In the future, machines will replace humans in jobs. This is not controversial: it’s what machines have done since well before the start of the industrial revolution. Petrol pump attendants were replaced by automated pumps, secretaries were replaced by Microsoft Office. This is what economists call the substitutive effect of automation: humans are substituted in jobs by machines.
From time to time, fears have been expressed that humans would run out of jobs entirely. I first wrote about this concern back in 1980, and like many other people at the time, I under-estimated the resilience of what economists call the complementary effect of automation. To simplify a little, this means that when a human job is automated, the amount of wealth generated in the economy increases. That increases demand, which in turn increases employment. The new jobs that are created are done by humans, because machines can not yet do everything that we can do for money.
The complementary effect probably won’t last forever. The amount of compute power you can buy for $1,000 continues to increase exponentially: Moore’s Law is evolving, not stopping. This means that in a decade, the machines we have will be 100 times as powerful as the ones we have today. In two decades the multiple will be 8,000, and in three decades, 1,000,000. There is no straight-line correlation between compute power and the ability of AI to carry out human tasks, but the two do travel together. People who airily dismiss fears about medium-term technological unemployment are committing the Reverse Luddite Fallacy. They are making four mistakes.
First, they forget that we have seen technological unemployment before. In 1915 there were 22 million horses working in America, pulling vehicles and working on farms. The horse population of America today is two million. That is severe technological unemployment.
Second, they argue that because automation has not caused technological unemployment in the past, it will not do so in the future. The past is often a good guide to the future, but it is far from perfect: if it was, we would not be able to fly.
Third and fourth, they are thinking too short-term, and they are not taking into account the astonishing power of exponential growth. Machines will not be able to do everything humans can do for money in the next year, nor in the next decade. It is also true that technologies are not implemented overnight. It takes time for companies to figure out how to use them. But three decades from now, when our machines are 1,000,000 times more capable than they are today, they will probably be cheaper, faster, and better at almost everything that humans can do for money.
Some people reply that we will endlessly create jobs which machines are unable to do, either because they are not creative, or because they are not conscious, and therefore have no empathy. These are likely to prove vain hopes: AI systems are creative already, and they can decode human emotions very effectively. They can also provide the same appearance of empathy that humans often do at work.
If technological unemployment is coming, we must transition to a very different type of economy. I call the transition the “economic singularity”, using a term from physics which was first applied to human affairs by computer pioneer John von Neumann. Back in the 1950s, he said we were “approaching some essential singularity in the history of the race beyond which human affairs, as we know them, could not continue.”
How should policy makers respond to the challenges presented by the economic singularity? Let’s look at a few options.
In America, some truckers’ groups want to ban the introduction of self-driving lorries. Not unreasonably, they argue that the technology will put millions of them out of work, with serious knock-on effects for their families and their communities. This could be an effective strategy if they were not in a competitive market: train drivers unions in the UK have prevented efficiency measures for years. But unlike trains, trucking is a competitive market. And even if the federal government were to wave a magic wand and give the truckers their wish, the US economy would become less competitive internationally. Protecting jobs stifles competition, kills innovation, and makes everyone poorer.
Similarly, taxing robots in order to fend off or mitigate technological unemployment would stifle both innovation and competitiveness. It also suffers from conceptual difficulties. Machines will rarely replace humans on a one-for-one basis, and most of the “robots” automating jobs in the coming years will be software, not androids. Humans will disappear from many call centres, for instance, and be replaced by AI systems running on banks of servers in big buildings with powerful air conditioning. Does the government tax one entity – namely the AI system - or each individual server? Or does it estimate the number of humans who have been laid off by the system and calculate the tax based on that?
If we are going to tax machines which automate jobs away, do we start with tractors and weaving machines? Or do we just tax new instances of replacement? Imagine two firms making the same product. One has been going for a few years, and recently replaced half its staff with machines. The other is a startup, and used machines from the outset. The former would be hit by a tax that the latter would escape. Not only is that patently unjust, it would simply mean the former would close, and the tax base would disappear anyway.
There are a number of problems with introducing UBI in the types of economy that exist today, which is why it has never been done, except in limited, temporary trials. The biggest problem is the little word in the middle: basic. Without spending implausible amounts, the best that UBI can do is to make its recipients poor. That is much better than being dead, if that were the alternative, but it is nowhere near good enough. As the noted economist John Kay said, “either the basic income is impossibly low, or the expenditure on it is impossibly high.”
Despite its failings, the proponents of UBI have hit upon a vital insight: in a post-jobs world, there are going to have to be substantial transfers of income and / or wealth from the minority who are still in paid employment, and from the owners of major assets. The problem is that if the cost of a good life remains high, then these transfers will be onerous, and therefore avoided. Wealthy people will move to jurisdictions which don’t enforce the taxes, or they will simply cease to work at all.
The solution is not to persecute the wealthy, but to reduce the cost of a good life. This means developing the economy of abundance, in which the prices of all the goods and services that you need for a very good standard of living are very low. Then the transfers from the wealthy need not be onerous – and avoided.
This does sound crazy when you first hear it, but if you think about the music industry, you can see how it could happen. Twenty years ago, not even a rich person could listen to any piece of music which took their fancy. Now it costs $10 a month, thanks to Spotify and similar services. Music, of course, is now digital, so it is dematerialised and non-rivalrous, as economists say. But more and more of what we value these days is digital, and this trend will continue. In another decade or so, many of us will probably spend hours at a time in virtual realities.
