A WORLD WITHOUT WORK
DIGITIZATION, DIGITALIZATION, AND THE AUTOMATION OF WORK
The displacement effect of technological disruption is always easier to predict than the compensation effect. That is because most new technologies are invented in order to generate cost efficiencies in the present, which is generally quite well understood. On the other hand, predicting how these technologies will generate a need for “new work” is a lot harder, as there are too many unknowns to calculate. This explains why most of the economic analysis on the forthcoming Fourth Industrial Revolution has concentrated on jobs lost rather than jobs gained. Easier does not, however, imply “easy”: predicting which jobs will be automated—or”displaced”—by intelligent machines, and how quickly this will happen, remains a difficult task wrought with many assumptions and oversimplifications and dependent on the methodological approach economists adopt in making their predictions. Many economist think that the displacement impact of automation will range from considerable to devastating, as, for example, in the famous 2013 paper by Oxford economists Frey and Osborne that predicted a loss of 47% of jobs in the United States by 2023. Using a related methodology, McKinsey put the same number at 45%, while the World Bank has estimated that 57%of jobs in the OECD member states will be automated over the next two decades.
Robot Pipe by Thor_Deichmann (Pixabay License / Pixabay)
Work automation not only reduces the number of jobs available but also negatively impacts wages. Research by economists Daron Acemoglu and Pascual Restrepo looked into US labor markets and estimated that one more robot per thousand workers reduces the employment-to-population ratio by up to 0.34% and wages by up to 0.5%. This seems peculiar: technological innovation has historically boosted human productivity, and one would expect that robots and AI would do the same now and that wages will rise. Indeed, a report by the consultancy firm Bain estimated that productivity would rise by 30% across all industries because of automation. And yet, if one combines these two findings, it looks as if the providers of work will not be sharing in the new bounty of automation. Wages will be depressed despite an increase in productivity. This result is even more puzzling if one factors in demographics: the number of workers will decrease in the next decades both in the West and in China, Korea, and Japan. And yet, the decrease in the supply of labor does not seem to lift wages either.
To understand this worrying trend, it is important to distinguish between digitization and digitalization and then proceed with analyzing how digitalization shifts the balance of contribution in economic value creation from human workers to software systems, that is, from labor to capital. Digitization is the process of rendering a physical object in a digital form as zeroes and ones. Scanning a paper document and recording a voice and storing the recording as a digital file are examples of digitization. Digitalization replaces methods of sharing information in a process with computer instruction code. For example, a process whereby a clerk would review a number of paper documents in multiple binders and then walk to the office of her supervisor, three floors up, to wait outside his door and ask for approval, could be digitalized using computer programs. One often needs the digitization of physical objects to happen before proceeding with the digitalization of a process. In the example of the clerical worker, digitizing paper documents and binders into digital files and folders provides the opportunity to develop computer code that digitalizes the process of review and approval using telecommunications instead of physically climbing stairs and waiting for doors to open. Digitalization “automates” tasks in a process, and often the whole process, reducing cost, increasing efficiency, and boosting productivity. Process tasks need not be trivial. For instance the digitization of radiology scans allows for a computer, instead of a medical expert, to perform diagnosis—that is, it allows for the digitalization of the diagnostic process. Digitalization’s other consequence is that it requires humans to acquire new skills in order to compete in the new, “automated” world. In the example of the clerk, the ability to access a computer system and manipulate digital files, use email, and so on are the new skills she needs to retain her job. The fact that the process no longer needs the clerk’s physical presence in the workplace is another profound consequence of process digitalization. Workers and workplaces can be disentangled. The clerk can work remotely and perform the new process. She can be anywhere in the world.
Given the interplay of digitization and digitalization, robots do not replace humans directly; it is the process of digitalization that does so in very unpredictable ways. The roboticist Rodney Brooks gives the example of the human toll collector. Developing a robot that would do exactly the same job is hard and inefficient. The dexterity required to reach out and meet the outstretched arm of a driver to collect coins and notes in a windy environment is hugely challenging for a robot with present-day technology. Nevertheless, the human toll worker can be replaced by digitalizing the toll-paying process. For this to happen a number of innovations have to occur. The car must acquire transponder capabilities and transmit digital information regarding ownership; credit cards have to be digitized so they can be charged without the need of physical contact; wages need to be credited digitally into a bank account connected to a credit card, and so forth. Digitalization is therefore an emergent phenomenon that is virtually impossible to predict, because it is the unpredictable combination of technological innovations.
