문으로]
13. range from ... to ; 범위가 ~에서 ~에 이르다 [본문으로]
14. legal brief ; a document stating the facts and points of law of a client's case [본문으로]
15. perform[carry out] tasks ; 과제를 수행하다 [본문으로]
16. transcribe ; 2. (전문 용어) (말소리를) 음성 기호로 표기[전사]하다 [본문으로]
17. police ; 2. (위원회 등이) (규칙 준수를) 감시하다 [본문으로]
18. rater ; 1. 평가[측정]자 [본문으로]
19. search algorithm ; (IT) <컴퓨터>탐색, 검색 알고리듬(探索~) [본문으로]
20. moderation ; 2. (英) (교육에서) (시험 채점 과정 등에서의 기준) 조정[관리] [본문으로]
21. face hefty fines ; 과중한 벌금에 직면하다 [본문으로]
22. moderator ; 3. (英) (시험 채점 평가 기준의) 조정자[관리자] [본문으로]
23. computing power ; 연산력 [본문으로]
24. distill ; 2. 증류하여 <불순물을> 제거하다 ((off, out)) [본문으로]
25. footage ; [U] (특정한 사건을 담은) 장면[화면] [본문으로]
26. road signs ; (교통) 도로표지 [본문으로]
27. in the loop ; (중요한 일을 다루는) 핵심[중추]의 일원인 ;; 반의어 ; out of the loop [본문으로]
28. classify ; 1. 분류[구분]하다 [본문으로]
29. sentiment ; 1. [C , U] (격식) 정서, 감정 [본문으로]
30. route ; [타동사][VN + adv. / prep.] (rout・ing 또는 route・ing , rout・ed , rout・ed) (특정한 루트를 따라 무엇을) 보내다[전송하다] [본문으로]
31. if ever ; (《삽입적으로》) 설사 …하는 일이 있다 해도 [본문으로]
32. accordingly ; 2. [특히 문장 첫 부분에 쓰여] 그런 이유로, 그래서 [본문으로]
33. collaborator ; 1. 공동 연구자[저자], 합작자 [본문으로]
34. split up into ; …로 갈라지다, 나뉘다 [본문으로]
35. direct ; 2. CONTROL | [타동사][VN] 지휘하다, 총괄하다 [본문으로]
36. customized ; [형용사] 개개인의 요구에 맞춘 [본문으로]
37. be configured with ; ~로 구성하다 [본문으로]
38. assemble ; 1. 모이다, 모으다, 집합시키다 [본문으로]
39. assign ; 1. ~ sth (to sb) | ~ (sb) sth (일・책임 등을) 맡기다[배정하다/부과하다] [본문으로]
40. proletariat ; [sing.+ sing./pl. v.] [특히 과거에 대해 말할 때 씀] (전문 용어) 프롤레타리아[무산 노동자] 계급 ;; 참고 bourgeoisie ;; US [|proʊlə|teriət] UK [|prəʊlə|teəriət] [본문으로]
41. burnout ; [C , U] 1. (신체적 또는 정신적인) 극도의 피로 [본문으로]
42. dodgy ; (英 비격식) 1. 의심스러운, 부정직해 보이는 [본문으로]
43. extended period ; 장기간, 오랜 시간 동안 [본문으로]
44. drive down ; [이율 따위]를 억제하다, 내리다 [본문으로]
45. unalloyed ; [형용사] (격식) 다른 것이 섞이지 않은, 순수한 [본문으로]
46. bane ; [sing.] the ~ of sb/sth 골칫거리 [본문으로]
47. printing press ; 인쇄기(특히 동력 인쇄기) [본문으로]
48. repetitive ; 1. (지루할・단조로운 정도로) 반복적인 [본문으로]
49. trade-off ; [명사] ~ (between sth and sth) (서로 대립되는 요소 사이의) 균형 [본문으로]