A World Without Work

Daniel Susskind, “A World Without Work: Technology, Automation, and How We Should Respond,” Metropolitan Books, 2020, 239 pp.

This is the most important and enlightening book I read this year.  It describes our exit from “The Age of Labor,” an age which mostly distributes income through wages for work, and entering an age of Artificial Intelligence, an age where income is only earned by a few; an age where many (maybe most) will become, not just unemployed but irrelevant--displaced by machines, worldwide. 

The labor force has been shaped by millennia of technological progress. Agricultural technology birthed the farming industry, the industrial revolution moved people into factories, and then globalization and automation moved them back out, giving rise, in western nations, to a nation of services. But, until now, these reshufflings increased the total number of jobs available.  What is now looming is different: an era of technological unemployment, in which computer scientists and software engineers essentially invent us out of work, and the total number of jobs declines steadily and permanently.

The author argues that in the next hundred years, technological progress resulting from Artificial Intelligence will make us more prosperous than ever before.  Yet that progress will carry us toward a world with less paid-work for human beings.  New problems will arise.  First, the problem of inequality, of working out how to share this economic prosperity with everyone in society.  Second, the problem of political power, of determining who gets to control the technologies responsible for this prosperity and on what terms.  And third, the problem of meaning, of figuring out how to use this prosperity not just to live without work but how to live well. P. 238). 

My Notes:

Pg. 8:  At the turn of the first century AD, if the global economic pie had been divided into equal slices for everyone in the world, each person would have received just a few hundred of today’s dollars per year.  As late as 1800, the average person was no more materially prosperous than her equivalent back in 100,000 BC.  But over the last few hundred years, economic growth has soared and this growth was driven by technological progress.  Today’s total GDP per capita is about $10,720 a year (an $80.7 trillion pie shared out among 7.53 billion people) and this explosive growth, propelled by technology, occurred only in the last few hundred years.

Pg. 13f:  Economists agree that this explosive growth was propelled by technological progress.  Opinion of why it occurred in Western Europe and not in other areas toward the end of the eighteenth century, however, is divided.  One reason may be geographical: certain countries had bountiful resources, a hospitable climate, and easily traversable coastlines and rivers for trade.  Another may be cultural: people had different attitudes toward the scientific method, finance, hard work, and each other (the level of ‘trust’ in a society is said to be important).  The most common explanation of all, though, is institutional: certain states protected property rights and enforced the rule of law in a way that encouraged risk-taking, hustle, and innovation, while others did not.

Pg. 24f:  Five hundred years ago the British economy was largely made up of farms; three hundred years ago, of factories; today, of offices.  In 1962, 82 percent of Chinese workers were employed in agriculture; today, that has fallen to around 31 percent.  In 1900 most English people worked on farms or in factories.  Few could have anticipated that in the future, a single ‘health care’ organization, the National Health Service, would employ far more people than the number of men then working on all the farms in the country combined.  

Pg. 36ff:  The Hollowing Out Of the Middle-Class*:  In many economies, if you took all the occupations and arranged them in a long line from the lowest-skilled to the highest-skilled, over the last few decades you would have often seen the pay and the share of jobs (as a proportion of total employment) grow for those at either end of the line, but wither for those near the middle. 

This phenomenon is known as ‘polarization’ or ‘hollowing out.’  The traditionally plump midriffs of many economies, which have provided middle-class people with well-paid jobs in the past, are disappearing.  In many countries, as a share of overall employment, there are now more high-paid professionals and managers—as well as more low-paid caregivers and cleaners, teacher’s aides and hospital assistants, janitors and gardeners, waiters and hairdressers.  But there are fewer middling-pay secretaries and administrative clerks, production workers and sales-people.  Labor markets are becoming increasingly two-tiered and divided.   leoxx

*the middle class is the middle three-fifths of the wealth spectrum.

Pg. 37:  When economists talk about ‘skill,’ they really mean ‘level of formal schooling.’  Using pay as a proxy measure for that particular definition of skill does turn out to be a reasonable thing to do.  In any event, whether you line up jobs according to their level of pay or the average number of years of schooling that people in them have does not really matter—the hollowing-out pattern looks roughly the same.

