Today, AI is advancing so fast that it beats people in many tasks. Concerns are raised whether AI will take our jobs in the future. In this paper, I reviewed recent AI progress and potential. Then I discuss AI’s impact on employment, including employment rate, wages, employment structure, and the nature of work. My opinion is AI will definitely cause technological unemployment, but not all jobs at risk will actually be substituted. Besides, AI will raise the wages of people who complement AI but reduce overall average wages, increasing inequality. Middle-class employment will decline, pushing workers down the occupational ladder or even out of work. In addition, AI will change the nature of work as well as the workers’ skills as more and more cognitive efforts are offloaded onto AI-related technologies.
I. Machines of Loving Grace: The Great AI Awakening
When John McCarthy defined Artificial Intelligence (AI) as something that enables machines to behave like humans in the Dartmouth Conference in 1956, three years before Bob Dylan started his music career, none of them would think sixty years later, an AI called Watson could converse with Dylan about the “time passes” and “love fades” themes in his songs after it reads all his lyrics in a fraction of a second. The “old-fashioned approach” of AI at that time, which requires programming very specific rules and the the-more-the-better knowledge about the world into big, clumsy computers, is extremely time-consuming, unable to be applied in fields with less explicit rules, let alone understanding lyrics.
Science fiction writers such as Isaac Asimov and Philip Dick, however, have long anticipated a future where we’re living with intelligent, emotional, or even conspiratorial machines, which act as our friends, enemies, servants, masters, or parts of our bodies, like what the new movie Ghost in The Shell has depicted. When I was a kid, I was amazed by Arcadia Darell’s “transcriber” described in Asimov’s Foundation Series. By definition, it was a speech recognition machine that could emancipate me from handwriting for homework. I always wanted one so badly, until one day, I found myself already had one even better, whose name was Siri, which cannot only transcribe my speech but also answer my questions. Nevertheless, to the futurists’ surprise, AI hasn’t gained much momentum until recently. The hypes in the popular press and investment rose and fell for several times due to the so-called AI Winters. Thanks to recent technological advances, such as the availability of big data, improved machine learning algorithms, much mightier computers, cloud computing, and sophisticated sensors, AI reaches a sweet point and is eventually awakening in the 2010s.
Today, AI is advancing so fast that you can see it making headlines every day. Because of deep learning, AI could easily identify “patterns of patterns” using its multi-layer neural networks4. It is now surpassing the human level in terms of many tasks, including image recognition, data analysis, navigation, and playing games like the Jeopardy!, chess, go, and Texas Hold’em poker. Google’s cat recognition algorithm, which has over 1 billion “synaptic” connections can figure out a “high-order human concept” with unlabeled data without supervision2. In addition to deep learning, other AI approaches such as Bayesian Program Learning are also making inroads deeper and deeper into realms that until recently only humans had occupied, such as learning to write letters from very few examples.
The implication is far-reaching. As we are stepping into the “second half of the chessboard,” AI is increasingly incorporating into nearly all industries. In hospitals, IBM’s Watson is now helping doctors diagnose cancer5. Paro, a cute seal robot with AI capacities to “learn” from user behaviors is being used in eldercare and psychological therapy39. Scientists are using machine learning algorithms to explore the ultimate nature of the universe, such as detecting gravitational waves and searching for new fundamental particles. Research and contests in robotics are in full swing in order to develop robots that could assist people in hostile environments such as seabed, space, and nuclear power stations. Lawyers are using e-discovery software to organize files and gain insights. Algorithms even can predict the Supreme Court’s decision with a high precision rate39. Not to mention the white-hot projects of self-driving vehicles that are underway in many tech giants like Google, Tesla, Baidu, and Toyota.
In addition to practical applications, AI is also revealing its potential to create artworks. Harold Cohen, a computer scientist at UCSD wrote an AI program called AARON to create beautiful artistic images. David Cope, a computer scientist at UCSC writes AI algorithms to analyze and compose music in a certain style. Similar AI music projects are in smooth progress in companies like SONY and Google.
As a consequence, concerns are raised that AI is going to be smarter than humans and replace humans in labor market, pushing more and more people out of work. Will computers be smarter than people? Jerry Kaplan, the famous computer scientist and entrepreneur at Stanford believes the answer is yes, but most likely “in limited ways39.”
