Whether AI narrows or widens the wage gap comes down to whose labour it substitutes for and whose it complements. AI is strong at routine cognitive work, the white collar tasks that have long commanded a wage premium, so it may compress the pay of that work by acting as a substitute and lowering the demand for it. Meanwhile, work that is hard to automate should hold up: jobs that need soft skills and dealing with people, and skilled trades like plumbing and carpentry that need hands on competence. On that reasoning the gap between high paid routine cognitive work and skilled manual or interpersonal work could narrow. The opposite case is real too: if the gains from AI flow mainly to capital owners and a small group of highly skilled workers, inequality could widen instead. It is a question of derived demand and wage differentials, and the honest answer is that it depends.
Almost every headline about artificial intelligence and jobs tells the same story: the machines are coming, inequality will widen, and the people at the bottom will be hit hardest. It is a reasonable fear, and I am not going to pretend it is baseless. But I want to put the other case, because I think it is underrated, and because arguing it properly is a near perfect exercise in A level labour market economics. My own view, and I will hold it loosely because nobody can be certain here, is that AI may actually narrow the wage gap rather than widen it.
The story, first
Here is what is actually happening. The latest wave of AI is unusually good at routine cognitive work: drafting, summarising, coding to a brief, first pass legal and accounting tasks, the kind of desk work that, until now, you needed a degree and a salary to do. For a long time that white collar work has commanded a wage premium over manual and service work. The interesting question is what happens to that premium when a tool can do a large slice of the work quickly and cheaply. That is not a moral question, it is a labour market question, and the syllabus already gives you the apparatus to answer it.
The economics: it is a labour market
Start from the foundation. A wage is a price, and like any price it is set by demand and supply, here the demand for and the supply of a particular kind of labour. The reason a surgeon earns more than a cleaner is not fairness or effort; it is that the demand for surgeons is high relative to a supply that is hard to expand, while the supply of people who can clean is large. Different wages for different kinds of work are wage differentials, and they exist because each kind of labour sits in its own market with its own demand and supply.
The second idea is the one that does the heavy lifting. The demand for labour is a derived demand: firms do not want workers for their own sake, they want what the workers produce. So the demand for any kind of labour depends on how productive and how valuable that labour is to the firm. Now drop a new technology into that picture. A technology can substitute for a kind of labour, doing what that worker did, which lowers the demand for them and pushes their wage down. Or it can complement a kind of labour, making that worker more productive, which raises the demand for them and pushes their wage up. Whether AI widens or narrows the wage gap is, at bottom, a question about which kinds of work it substitutes for and which it complements.
- Derived demand
- The demand for labour is derived from the demand for what it produces. Firms hire a kind of worker for the value that worker adds, so anything that changes that value changes the demand for them, and so their wage.
- Wage differential
- The difference in wages between different kinds of work. It arises because each type of labour has its own demand and supply, so a shock that shifts demand for one type relative to another changes the gap between them.
Mr Toh's take
So here is my actual view, and I want to be honest that it is a hypothesis I am arguing, not a fact I am reporting. I think AI is going to compress the wage gap rather than widen it, and the reason is the specific thing it is good at. AI is very good at exactly the routine cognitive work that has carried a wage premium, the white collar desk tasks. If a tool can do a meaningful share of that work, the premium on it is the premium most under pressure. The market that gets the new cheap substitute is the one whose relative wage I would expect to fall.
Meanwhile, two groups of workers should do relatively well, because their work is genuinely hard for a machine to do. The first is people with soft skills, the ability to talk to people, read a room, persuade, comfort, manage and negotiate. That is the hardest thing to automate, because it is not really a routine at all. The second is people with specific hands on competencies, the skilled trades, a good plumber, a good electrician, a good carpenter. A machine cannot yet crawl under your sink and fix the leak. So the work that AI struggles with is, interestingly, found at both the interpersonal end and the manual end, and it is the comfortable middle of routine cognitive work that is most exposed.
Put those together and you get the compression. If routine cognitive work loses some of its premium while skilled trades and people facing work hold their value, then the gap between the highly paid desk worker and the skilled plumber or the warm, capable carer narrows. That is the case I would argue. But I would be doing you a disservice if I stopped there, because the opposite case is just as arguable, and a good economist holds both.
The honest counter argument runs like this. The gains from AI may accrue not to ordinary workers at all but to the owners of capital, the firms and shareholders who own the AI, and to a small group of very highly skilled workers who design and direct it. If output rises but the returns flow to capital and to a superstar few, the share going to labour falls and inequality widens, not narrows. There is also the transition: even if the long run picture is benign, the workers whose tasks are automated face real disruption now, and whether they can retrain into the safer kinds of work depends heavily on education and access. So the same labour market logic can point either way. That is exactly why it is an evaluation question, not a slogan.
If a question asks whether technology raises or reduces inequality, do not pick a side and assert it. Show that the demand for labour is a derived demand, then split the labour market by how exposed each kind of work is to the technology, and argue both directions before you judge. A model sentence: "Because the demand for labour is derived from its productivity, a technology that substitutes for routine cognitive labour lowers the demand for it and so its relative wage, narrowing the differential with hard to automate trades and interpersonal work; however, if the gains accrue chiefly to capital and a small pool of complementary high skill workers, the wage gap may widen instead, so the net effect depends on the balance of substitution and complementarity and on the scope for retraining."