WHEN HUMANS ARE NO LONGER NEEDED
Automation technologies are advancing apace. There can be no doubt that they will impact the current economic model of growth in a profound and unprecedented way. It is not necessary for AI to reach, or overtake, “human-level” intelligence, or for the robots to become as”dexterous” as human beings, for human work to be replaced almost entirely by machines. Digitalization does not replace human workers by simulating how humans perform work tasks; it does so by reconfiguring how work is done in a process. This reconfiguration can be significant and replace many tasks that we currently regard as uniquely human. Take, for example, how machine-learning systems diagnose cancer by looking into medical images. They do not do so by emulating human experts. Instead, they do so by scanning thousands of medical images and building their own internal ways of drawing inferences from data correlations. They do not need to understand what they do or why they do it. Their “intelligence” is different from ours—it has no semblance of consciousness or self-awareness—but none of that is important. What is important is that intelligent machines can do diagnosis better than human experts, do it faster, and do it at a scale. Also important to note is that a machine is a capital asset. In hospital systems we have capital assets, such as buildings and equipment, and human resources, such as doctors, nurses, orderlies, janitors, and so on. Given what cognitive technologies are capable of, it is not too hard to imagine a hospital system that is highly digitalized and where markedly fewer human resources are needed. In that future hospital the ratio between capital assets and labor will have shifted significantly in favor of the former. In fact, the future hospital may be completely made of software.
The compensation effect has always created demand for new work during past industrial revolutions. But given the exponential rate by which automation technologies improve themselves, we may be reaching a tipping point in history where most of the new work created by AI will be done by AI! Take, for example, Uber drivers: a software-based platform in combination with satnav systems and the digitization of payments has delivered a high degree of digitalization of the process of finding a taxi when you need it. As a result, many microtasks of the process have been automated: as a passenger I do not need to step out onto the street and wait in hope for a passing cab, a taxi driver does not need to know all the streets and alleys of a city to get me to my destination, I don’t have to dig into my pockets for bills and coins to pay the driver, and so forth. This “first wave” of digitalization due to AI has created some new jobs for humans, albeit low-paying ones. Nowadays, almost anyone can bean Uber or a Lyft driver, and many have done exactly that, either as a means of supplementing their income or as their main job. Who doubts, however, that the next wave of the digitalization of the taxi process will be the total elimination of human drivers and their replacement by powerful AIs?
In fact, few people do. According to research by Pew, “most [citizens]believe that increasing automation will have negative consequences for jobs” and “relatively few predict new, better-paying jobs will be created by technological advances.” Citizens are clearly aware of the danger that AI-powered automation will destroy their livelihoods, and the data back up their perception. Research on US Department of Labor data by Axios revealed that three-quarters of US jobs created since the 2008–2009 financial crisis pay less than a middle-class income economy is on a record-breaking streak of adding new jobs into the labor market, professions that were once the backbone of the middle class have been vanishing. This “hollowing out” phenomenon is reflected in the current composition of the workforce, which is concentrated mostly on high-skill, high-wage and low-skill, low-wage jobs. Middle-skill, middle-wage jobs are vanishing. Manufacturing—because of the degree of automation already in place—is a good proxy indicator for the future of work, as technological disruption is already impacting more middle-skill, middle-wage jobs across all industries. According to data by the Federal Reserve, 25%of jobs have disappeared from manufacturing in the last two decades.Given the exponential rate of technological change, we should expect a much higher percentage of hollowing out of the labor market in a much shorter time frame.
The Fourth Industrial Revolution could lead to the massive digitalization of business processes across every industry. If that happens, human work—as we know it—could become unrecognized. Full-time jobs would become a rarity, and most of us would be working part-time for a wide variety of clients. Meanwhile, society would be reaping handsome dividends rom this new transformation of the economy. Machines will be producing goods at a very low cost in fully automated factories—or indeed much closer, in our homes, using 3D printing. Digital assistants powered by AI will be our doctors, financial advisors, and personal agents looking after every aspect of our lives. Innovation will be accelerated across every industry. We can imagine AI systems used by scientists to synthesize new materials that mimic plants and make super efficient solar panels. Solving our planet’s energy and environmental problems at a stroke, we can proceed by using AI to optimize food production and every other process too, so that human productivity rises exponentially. In such a scenario we could, theoretically, fulfill most of our material needs without the need to work and, in effect, return to an era of economic abundance. But there is a caveat.
Unlike our prehistoric ancestors, we now live in highly complex societies with sophisticated power structures and institutions, where wealth redistribution is largely controlled by governments and whoever has the most influence on their decisions. Therefore, a key political question about the Fourth Industrial Revolution is how governments will enact policies that encourage the widest possible sharing of the economic bounty that intelligent machines will deliver to the economy. The risk of not sharing the spoils of the AI economy in a democratic way is rather obvious: ordinary citizens whose only way to acquire income and wealth is through work would sink into a spiraling abyss of abject poverty. Such an outcome would most certainly destroy the last vestiges of trust in liberal democracy. Democratic governments would therefore need to step in and manage the impact of automation on work. They would also need to deal with many other risks and challenges that AI brings, including, the ethics of AI algorithms and the use of personal data in business and government. Governments would need to adopt AI systems in their processes, in order to improve their own performance and deliver better services to citizens. But how far can we—or should we—go with the automation of government? And what should the role of government be in an automated future? Given the potential of AI to automate many cognitive processes at very low cost, as well as its superhuman power for prediction and strategic planning once it processes vast amounts of data, can we imagine a fully automated government, one that is run by intelligent machines rather than human civil servants and politicians? Does it make sense to relinquish control of our economies to intelligent machines with full autonomy? The next chapter will examine in more detail such questions, as well as the different approaches between liberal democracies and authoritarian regimes in shaping the future of AI.
* * *