The hollowing out of the labor market was a new puzzle.  And the canonical model that dominated economic thinking in the late twentieth century was powerless to solve it.  It was narrowly focused on just two groups of workers, the low-skilled and the high-skilled, and had no way to explain why middling-skilled workers were being phased out.  In the past decade, support has emerged for an entirely different way of thinking about technology and work, known as the ALM hypothesis.  

The ALM hypothesis is built upon two realizations.  The first of these was simple: looking at the labor market in terms of ‘jobs’ is misleading.  When we talk about the future of work, we tend to think in terms of journalists and doctors, teachers and nurses, farmers and accountants; and we ask whether, one day, people who have one of these jobs might wake up and find a machine in their place.  But thinking that given jobs are uniform, indivisible blobs of activity: lawyers do ‘lawyering,’ doctors ‘doctoring,’ and so on is unhelpful.  To think clearly about technology and work, we have to start from the bottom up, focusing on the particular tasks that people do, rather than looking from the top-down, looking only at far more than job titles.

The second realization is subtler.  It became clear that the level of education required by human beings to perform a given task—how ‘skilled’ those people were—was not always a helpful indication of whether a machine would find that same task easy or difficult. Instead, what appeared to matter was whether the task itself was what the economists called ‘routine.’  By ‘routine,’ they did not mean that the task was necessarily boring.  Rather, a task was regarded as ‘routine’ if human beings found it straightforward to explain how they performed it—if it relied on what is known as ‘explicit’ knowledge, knowledge which is easy to articulate, rather than ‘tacit’ knowledge, which is not. The ALM hypothesis is that machines can readily perform the ‘routine’ tasks in a job, but would struggle with the ‘non-routine’ tasks.  Many of the activities that middling-paid people performed in their work turned out to be ‘routine,’ whereas those done by the low-paid and high-paid were not.  That was why labor markets around the world were being hollowed out.  Technological change was eating away at the ‘routine’ tasks clustered in the middle. (It is important to note that by 2017 many ‘non-routine’ tasks are now within the grasp of machines, something previously unthinkable p. 68)

Technological progress was neither skill-biased nor unskilled-biased.  Rather, technological progress is task-biased, with machines able to perform certain types of tasks but unable to perform others. 

Pg. 50:  In 1997 IBM’s ‘Deep Blue’ beat world chess champion Garry Kasparov using vast amounts of processing power and data storage to crunch up to 330 million moves in a second.  This was the ‘pragmatist revolution:’ instead of attempting to copy the way a human thinks, they designed it to follow and entirely different route.  Since Deep Blue, machines are judged not by how they perform a task but how well they perform it.  For instance, translations are accomplished by having computers scan millions of human-translated pieces of text to figure out interlingual correspondences and patterns on their own.  The same with pictures: now computers review millions of previously labeled pictures and hunt for similarities between those and the particular photo in question.  In 2017 image identification by computer reached 98 percent accuracy, beating humans ever since 2015.

Pg. 52:  Computer algorithms today are specifically aimed at letting systems learn from their experience instead of being guided by explicit rules.  For instance, AlphaGo Zero was only given the rules of Chess and then played itself for three days to generate its own data—it then thrashed its older cousin AlphaGo.  Of course, Go and Chess are games of perfect information and do not represent the complexity of real-life.  Yet, in 2019 Pluribus was created and could defeat the best professional poker player in multiplayer competitions.   The system learned poker purely ‘from scratch,’ without any human input, just by knowing the rules and playing hand after hand against copies of itself over the course of a few days.

Pg. 56:  DeepMind, the AI company that developed AlphaGo, was bought by Google in 2014 and is now staffed by the leading minds in the AI field poached from top academic departments, with average pay packages of $345K per employee.  Their machines, such as AlphaGo are only ‘intelligent’ in the sense that they are very capable, but they do not think or reason like human beings. 

Pg. 73:  The temptation is to say that because machines cannot reason like us, they will never exercise judgment; because they cannot think like us they will never exercise creativity; because they cannot feel like us, they will never be empathic.  And all that may be right.  But it fails to recognize that machines might still be able to carry out tasks that require empathy, judgment, or creativity when done by a human being—by doing them in some entirely other fashion.  Apparently, we do not need to solve the mysteries of how the brain and mind operate to build machines that can outperform human beings.  Nor must it be true that machines are limited to what a human can do.