II. AI’s impact on employment rate
Will AI take people’s job? Absolutely yes. It’s already happening. Call your bank and you’ll hear a sweet female voice asking you questions, trying to identify the reasons for your call based on your answers. This kind of speech recognition systems reduce the number of employees in call centers.
But while new technologies destroy old jobs, they tend to create new jobs and increase the demand for some existing jobs, a process known as “creative destruction”. For example, as the demand for AI-based products increases, the need for computer scientists who know how to incorporate AI into existing systems is also boosted. Accordingly, the demand for people who train and assess AI-related skills grows, too. “Who would have thought that there be this job category called Search Engine Optimization,” had there been no search engines in the first place, said John Markoff in an interview with me, “that’s a new kind of labor.” He is a New York Times writer who won Pulitzer Prize for his reporting about tech companies, and one of the first journalists who wrote about Google’s self-driving cars.
So, the most important question becomes, whether the speed AI creates new jobs will outpace the speed it destroys jobs. In other words, will AI create technological unemployment? Researchers fell into two opposing camps.
The optimistic camp believes AI will not bring a good deal of damage to the labor market. Most of the proponents are economists. John Keynes, who coined the concept of “technological unemployment” in his Economic Possibilities for Our Grandchildren in 1930, would agree with this view. In Keynes’s discourse, technological unemployment is “due to our discovery of means of economising the use of labour outrunning the pace at which we can find new uses for labour.” However, Keynes considered it a “temporary phase of maladjustment” and believed it would be solved within a hundred years. Actually, that is what happened in the economic history. In the past, the jobs eliminated by new technologies were usually balanced by the jobs newly created, even though the transitions were not without pain.
The pessimistic camp considers AI as a substitute for human labor. This position can trace back to the age of industrial revolution when the Luddites broke looms in the fear of being replaced by machines. Decades later, Karl Marx deemed automation of the proletariat was an indispensable trait of capitalism. In 1964, a group of Nobel Laureates wrote to the President Lyndon B. Johnson to warn him about three revolutions that could significantly impact the society. The first one is the Cybernation Revolution, in which the extremely productive system comprising of computers and “self-regulation machines” would devalue human labor, potentially putting many people in poverty. According to a quantitative research conducted by Carl Frey and Michael Osborne at Oxford University, 47% of the U.S. employment is at high risk of automation in one or two decades. Although past experience points to the optimistic direction, history will not necessarily repeat itself, as Yuval Harari, the author of Sapiens: A Brief History of Humankind said in a lecture in 2016. He warned governments to be prepared for the future “unworking class” because he foresaw a massive unemployment coming.
A famous pessimistic analogy is horses. Wassily Leontief, the Nobel Laureate argues that workers whose job is “following specific instructions” will in principle be substituted by machines and the subsequently unemployed people will not be able to find new jobs in other industries because the value of human labor is diminished by computers, just like horses became useless in almost all industries with the advent of automobiles, which were much faster, safer, more cost-effective, and therefore more valuable than horses.
To the contrary, the optimistic camp thinks the number of jobs available will grow unceasingly. MIT economists Erik Brynjolfsson and Andrew McAfee disagree with Leontief’s horse analogy because they think people are fundamentally different from horses. People can have capitals, vote for their rights, and revolt, while horses just don’t care about whether they have “jobs20.” In particular, machine is a kind of capital. Having capital means people can buy machines to compensate the lost in the value of their labor.
So, will AI create technological unemployment? My view is one of an optimistic pessimist. That means, I believe more than 47% jobs will be at risk of AI automation, but not all of them will actually be replaced in the recent decades or two. I’ll explain as follows.