What form of economic and political governance will best suit this post-jobs world? Some have argued for a version of communism, called fully automated luxury communism. While this approach deserves credit for facing up to the exponential future, it overlooks two important facts. First, it is no coincidence that wherever communism has been tried, it has degenerated into some of the worst regimes the world has ever seen. Communism grants absolute power to a ruling elite, and we all know what absolute power does to people.
The other fact is the astonishing power of markets. Markets incentivise people and firms to provide goods and services that are genuinely valued, and to do it efficiently. Allied with technology, free markets (with appropriate welfare safety nets, and with regulation to ensure that markets are not suborned) have made this the best time ever to be a human. Technology provides potential solutions to problems, and markets provide the incentive for people to develop and invest in those solutions. This is how China has gone from backwater to superpower in a single generation.
Even in the abundance economy, it will be a very long time before we have the technology to build the matter replicators found on Star Trek’s USS Enterprise, so goods and services will impose a residual cost. And as long as we are human, we will also face shortages of things like attention, artisanal goods, and works of art. Markets are the best system we know of for resource allocation, so to start with at least, the optimal system for the economy of abundance is likely to be a form of capitalism: let’s call it “fully automated luxury capitalism”.
In fully automated luxury capitalism, people would continue to trade and innovate. Fortunes (and also more modest wealth) would continue to be made by creating art, artisanal goods, and other forms of intellectual property. Even when AIs can create objects more beautiful than the ablest human craft worker, there could well be a premium attached to items which are “made by hand”. Judiciously trading scarce assets such as beach-front properties, and original Aston Martin DB5s could remain another source of wealth and income.
But not everyone will be able to or inclined to participate in this reduced commercial world. The three central insights underpinning the idea of fully automated luxury capitalism are (1) that most humans will not have jobs, (2) that everybody must be wealthy – or at least comfortable rather than poor, and (3) that the taxes levied on wealthy people people and organisations must be affordable. The only way to achieve this is to drive prices down – to evolve the economy of abundance.
If we can achieve the economy of abundance, then we should be thankful that humans are not condemned to do jobs forever, because although some people love their jobs, plenty of research shows that most people don’t. Many people will take time to adjust, and some will need help, but a post-jobs world could be one in which humans do whatever we want to. We could have a second Renaissance. The Reverse Luddite Fallacists are not only probably wrong – they are also profoundly pessimistic.
Is the economy of abundance possible? Obviously, not everything is digital, like music. Can we reduce the cost of non-digital goods and services too, like housing and clothes? Can we create their Spotify equivalents, “Constructify” and “Clothify”?
We can, if we do three things. First, we must take the expensive humans out of the production process for all goods and services. That is exactly what automation does, so this is a case where the problem is also part of the cure.
Second, we must make energy very cheap. The cost of solar cells is falling fast, and battery technology is improving. Many observers think that within 20 or 30 years, electricity could not only be much cheaper to generate, store and transmit than the oil and gas we get by digging up dead dinosaurs – it could be almost too cheap to meter. This would also reduce CO2 emissions much faster than most people today think possible.
Third, we must use AI to make all production processes as efficient as possible, requiring minimal material inputs.
Andrew McAfee pointed out in his recent book “More From Less” that we are already on this path, but there is much further to go. We should therefore not hold automation back; instead we should accelerate it. The mantra for companies and for countries should be: “automate and redeploy, rinse and repeat”. Companies which automate rapidly will prosper, but the outstanding ones will be those which also genuinely value their people: companies where people actively seek to work out how to automate their tasks, in the knowledge that they will be rewarded both with pay and with more interesting work in the future.
For a decade or three before technological unemployment arrives, AI-driven automation will have an increasingly disruptive effect on the job market: the Churn. People will lose their jobs to machines, and they will have to be redeployed within companies, between companies, and between industries. With increasing frequency. Workers will have to learn how to work with machines: AI won’t replace humans in all jobs, but increasingly, humans who can work with AI will replace those who cannot. We cannot yet say what jobs people will be re-trained to do; no doubt some of them will surprise us.
This process will be disturbing and frightening for many of us. Governments will need to support their people through the Churn with enhanced welfare programmes and other methods of economic support. The experience of massive government intervention during the corona virus crisis should teach us valuable lessons about what works and what doesn’t.
During the period of Churn, we will need to re-train ourselves more and more frequently, and do it faster each time. Education and training are notoriously hard industries to reform. Here again, AI will provide solutions as well as raising the challenge in the first place. Education and training will become personalised, as we all acquire learning assistants - digital coaches which know better than we do (and better than any human teacher could do) what we already know, what we need to learn next, and how to optimise that learning process. For many years, these assistants will work alongside human teachers rather than replacing them. If we manage the transition smoothly, then by the time the replacement happens, we will hardly notice, and we won’t object.
Once we arrive at the economy of abundance, some people will continue to re-train for economic activity, but for most of us, educational will become vacational, not vocational. Which is what many of us wanted it to be all along.
To survive the economic singularity, we need to identify the optimal outcome in a largely post-jobs economy. If it is indeed the economy of abundance, and fully automated luxury capitalism, then we need to build a consensus about that, and a plan for how to get there without panic. Panic at the thought of racing toward technological unemployment without a plan could be almost as bad as the fact itself. But if we can manage the transition through the economic singularity successfully, the outcome will be wonderful.