Pg. 79:  The three main capabilities that human beings draw on in their work: manual, cognitive, and affective capabilities are each under increasing pressure by machines. 

Pg. 80:  Driverless cars now learn to navigate themselves, from the bottom up, drawing on sensor data from millions of real and simulated test drives.  Given that, on average, one person is injured in a traffic crash somewhere in the world every second—and one is killed every twenty-five seconds—the prospect of self-driving vehicles is something to be welcomed.  A convoy of semiautonomous trucks completed their first trip across Europe in 2016, ‘platooning’ with one another: the front vehicle controlled the speed, and the others automatically followed its lead.  (During this test there was a driver in each seat).

Pg. 81:  Today robots now account for 80 percent of the work undertaken in manufacturing a car. 

Pg. 84:  Kahn Academy has about ten million unique visitors a month, a higher effective attendance than the entire primary and secondary school population of England.  (I belong to this: well worth looking into).

Pg. 85:  In 2011 a Catholic bishop issued a mobile app that is meant to help its users prepare for confession; among its assorted functions, it provides tools for tracking sins and a drop-down menu with various options for acts of contrition.  The app caused such a stir that the Vatican itself felt it had to step forward, noting that while people are allowed to use this app to prepare for confession, it is not a substitute for the real thing. (P. 85).    (Dan Gilgoff and Hada Messia, “Vatican Warns About iPhone Confession App,’ CNN, 10 February 2011).  (I am hopeful that the next generation, called Gee Whiz rather than G5, will have a Drop-Down menu with suggested sins.  Kind of a guidebook for adolescent boys who may find things on the menu they had never thought of.)

Pg. 86:  Beyond the physical world and the cognitive sphere, machines are now also encroaching on tasks that require our affective capabilities, our capacity for feelings and emotions.  This field of computer science is known as ‘affective computing.’  For example, there are systems that can look at a person’s face and tell whether they are happy, confused, surprised, or delighted.  Another machine can tell whether a person is lying I court with about 90 percent accuracy—whereas human beings manage about 54 percent, only slightly better than chance.  A Chinese insurance company uses a system like this to tell whether loan applicants are being dishonest.  There are also ‘social robots’ now used in health care.  They comfort people with cognitive disorders like dementia.

Pg. 91:  The sorts of tasks that the OECD consider to be easiest to automate are disproportionately found in poorer countries as well as developing countries.  Also, the top 1 percent of the most highly cited papers in mathematics and computer science are coming from two universities in China, ahead of Stanford and MIT.

Pg. 103:  In the 1950s and ‘60s fast-food jobs were seen as largely for ‘teens in the summertime,’ but in the US today only a third of fast-food workers are teenagers, 40 percent are older than twenty-five, and nearly a third have some college education.  A third of Americans with degrees in STEM subjects are now in roles that do not require those qualifications. 

Today, it is estimated that there are only twenty-two thousand PhD-educated researchers in the world who are capable of working at the cutting edge of AI, and only half are based in the US.

Pg. 106:  From 2000 to 2010, the US areas with the biggest fall in population (aside from New Orleans, battered by Hurricane Katrina) were Detroit, Cleveland, Cincinnati, Pittsburg, Toledo, and St. Louis.  These Rust Belt cities, traditionally dependent upon manufacturing, lost up to 25 percent of their population. 

Pg. 110:  In the US today driverless vehicles threaten 2.2 to 3.1 million jobs.

Pg. 145:  In the US today, the poorest 50 percent of Americans own only 2 percent of the country’s wealth, but the richest 1 percent, who owned a bit less than 25 percent in 1980, now own more than 40 percent.  Relative income equality began a dramatic divergence in 1980.  America now finds itself in a situation where the richest 0.1 percent (about 160,000 people) own about 22 percent of all the wealth in the country.  For the most part, economies around the world are becoming more prosperous but also more unequal.  And the main culprit in this is technological progress.

Pg. 154:  During the eighteenth and nineteenth centuries a country’s prosperity depended on its willingness to invest in traditional capital, its factories and industrial machines, but in the twentieth century that changed, and prosperity started to depend far more upon an eagerness to invest in human capital, the skills and capabilities of its workers.  New technologies increasingly required specialized skills, and countries that had better-educated workforces were better equipped to put these technologies to proper use.  Simple literacy and numeracy were no longer sufficient for economic success.  More education was needed.