- AI will produce technological unemployment. First of all, companies’ purpose is to make profits, so they prefer more cost-effective technologies. If AI saves money, it’s inevitable to use AI rather than humans. For example, Vanguard Plastics is a factory that uses AI robot “Baxter” to pack cups. The company spent $25,000 to buy it, approximately equal to a blue-collar worker’s annual salary. However, the maintaining cost is much lower than a worker’s salary. So, business owners have the incentive to use robots rather than human, as long as their jobs are indistinguishable. Secondly, although current AI usually substitutes certain tasks, not the entire job, it still suggests fewer jobs. This is a point made by economists Frank Levy and Richard Murnane in their Dancing with Robots40 and demonstrated in cases like call centers. Thirdly, just as Martin Ford argues in his Rise of the Robots, the new jobs and industries born in the creative destruction process of AI demand much less human labor5. For example, many local governments hope they can attract tech giants like Google and Facebook to build facilities in order to boost local employment, but only to find things going athwart. According to the Washington Post, Apple built a massive data center in the town of Maiden, North Carolina in 2011, only to create fifty full-time jobs. Jerry Kaplan warns in his Humans Need Not Apply a situation in which machines boost productivity in a way that no matter how expanded a warehouse is, no extra employees will be hired. That is exactly what is happening in Amazon’s fulfillment center, where thirty thousand Kiva robots are working without salary and insurance. In addition, recent jobless recoveries imply a disturbingly decoupling trend between the aggregate employment and aggregate output. In other words, we just don’t need so much human labor to produce an abundant society despite the rising inequality due to maldistribution.
- More than 47% jobs will be at risk of automation. Frey and Osborne famously claim that 47% of U.S. employment are at high risk in one decade or two. I think the figure will be higher than that. I agree that jobs that need creativity and social intelligence are not easy to be replaced by computers, but as the AI keeps evolving, it won’t take long for AIs to be able to at least act like creative and socially intelligent. The AIs that could paint or compose music I mentioned before herald their creativity. Brynjolfsson and McAfee think humans are “deeply social species” that needs human connection20. In my opinion, however, it’s just a temporary need, like all the historical needs we had before but no longer needed much today. A chatbot called Xian’er developed in Longquan Temple, a Buddhism temple in Beijing, China, is considered by many people to be very smart in social intelligence. It can converse with people in a simple but very insightful way. Many people said they were enlightened, gained delight, and saw Buddhist wisdom and life philosophy in Xian’er’s responses. So I think in principle, people don’t care about human connection. They just care about being amused and acknowledged, regardless by another human or a machine. Therefore, I think jobs that need creativity and social intelligence will also be at high risk.
- Not all jobs at risk will actually be replaced by machines. Being at risk, however, doesn’t necessarily mean being replaced. Governments have the power to intervene job market. For instance, Queen Elizabeth I refused to grant a patent to a knitting machine, concerning it would impact employment negatively22. In addition, since cost-effective and commercially valuable AI applications require massive training database and enormous efforts in R&D, it takes time and resources. For example, facial recognition algorithms achieved high scores in standard training datasets. But they cannot be used immediately in practice until being incorporated into systems comprising of peripheral support components such as cameras, computers, LAN, user interfaces, and servers. Besides, they need to be adjusted according to scenarios. In addition, they need to learn to deal with real life situations and improve accordingly. Dr. Yu Hu, the SVP of a Chinese voice recognition software company called iFlytek once said, their software didn’t do well in real life scenarios at first even though their performance in lab tests were pretty good. It took more than five years for the the software to evolve as more and more user data is collected. It’s really a long and difficult process. So the massive substitution for human labor will not happen very soon.
In conclusion, I think AI will create technological unemployment. More than half of today’s jobs will be at high risk of being replaced by machines. But it will not happen very soon.
III. AI’s Impact on wages
One of AI’s greatest capabilities is predicting the future. Amazon provide you with recommendations based on your previous purchases simply because their AI algorithms predict you have a high probability to buy those relevant goods in the future. AI’s prediction ability is so superior and scalable that it will decrease the prize of the goods and services relying on prediction.
Then, there will be two economic implications, according to Ajay Agrawal, Joshua Gans and Avi Goldfarb in their The Simple Economics of Machine Learning. The first implication is that prediction will be used in tasks that don’t need prediction in the past, while the second one is the value of the things that “complement prediction” will increase. So, the income of the people who have the skill to complement AI will rise. For example, developers of AI-related applications, robotics engineers, and data scientists with machine learning skills are among the highest-paid occupations today. Another valuable skill is human judgment. At the same time, the wages of the people whose skills are at high risk of automation will plummet.
