The UK recently held an international summit about artificial intelligence (AI). Leaders and large firms have been taking AI extremely seriously.
Now, economists are jumping into AI research too.
Economists’ interest in AI has exploded recently, with tons of new research coming out about its consequences.
So, should we worried about AI taking our jobs? Will AI worsen inequality?
What does economic theory say about the effects of automation on jobs?
How does automation affect employment in theory? Here are a few possible ways:
Channel 1: Replacing jobs
Automation encourages firms to replace workers with machines. This reduces demand for labour and employment falls.
Generally, this occurs where workers and capital are substitutes.
Channel 2: Complementary work
Sometimes labour and capital are complements. For example more engineers and IT workers may be needed to fix and program machines.
So automation may increase demand for some worker types.
Channel 3 – Automation and prices
Automation, as it cuts costs, may reduce prices. This leaves more income for consumers to spend on other goods, increasing consumption of other goods. Via derived demand, employment in these other sectors rises.
Other effects – inequality and firm-specific consequences
Automation can increase inequality. This depends on which jobs automation replaces and creates.
Automation can also have different effects for different firms.
Firms that automate can raise productivity and lower costs, raising profits. This allows the firm to expand and hire more workers.
What does economic history say about automation?
One of the most well-known examples of technological change was the Industrial Revolution.
There were new technologies influencing the entire economy, like the steam engine. Also new technological developments appeared for specific industries. For example, machinery for spinning cotton.
What effect did this have on workers?
During the Industrial Revolution, there were complaints about long hours and urbanisation. This seems difficult to square with the idea of automation reducing hours worked.
On the balance of evidence, it is likely that real wages did not grow in this period (though this is disputed).
The Industrial Revolution did see higher demand for complementary labour.
Not only mechanics to fix machines but also supervisors and accountants. The latter occupations could control and manage the large scale factories.
Also, technological innovation can create entirely new sectors and methods. In doing so, innovation generates demand for new types of work. This included new chemical manufacturing methods, mechanised factories and steam power.
Not everybody benefitted from cost-saving machinery. Workers were displaced. Broadly throughout the economy, factory work was replacing handicraft work.
For example, inventions in textiles led to the mechanisation of spinning, reducing wages in artisan work. Handloom weavers made up 10% of the male workforce alone in 1820 (Allen 2021).
A group of English textile workers, called the Luddites, protested mechanisation. Their tactics even included destroying some of the machines.
The economy may benefit from automation over time through rising productivity Nevertheless it would take longer than the average working generation in the 19th century, for all to benefit from this change. In the meantime, some were worse off while others were better off.
Into the late 19th and 20th centuries
Strains of “Luddite” thought occasionally reappeared throughout the 19th and 20th centuries.
Hours worked have fallen. From 1870 to 1998 annual hours worked fell from 2950 hours per worker to 1500 per worker in highly industrialised western economies.
This could be a good thing or a bad thing. For instance, lower hours worked means more leisure time. But people may work to feel accomplished or to make a contribution.
The reduction in hours worked has fallen most on those without college or university degrees. This group saw their hourly week fall by 10 hours from 1965 to 2003, compared with a less than one hour fall for college graduates. This can lead to higher inequality in terms of income and wealth (Mokyr).
A caveat is that other factors will also determine hours worked.
However automation could reduce inequality in terms of access to primary resources. Primary resources could include basic needs like including food. This is because if automation lowers costs for firms, it may also lower prices.
Automation has also contributed to rising inequality from the 1980s. Consider deindustrialisation in the UK and the loss of jobs along the “rust belt” in the US. Deindustrialisation has occurred due to a combination of automation and globalisation, among other things.
Any retraining schemes, to find workers new jobs, were broadly insufficient.
This contributed to regional inequality, persisting until this day.
New jobs
Autor (2015) titles his paper on automation “Why are there still so many jobs?”.
In spite of the predictions of zero jobs by the 21st century because of automation, there are very still very many jobs.
As of writing, unemployment rates are low in advanced economies. Though the phenomenon of “underemployment” may tell a different story.
Automation can replace some jobs but critically, it creates new ones too. Moreover, journalists tend to ignore how labour and automation can complement one another.
We mentioned already the complementary labour in the industrial revolution. Yet we can also come up with several more recent examples.
The IT revolution has led to higher demand for coding and machine learning skills.
Even the creation of social media, for instance, has led to several new jobs. Jobs at social media companies, advertising jobs, running social media groups and social media profile managers.
Also, if automation reduces costs and lowers prices, this frees up income to spend on other goods and services. This could include higher demand for leisure activities. This could raise employment in the entertainment, tourism and health and fitness industries.
Why elasticities matter in automation
Elasticities capture the sensitivity of supply or demand to price / wage / income changes.
For understanding the economic consequences of automation, elasticities are critical (Autor 2015). Here are a few examples:
The wage elasticity of labour supply:
- Suppose automation increases the demand for workers in a certain sector. For example, IT.