Pg. 181:  In the US about forty million people already receive ‘food stamps’ to receive basic sustenance for free, worth about $1,500 a year.  And Native Americans’ have the highest poverty rate of any race (26.2 percent compared to the US national average of 14 percent), and a suicide rate that is 1.5 times the national average.  

Pg. 186:  A guaranteed basic income in a world with less work should have conditions attached, but not those from a labor-intensive society.  The author proposes a Citizen Basic Income (CBI) rather than a UBI “Universal Basic Income.”  A CBI would emphasize efforts to support the community rather than pursue employment, employment which may be non-existent in an AI world.  Under this system, those without traditional work are likely to be divided in two: not between leisure and paid work, but between activities that they choose and others that their community requires them to do. (p. 233).

Pg. 197f:  The “Big Five” technology companies today are Amazon, Apple, Google, Facebook, and Microsoft.  In the US, Google gets 62.6 percent of search engine traffic and has 88 percent of the market in search advertising, Facebook is used by almost a third of the world’s human beings, and across its various platforms (like Instagram and WhatsApp) it is the gatekeeper for 77 percent of mobile social traffic.  Amazon is the shopkeeper of choice for 43 percent of all online retail and 74 percent of the e-book market.  Apple and Google combine to control 99 percent of mobile phone operating systems.  Apple and Microsoft account for 96 percent of desktop operating systems.  In 

2018, these five companies were among the top ten most valuable in the world.  leoxx

Pg. 200:  To test driverless cars and improve their performance, Uber built an entire mock town, complete with plastic pedestrians that occasionally throw themselves into traffic, and gathers data as their cars drive around it. Tesla, meanwhile, collects data from its non-autonomous as they are driven by their owners, with about a million miles’ worth of data reportedly flowing in every hour.  Google in yet another approach to the problem, creates entire virtual worlds to gather data from cars driving around in these simulations.

Pg. 211f:  China’s social credit system:  China hopes by 2020 that all Chinese citizens will be scored and ranked according to information about them stored in a national database.  In one pilot, scores are determined by acts as trivial as getting a traffic ticket (minus 5 points) and as ambiguous as ‘committing a heroic act’ (plus 30 points).  Officials say that they want the system to ‘allow the trustworthy to roam everywhere under heaven while making it hard for the discredited to take a single step.’  (In the US we already have FICO scores).

Pg. 220:  Up until the 1960s it was thought that hunter-gatherers must have lived highly laborious lives, but recent anthropological research has shown that they probably did surprisingly small amounts of work, certainly much less than workers in the US currently do.

Addenda:

• The spread of nonworking men and underemployed youth. The share of prime-age Americans (25 to 54 years old) who are working has been trending down since 2000. Among men, the decline began even earlier: the share of prime-age men who are neither working nor looking for work has doubled since the late 1970s, and has increased as much throughout the recovery as it did during the Great Recession itself. All in all, about one in six prime-age men today are either unemployed or out of the workforce altogether. This is what the economist Tyler Cowen calls “the key statistic” for understanding the spreading rot in the American workforce. Conventional wisdom has long held that under normal economic conditions, men in this age group—at the peak of their abilities and less likely than women to be primary caregivers for children—should almost all be working. Yet fewer and fewer are.

More people are pursuing higher education, but the real wages of recent college graduates have fallen by 7.7 percent since 2000. In the biggest picture, the job market appears to be requiring more and more preparation for a lower and lower starting wage.

In 2013, Oxford University researchers forecast that machines might be able to perform half of all U.S. jobs in the next two decades. The projection was audacious, but in at least a few cases, it probably didn’t go far enough. For example, the authors named psychologist as one of the occupations least likely to be “computerisable.” But some research suggests that people are more honest in therapy sessions when they believe they are confessing their troubles to a computer, because a machine can’t pass moral judgment. Google and WebMD already may be answering questions once reserved for one’s therapist. This doesn’t prove that psychologists are going the way of the textile worker. Rather, it shows how easily computers can encroach on areas previously considered “for humans only.”

The most-common occupations in the United States are retail salesperson, cashier, food and beverage server, and office clerk. Together, these four jobs employ 15.4 million people—nearly 10 percent of the labor force, or more workers than there are in Texas and Massachusetts combined. Each is highly susceptible to automation, according to the Oxford study.