Another trend is that job opportunities and the wages of middle-income people will drop, as David Autor put it in 2000. In the most recent recession, a lot of people in well-paying positions have lost their jobs, while 3/4 of the newly created jobs are in the “low-paying sectors38.” Hence, the overall average wage will decrease.
In conclusion, AI’s impact on wages is a double-edged sword. On the one hand, it decreases the value of the jobs at risk of automation; on the other hand, it increases the value of the jobs that complement AI. Collectively, the low-income population grows while the average wage decreases, increasing inequality.
IV. AI’s impact on employment structure
AI will not only cause technological unemployment but also change the employment structure.
The first impact on employment structure is the “job polarization”, as MIT economist David Autor put it. It’s already happening in the last thirty years due to the computerization and automation of routine works and international trade, especially offshoring. The employment of high-skill-high-income jobs and low-skill-low-income jobs both increase, while the ones in the middle decrease, a phenomenon called middle-class “hollowing out.” AI has the same consequences. For example, some AI is now helping lawyers sort through and get insights from a huge amount of files, while others assisting oncologists in cancer diagnosis, reducing the demand for solicitors and clinicians, which are both middle-class jobs that were thought to require high educational attainment and great human intelligence. Although now many AIs are not competent enough to work alone and still need humans “in the loop”, I believe it’s just a matter of time for many AI applications to reach a fully automated level.
Another impact is the descending of labor through “occupational ladder22.” High and middle skilled workers begin to take low-skill jobs because their previous jobs were computerized, pushing low skilled workers “even further down”, or even out of work.
I think there’s more than that. Lewis Mumford said there are two kinds of technologies—authoritarian and democratic. Both of the two opposing features are observed in AI.
- Authoritarian AI: A good AI needs a huge amount of data to train, so it requires massive infrastructures, which only can be built by big companies and governments. Unless Bayesian approach that need less training data enters the AI mainstream, this trend can cause centralization of capital and power, potentially reducing intergenerational social mobility.
- Democratic AI: There are several movements in AI industry such as open source, cloud computing, and distributed computing, all of which are decentralizing and enable average people to access the platforms built by tech giants. A college student can get an API from Google’s cloud and use TensorFlow—Google’s open source software library for machine learning to do a lot of things. It will certainly increase social mobility.
However, these movements are commercial operations aiming at future profit, not inherent qualities of AI. Only big companies can build colossal cloud servers and open source platforms. Furthermore, data and programs running on the platform are assimilated into the platforms, making the powerful companies even more powerful. So, it’s hard to say which direction AI will take.
In brief, AI will change the employment structure, pushing people down the occupational ladder, hollowing out middle-class jobs. AI is more likely to thrive in big companies and will empower them with much more muscle.
V. AI’s Impact on the nature of work
Besides changing employment structure, AI will also change the nature of work, because they are more likely to substitute certain types of tasks.
First of all, jobs will be more and more non-routine. That is to say, there will be fewer and fewer tasks that “follow explicit rules”. Task framework approach was proposed by David Autor, who divided jobs into four types with two dimensions: routine-nonroutine and cognitive-manual. According to his observation, routine jobs are much easier to be replaced by computers, while non-routine jobs, no matter cognitive or manual, are at lower risk of automation. It also agrees with Deloitte economists’ observation on U.K.’s employment data from 1992. I think this trend of non-routinization will go on with AI-related technologies in the future.
In addition, many workers will be deskilled by AI. Deskilling accompanied all the way with automation from the Industrial Revolution. At the beginning of the 20th century, deskilling even became the goal of industrial production. In his Design Interaction, Donald Norman reasoned that cognitive artifacts on which we offload mental efforts could merely change “the nature of task performed by the person“ instead of improving an individual’s capacities. What is even worse, they can even weaken people’s capacities. As more and more tasks are delegated to machines, the things left to humans need less and less mental power. Workers could possibly lose their skills due to lack of practice. In his The Glass Cage, Nicholas Carr cited several aviation accidents due to pilots’ rusty manual skills degraded by autopilot system. He insightfully distinguishes automatization from automation and emphasized the importance of the former, which internalizes skills into muscle memories in stead of computer programs. Therefore, like Karl Marx, Carr thinks automation will deprive workers of autonomy and independence, becoming “a barrier to higher thought rather than a spur to it.” I guess Carr thinks there’s something special and irreplaceable in human’s work, just as John Searle thinks there’s something special in human’s cognition.
