In a previous column, I looked at the way automation and AI are likely to transform the world of work and employment. There is a lot of discussion about this, most of which focuses on the likely impact in terms of the kinds of paid work that will disappear. What there is much less of is discussion of the new kinds of paid work that will come into being.
If the result of automation is to create jobs more than to destroy them, then what kinds of work are likely to expand in the future? This is related to but distinct from the first question. In one way, this is a very hard question to answer. Many of the new kinds of employment that will appear are literally unimaginable — if we could imagine them, they would already exist.
Back in the 1980s, nobody could have told people worrying about the decline of jobs in the steel industry that there would be work designing apps for mobile phones, for example. So we can be confident that new kinds of work will appear but have no idea about what it will be — it’s for entrepreneurs to invent and discover that.
However, we can do some thinking about it because while the details may not be clear, there are cases where we can have a strong notion as to what will appear. In the 1900s, for example, there were a large number of jobs associated with horses, at that time still the main power for transport. Almost all of them were gone by 1930, but people could guess that a lot of new work would be created servicing and supporting (as well as producing) the motor vehicles that were replacing horses. Thinking like this about the present situation should lead us to a number of conclusions and to one in particular that many will find both surprising and heartening.
If you read the various studies that have been done over the last five years, there is widespread agreement about the kinds of jobs that are “at risk.” A recent study by the Brookings Institution estimated that 25 percent of current US jobs are at greater than 50 percent risk of automation. Some are not surprising. Any job that is both boring and repetitive is likely to be at risk. You might suppose that this would mean low-paid manual occupations would be at high risk, and indeed many are — shelf stacking, waitressing, and data entry are all at high risk.
On the other hand, many better-paying jobs are at considerable risk of disappearing. A range of jobs in the transport industry, from truck and taxi drivers to train and bus drivers, are likely to go in the medium term because of the rise of autonomous vehicles (most new metros around the world already have driverless trains). A wide range of clerical and administrative tasks are also likely to be handed over to algorithms, from financial services to company administration and financial advice.
The last example brings up another point. A recent study by the OECD argued that jobs involving face-to-face contact were more likely to survive — which suggests a rosier future for financial advisors. However, experience suggests this is actually not true. When the ATM was introduced, some argued that it would not catch on because customers preferred the human interaction with a teller. Experience suggests that actually the opposite was the case. The same is likely to be true in a range of occupations and not just financial advice and wealth management.
The common factor is that these are activities that can be readily reduced to a tick list of standard questions and hence an algorithm. Routine medical care and diagnosis is one; another is most standard legal work. This suggests that the risk of automation is actually high for many professional occupations such as medical general practice and routine attorney work. In the future, we will probably consult an algorithm rather than a human doctor or lawyer or accountant. However, surgery and nursing are still almost certain to be done by flesh and blood humans.
That particular example can lead us to the surprising and heartening conclusion mentioned earlier. Much of the commentary argues that we are moving into a world where the labor market will be dominated by two kinds of employment. There will be creative jobs that are open to highly educated people and which pay very well, and there will be unskilled and low-productivity jobs (hence low paying), but there will not be a range of middle-skill jobs that pay a decent or even high wage. The view is captured in the title of Tyler Cowen’s work Average Is Over. This has a number of alarming implications, most notably that access to high-paying work is going to become even more dependent than it already is on higher-education qualifications. We should be more sanguine, however.
A Heartening Conclusion
Economic theory, confirmed by empirical research, tells us that people will in general only adopt a new technology when the expected gain from doing so is greater than the cost (technically, when the marginal gain exceeds the marginal cost). This means there are many things that are technologically feasible that do not happen because they do not pass the test of their benefit being greater than their cost.