Notice what that model sentence does. It does not gamble on a prediction. It names the mechanism, derived demand, then shows the shock pulling in two directions, and makes the judgement turn on which force dominates and on whether workers can move. That is what the top band rewards, and it is also, as it happens, the honest state of the actual debate.
| Kind of work | Exposure to AI | Likely wage effect |
|---|---|---|
| Routine cognitive, white collar desk work | High: AI substitutes for much of it | Premium may compress as relative demand falls |
| Interpersonal and care work, strong soft skills | Low: hard to automate the human element | Relatively protected; demand may even rise |
| Skilled trades, plumbing, carpentry, wiring | Low: hands on, physical, hard to automate | Relatively protected as a substitute is scarce |
| Owners of capital and a few high skill specialists | Complemented, not replaced | Could gain most, which is how the gap might widen |
How different kinds of work are exposed to AI, and why the same labour market logic can compress the gap at one end and widen it at the other. These are reasoned directions, not precise forecasts.
A wage is a price. To ask whether AI narrows the gap is to ask whose labour it substitutes for and whose it makes more valuable.
- Wages are set by demand and supply for each kind of labour, so a wage gap is a gap between separate labour markets, not a single number.
- The demand for labour is a derived demand. A technology that substitutes for a kind of work lowers the demand for it; one that complements it raises the demand.
- My view: AI may narrow the gap. It is strong at routine cognitive work that has carried a premium, so it may compress that premium while skilled trades and soft skill work hold up.
- The opposite case is real. If the gains flow to capital and a few high skill workers, inequality could widen instead, and the transition is hard for those displaced.
- In the exam, argue both directions. Use derived demand and wage differentials, weigh substitution against complementarity, and let education and retraining carry the judgement.
Want this on paper? Grab the free 112 page Summary and Diagrams pack.
Frequently asked
Will AI narrow the wage gap?
It might, but it is genuinely uncertain and the answer depends on whose labour AI substitutes for and whose it complements. The case for narrowing is that AI is strongest at routine cognitive, white collar work that has long carried a wage premium, so it may act as a substitute that lowers the relative demand for that work and compresses its premium, while work that is hard to automate, soft skill and people facing roles and skilled trades like plumbing and carpentry, holds its value. The case for widening is that the gains from AI may flow mainly to the owners of capital and a small group of very highly skilled workers, in which case the share going to ordinary labour falls and inequality grows. In labour market terms it is a question of derived demand and wage differentials, and the net effect depends on which force dominates and on how easily displaced workers can retrain.
Will AI cause unemployment?
AI is likely to change the mix of jobs more than it removes the need for work altogether, but the transition can be painful. Because the demand for labour is a derived demand, when a technology can do a task more cheaply the demand for the workers who did that task falls, which can cause structural unemployment for people whose skills no longer match what firms need. At the same time, new tasks and new roles tend to appear, and workers whose skills complement the technology can become more productive and more in demand. Whether the overall effect on employment is small or large depends on how fast the change comes and on how well education and retraining help workers move into the kinds of work that are less exposed. The economics does not promise mass unemployment, but it does point to real adjustment costs that policy has to manage.
What jobs are safe from AI?
No job is completely safe, but the work that is hardest to automate tends to share two features: it relies on dealing with people, or it relies on physical, hands on skill in an unstructured setting. Roles that lean on soft skills, persuading, caring, managing, negotiating and reading a situation, are difficult to automate because they are not really routines. So are many skilled trades, such as plumbing, electrical work and carpentry, where the task is physical, varied and done on site. The most exposed work, by contrast, is routine cognitive work that can be specified clearly and done at a desk. In labour market terms, the safer work is the kind for which AI is a poor substitute and, ideally, a complement that makes the worker more productive rather than replacing them.
How does technology affect wages?
Through the demand for labour, which is a derived demand: firms hire workers for the value they add, so anything that changes that value changes the wage. A technology can substitute for a kind of labour, doing what that worker did, which lowers the demand for them and pushes their wage down. Or it can complement a kind of labour, making that worker more productive and more valuable, which raises the demand for them and pushes their wage up. This is why the same technology can lift some wages and depress others at the same time, and why it can either widen or narrow wage differentials depending on which kinds of work it substitutes for and which it complements. It is also why education and retraining matter so much: they decide whether a displaced worker can move into the work the technology rewards.
Ds and Es to an A
"Mr Toh is an amazing econs tutor. He helped me jump from Ds and Es to an A for A Levels. Very understanding and patient, super caring too."Jamie Ng
The standard, every week.
One essay or case study a week, personally marked with a worked model and a video walkthrough, from materials written by the author of the H1 and H2 TYS answer keys sold at Popular. This is the core JC1 and JC2 programme.
Weekly, marked, everything included
- A marked essay or case study each week
- Worked model plus a video walkthrough
- Onsite, live Zoom or recordings