However, not everyone thinks AI-related-deskilling is necessarily a bad thing. English mathematician and philosopher Alfred Whitehead considered automation a benign thing, a way of liberating, enabling us to pursue things that “requiring greater dexterity, richer intelligence, or a broader perspective38.” In addition, the nature of work keeps transforming from the dawn of civilization. A hunter-gatherer thousands of years ago might concern her descendants’ hunting skills be degraded by the technology of raising domesticated livestock. She might worry that in an emergency situation such as a plague of livestock, people would have lost the skill to hunt wild animals for food. But raising livestock can support much more people with limited resource than hunting. One valuable skill may be considered useless in the future, for instance, the skill to drive an aircraft or a car, or the Luddites’ exquisite craftsmanship to manually produce fabrics without Jacquard looms. That’s just how history works. It’s inevitable.
In my opinion, it doesn’t matter which camp is correct. The relationship between AI and workers will become tremendously complicated. The consensus is that AI will definitely change the skill and role of human workers. Just like James Bright said in his book Automation and Management, because of automation, a metalworker’s role had transformed from a “machinist” to a “machine operator”, and then a “patrolling”, and then “a sort of watchman, a monitor, a helper.” As Carr put it, computer has a dual role—an enforcer as well as emancipator38. I think AI will serve these two roles in the future as well.
So, what work will be left to humans? Moravec’s paradox famously states that what’s easy for humans is hard for AI and what’s hard for humans is easy for AI. Frey and Osborne showed three bottlenecks for AI—perception and manipulation tasks, creative intelligence tasks, and social intelligence tasks. Likewise, Brynjolfsson and McAfee think the skills of ideation, large-frame pattern recognition, and complex communication are challenging for AI11. Levy and Murnane showed two types of work that is very hard for AI—high skilled unstructured tasks and low skilled work, such as folding towels. So they think future labor market will focus on three types of work: solving unstructured problems, working with new information, and carrying our non-routine manual tasks.
As I mentioned in the first part, I think AI is not necessarily incompetent in creative and socially intelligent tasks. Before AI figures out how to produce really artistic works, there’s still much room for humans. Even if AI can create very artistic works, we still can spend our extended leisure time on creating arts, since we’ll have plenty of leisure then and no one would care about whether machines’ works are better than ours. After all, we don’t do that for money. Basic income project will take care of us, right? And the boundary between human and machines will “blur into irrelevance”, as Kaplan put it.
In conclusion, AI will have huge impact on employment.
- AI will cause technological unemployment. More than half jobs will be at high risk of automation but not all of them will actually be automated. It depends on government regulation and the commercialization of AI applications.
- The wages of people who can complement AI will raise, but because of the increase of low-income employment, the overall average wages will drop.
- High-skill-high-income jobs and low-skill-low-income jobs will increase while middle-class jobs will go on “hollowing out.” Middle-class workers will move down the occupational ladder, pushing low-income workers out of job.
- AI will centralize power and authority, rising inequality.
- AI will change the nature of work. There will be less and less routine work.
- AI will change or degrade workers’ skills. Human’s autonomy and independence will be in danger.
There are many questions we should ask as soon as possible. For example, what should government do to deal with AI-related technological unemployment? Do we need human “in the loop” when engineers designing AI systems? How to deal with the increasing inequality caused by AI? What do we do with the extra leisure time caused by unemployment and underemployment? Should we change our modern Puritan work ethic24?
Just as Brynjolfsson and McAfee said, horses can’t give us any answer, neither do machines. The answer can only come from the “the goals we set for technologically sophisticated societies and economies we are creating and the values embedded in them20.”
 Funny or Die. 2017. Bob Dylan’s New IBM Commercial (Extended Version). Accessed April 2. https://www.youtube.com/watch?v=wtyORRJ_gzM.
 Lewis-kraus, Gideon. 2016. “The Great A.I. Awakening.” The New York Times, December 14. https://www.nytimes.com/2016/12/14/magazine/the-great-ai-awakening.html.