One example is supersonic passenger flight. This is certainly technically feasible — we know this because two such aircraft were in commercial service for some time. However, there are none now and no prospects for any. The reason, as Boeing discovered while trying to develop a supersonic transport, is that the benefit (getting from London to New York in three hours rather than seven, for example) is not valuable enough to consumers to exceed the costs of such an aircraft, which are due to the technology available and the unavoidable challenges of traveling at such a speed in the Earth’s atmosphere in a way that will meet standards of comfort and safety for passengers. (Military personnel have a different set of criteria, which is why supersonic combat aircraft are commonplace; plus, the buyers of such aircraft, namely governments, are not as price sensitive as airlines.)
One classic example of this is the artificial reproduction of manual dexterity or, to put it another way, of the combination of the human hand with the human brain and the complex feedback and control system (touch and sight) that connects the two. Despite much effort and research, this has proven incredibly difficult to reproduce artificially. Consequently, for most tasks involving manual dexterity and manipulation, it is still cheaper to use a human rather than a robot, and this seems likely to be the case for a long time. This explains why nursing and surgery are both at very low risk of being automated by everyone’s estimation, despite the fact that most surgical procedures and nursing tasks are standard and routine in many ways.
Manual Trades and Personal Services
So, there is a wide range of tasks and work that will not be automated because of this. However, the response might be that the kinds of jobs this applies to are precisely the low-productivity and low-skill jobs mentioned earlier, such as cleaning. Certainly, this is true, but it is not the whole truth. There is also a wide range of work involving manual dexterity that is skilled and highly paid, and the likely impact of AI will actually be to make that kind of work more productive and hence higher paid. Meanwhile, other foreseeable changes will increase the demand for these kinds of work and hence the number of employment opportunities, even allowing for the increase in productivity of individual workers.
This kind of work is that of skilled manual trades such as plumbing, painting and decorating, electricians, and construction work of all kinds. Another is personal-service work such as personal trainers or coaches. Teaching and researching are other examples (at the moment, these are thought of as jobs that require a degree, but that is more about rationing access than reality). Manual trades, for example, are very difficult because of the need for close-up manipulation — a robot that could do an electrician’s or plumber’s job would be seriously expensive.
At this point, another feature of innovation comes into play. What much innovation does is not so much replace human labor as enhance it and make it more productive. AI and associated control systems are a classic example of this. You will still need the manual dexterity of the surgeon or plumber, but the AI and associated technology will increase the range of things that they can do and make them much more effective. In other words, it will increase the value of the service they provide as well as the quantity per unit of time worked — which is the real definition of increased productivity. This translates into higher incomes for people delivering this kind of service.
Moreover, the demand for this kind of labor and service (as well as the others mentioned) is almost certain to increase. For one thing, the people earning very high incomes doing creative knowledge work will want to employ those providing these services in very large numbers (not least because the principle of comparative advantage means it makes sense for them to do this so they can concentrate on their own work). Another feature of AI is that it will make it much cheaper to personalize skilled work and services and so make it more valuable to the end consumer.
What we are likely to see, in fact, as well as the disappearance of a range of familiar jobs, is a revival in the value (and maybe the status?) of manual trades and personal services of all kinds. These will also become higher paying than many are at present (some, of course, already pay well). To give just one example, nursing and personal care are going to have their productivity significantly increased, while the demand for such services is going to rise organically because of the rise in the average age.
This will also mean an increase in the kind of work that requires a trade education and a relative decline in the need and demand for academic higher education. That sector will have to find another market to replace or supplement people looking for certification to have a shot at knowledge or creative work — the business of providing education and tutoring as a leisure and consumption good is one possibility.
In fact, one outcome of AI and automation may well be a revival of manual labor and of the traditional working class — maybe becoming more like an artisan class again. It is actually the credentialed and salaried white collar middle class that is more at risk in the years to come.
The overall effect will be massively positive, as economics leads us to expect. Thus a recent study by Price Waterhouse predicts that automation and AI will contribute an additional $15.7 trillion to the global economy by 2030, with a boost to local GDP of up to 26 percent by the same date. We should be sanguine about the impact of this latest wave of innovation, not just in terms of its overall impact on the wealth of the world but also in terms of its likely sociological impact.