- This is likely to raise wages for this sector.
- How easily can workers from other sectors retrain to become IT workers?
- If it’s easy to retrain, then the wage gain from automation will be small. But if it’s hard to retrain, the wage gain for IT workers is much larger.
- This is the concept of “wage elasticity of labour supply” – how much does labour supply increase, following a wage rise?
The income elasticity of demand
- Automation can lower costs and prices, freeing up income to spend on other goods. At least in theory.
- But where do consumers spend their extra funds?
- This is likely to be on income-elastic goods – goods whose demand is very responsive (positively) to income rising.
- Examples of such sectors include restaurants and personal fitness.
- Even though these sectors do not tend to face heavy automation, they could indirectly benefit from automation. These growing sectors could then create more jobs in the process.
Firm level evidence
Acemoglu et al. (2020, 2023) look at the effects of automation on employment, specifically at the level of the firm.
In France, they find that firms adopting robots leads to a falling labour share of total income, as well as higher productivity. Employment increases in robot-adopting firms. This also occurs in data from the Netherlands.
What about competing firms that do not adopt robots? In the France data, these firms employ fewer workers.
At the level of the entire market, employment falls. The negative employment effect from non-adopters outweighs the positive employment effect in robot adopters.
The Netherlands’ data also breaks down types of workers. Workers, whose jobs are directly affected by automation, see lower earnings and lower employment rates.
Here, directly affected workers refer to the workers whose jobs are most likely to be replaced. However other workers indirectly gain from automation.
In summary, there are clear imbalances in terms of the effects of automation on workers, depending on the type of firm and type of job.
Developments in recent years: AI and large language models
More recent developments in automation include machine learning and large language models. What are the effects of these types of automation, so far?
ChatGPT has reduced monthly earnings and tasks for freelance writers and image creators. This effect does not appear to depend on the skill level of the freelancer.
This paper involves using data from Upwork, a freelancing platform, to generate a “natural experiment“.
Specifically, the authors Hui. et al. (2023) compare A) the professions affected by generative AI, like ChatGPT, to B) professions that are less likely to be affected.
Suppose we start at the time ChatGPT and other generative AI software are introduced. From then onwards, we observe a divergence in the earnings and tasks of the two types of professions.
Another paper finds that using ChatGPT increased productivity among writers. This occurred particularly for writers with low ability levels.
This shows that generative AI may help those at the bottom of the skill distribution “catch up” with their peers. Potentially, automation can reduce inequality.
This is somewhat unexpected, as often we associate automation with inequality. But automation seems to be narrowing the gap between workers in the market for freelance writers.
Evidence is emerging on the impact of artificial intelligence more broadly on employment. Yet so far, impacts have been restricted and evidence has been mixed (see Acemoglu et al. 2021 and Albanesi et al. 2023).
What are possible policies to protect workers from automation?
Automation in the past has contributed to rising income inequality in the past.
Automation more likely to eliminate repeatable jobs. With such jobs often being lower paid, inequality may increase inequality.
Several papers in the area of automation make the case for some kind of redistribution.
This could include a more progressive income tax schedule, wealth taxes or software taxes.
Acemoglu, Autor and Johnson (2023) discuss how policy can move AI onto a pro-worker path. Key policies could include:
- Equating taxes paid by firms for labour and software. Currently labour taxes are more burdensome.
- Restrictions on deploying untested AI in areas that could affect workers’ jobs.
- Funding to support research into human-complementary AI.
- AI experts in government, who are independent of tech companies.
What are the future effects of automation on employment? The bottom line
What happened in the past, will not necessarily happen in the future.
Nevertheless, the evidence on automation so far gives clues as to the possible future effects on employment.
In particular, there are the following key questions to answer:
- Which jobs are most prone to being automated?
- Which new jobs will appear? I think this is a tougher question, as it requires foreseeing the new technologies that may appear.
- Will governments actually intervene to turn AI “pro-worker”? Or will it be left to the free market entirely to decide how automation affects jobs?
There is some evidence on the first question already. But thinking about the other questions will prove more difficult.
If requested, I will write another article on this topic. For now, I leave you to think about the impacts of AI on jobs in the future.
In particular, remember the following:
- AI can both destroy and create jobs.
- Automation has contributed to higher inequality historically and this is likely to continue. But this need not be the case with all new technologies.
- Elasticities influence the effect of automation on jobs, wages and inequality.
- Effects of automation may differ not only by industry but also by firm.
Finally, by focusing on jobs alone, we may fail to see the wood for the trees.
Even if at the economy-wide level, unemployment does not fall, there will be other consequences.
Wages, hours worked and importantly, the distribution of impacts matter too.
Related posts
How the UK tackles market power, mergers and big tech
How did Amazon grow so big? | 4 key economic factors
What economic theory says about immigration
Behavioural biases in decision making and how to avoid them
Why Goldin won the Economics Nobel Prize | 5 Key Takeaways
Key “marginal” economics equations and how they apply to the real world