 Asimov, Isaac. 1964. Second Foundation. Avon Science Fiction, N306. New York: Avon.
 Executive Office of the President. 2016. “Preparing for the Future of Artificial Intelligence.” https://obamawhitehouse.archives.gov/sites/default/files/whitehouse_files/microsites/ostp/NSTC/preparing_for_the_future_of_ai.pdf.
 Ford, Martin. 2016. Rise of the Robots: Technology and the Threat of a Jobless Future. Reprint edition. Basic Books.
 “Using Big Data to Analyze Images, Video Better than the Human Brain.” 2017. ScienceDaily. Accessed April 2. https://www.sciencedaily.com/releases/2017/03/170320090439.htm.
 Kasparov, Garry. 2017. “The Chess Master and the Computer.” The New York Review of Books. Accessed March 24. http://www.nybooks.com/articles/2010/02/11/the-chess-master-and-the-computer/.
 Koch, Christof. 2017. “How the Computer Beat the Go Master.” Scientific American. Accessed February 25. https://www.scientificamerican.com/article/how-the-computer-beat-the-go-master/.
 “This AI Can Supposedly Beat Experts at No-Limit Texas Hold’em Poker.” 2017. The Verge. January 10. http://www.theverge.com/2017/1/10/14220578/ai-deepstack-beats-poker-pros-no-limit-texas-hold-em.
 Lake, Brenden M., Ruslan Salakhutdinov, and Joshua B. Tenenbaum. 2015. “Human-Level Concept Learning through Probabilistic Program Induction.” Science 350 (6266): 1332–38. doi:10.1126/science.aab3050.
 Brynjolfsson, Erik, and Andrew McAfee. 2014. The Second Machine Age: Work, Progress, and Prosperity in a Time of Brilliant Technologies. First edition. New York: W.W. Norton & Company.
Wang, Jieshu. 2016. “机器学习：引力波探测的幕后功臣.” Jieshu’s Blog. February 29. http://wangjieshu.com/2016/02/29/p7jxlb-1p-2/.
Castelvecchi, Davide. 2015. “Artificial Intelligence Called in to Tackle LHC Data Deluge.” Nature News 528 (7580): 18. doi:10.1038/528018a.
 “AARON.” 2017. Wikipedia. https://en.wikipedia.org/w/index.php?title=AARON&oldid=758645389.
 “AI Makes Pop Music in Different Music Styles.” 2016. Flow Machines. September 19. http://www.flow-machines.com/ai-makes-pop-music/.
 Wang, Jieshu. 2017. “The Semiotics of Music: From Peirce to AI.” Jieshu’s Blog. February 15. http://wangjieshu.com/2017/02/15/semiotics-of-music/.
Wang, Jieshu. 2016. “Human or Machine, Who Will Dominat the Future? An Interview with John Markoff.” Jieshu’s Blog. January 21. http://wangjieshu.com/2016/01/21/p7jxlb-1j/.
 Keynes, John Maynard. 1931. “Economic Possibilities for Our Grandchildren (1930).” Essays in Persuasion, MacMillan. http://gutenberg.ca/ebooks/keynes-essaysinpersuasion/keynes-essaysinpersuasion-00-h.html#Economic_Possibilities.
 Kolbert, Elizabeth. 2017. “Our Automated Future.” The New Yorker. Accessed March 20. http://www.newyorker.com/magazine/2016/12/19/our-automated-future.
Brynjolfsson, Erik, and Andrew McAfee. 2015. “Will Humans Go the Way of Horses?” Foreign Affairs, June 16. https://www.foreignaffairs.com/articles/2015-06-16/will-humans-go-way-horses.
 Pauling, Linus, and Ad Hoc Committee for the Triple Revolution. 1964. “The Triple Revolution.” http://scarc.library.oregonstate.edu/coll/pauling/peace/papers/1964p.7-06.html.
 Frey, Carl, and Michael Osborne. 2013. “The Future of Employment: How Susceptible Are Jobs to Computerization?” Oxford University. http://www.oxfordmartin.ox.ac.uk/downloads/academic/The_Future_of_Employment.pdf.
 Wang, Jieshu. 2016. “21世纪将是史上最不平等的世纪.” Jieshu’s Blog. April 22. http://wangjieshu.com/2016/04/23/yuval-harare-new-inequality/.
Leontief, Wassily. 1983. “National Perspectives: The Definition of Problems and Opportunities.” In The Long-Term Impact of Technology on Employment and Unemployment: A National Academy of Engineering Symposium, June 30, 1983. National Academies.
 Knight, Will, and Jieshu Wang. 2016. “Baxter, the Blue Collar Robot That Could Transform Manufacturing.” In 科技之巅:《麻省理工科技评论》50大全球突破性技术深度剖析, 第1版. 人民邮电出版社.
 “Cloud Centers Bring High-Tech Flash but Not Many Jobs to Beaten-down Towns.” 2017. Washington Post. Accessed April 3. https://www.washingtonpost.com/business/economy/cloud-centers-bring-high-tech-flash-but-not-many-jobs-to-beaten-down-towns/2011/11/08/gIQAccTQtN_story.html.
 Kaplan, Jerry. 2015. Humans Need Not Apply: A Guide to Wealth and Work in the Age of Artificial Intelligence. New Haven, US: Yale University Press. http://site.ebrary.com/lib/alltitles/docDetail.action?docID=11083687.
 Jaimovich, Nir, and Henry E. Siu. 2012. “The Trend Is the Cycle: Job Polarization and Jobless Recoveries.” Working Paper 18334. National Bureau of Economic Research. doi:10.3386/w18334.
 Autor, David H. 2015. “The Paradox of Abundance: Automation Anxiety Returns.” In Performance and Progress: Essays on Capitalism, Business, and Society. Oxford University Press, Subramanian Rangan (ed.).
狄雨霏2016年4月28日. 2016. “来北京龙泉寺，和‘贤二机器僧’谈谈人生.” 纽约时报中文网 国际纵览. April 28. http://m.cn.nytimes.com/technology/20160428/t28chinamonk/.
 Agrawal, Ajay, Joshua Gans, and Avi Goldfarb. 2016. “The Simple Economics of Machine Intelligence.” Harvard Business Review. November 17. https://hbr.org/2016/11/the-simple-economics-of-machine-intelligence.
 Autor, David. 2010. “The Polarization of Job Opportunities in the U.S. Labor Market: Implications for Employment and Earnings.” In Discussion Paper Series (Hamilton Project); Washington, 0_3,1-40,42. Washington, United States: Brookings Institution Press. http://search.proquest.com.proxy.library.georgetown.edu/docview/845241447/abstract/4ADA7107F54E4A39PQ/1.
 Beaudry, Paul, David A. Green, and Benjamin M. Sand. 2013. “The Great Reversal in the Demand for Skill and Cognitive Tasks.” Working Paper 18901. National Bureau of Economic Research. doi:10.3386/w18901.
 Winner, Langdon. 2010. “Do Artifacts Have Politics?” In The Whale and the Reactor : A Search for Limits in an Age of High Technology, 19–39. Chicago, US: University of Chicago Press. http://site.ebrary.com/lib/alltitles/docDetail.action?docID=10402621.
 Autor, David H., Frank Levy, and Richard J. Murnane. 2003. “The Skill Content of Recent Technological Change: An Empirical Exploration.” The Quarterly Journal of Economics 118 (4): 1279–1333.
 “Technology and People: The Great Job-Creating Machine | Deloitte UK.” 2017. Deloitte United Kingdom. Accessed February 12. https://www2.deloitte.com/uk/en/pages/finance/articles/technology-and-people.html.
 Norman, Donald A. 1991. “Cognitive Artifacts.” In Designing Interaction, 17–23. New York: Cambridge University Press.
 Carr, Nicholas G. 2014. The Glass Cage: Automation and Us. First edition. New York: W.W. Norton & Company.
 Kaplan, Jerry. 2016. Artificial Intelligence: What Everyone Needs to Know. What Everyone Needs to Know. New York, NY: Oxford University Press.
 Levy, Frank, and Richard J. Murnane. 2013. “Dancing with Robots: Humans Skills for Computerized Work.” Third Way. http://content.thirdway.org/publications/714/Dancing-With-Robots.pdf.