Q: I am now reading the latest book by Daron Acemoglu: what happened to liberal democracy. He argues that blue collar jobs have declined in industrialised countries due to automation, off shoring and competition. it also led to more inequality as more income went to owners of capital. What do you think will happen with AI. I think that now jobs of white collar workers will become affected, whereas more complicated blue collar work as plumbing and nursing care will survive. What will be the overall impact on inequality?
A: Yes. I think your hypothesis captures something quite important: AI is likely to hit the labour market in almost the reverse direction from the first big wave of globalization and industrial automation. That makes its effect on inequality much less straightforward than “AI will make inequality worse.”
Acemoglu’s new book, published on August 11, explicitly connects the earlier loss of industrial and clerical middle-class jobs with the weakening of shared prosperity and the growth of inequality. AI could now disturb the ‘other side’ of that post-industrial settlement.
The surprising part: AI may compress some wage inequality
For roughly the last 40 years, advanced economies rewarded education enormously. A lawyer, consultant, programmer, financial analyst or senior civil servant could command a large wage premium because certain cognitive skills were scarce.
AI makes many of those skills less scarce.
Writing a competent report, analysing a spreadsheet, producing software code, translating something, doing preliminary legal research, creating a presentation or making a first medical interpretation can increasingly be done by a moderately skilled person assisted by AI.
Meanwhile, try asking an AI to:
* repair a leaking pipe behind a wall;
* help a frail 85-year-old shower without frightening her;
* diagnose why a heating system behaves strangely in a particular house;
These require physical dexterity, changing environments, interpersonal trust and often accountability. Robotics will eventually affect some of them, but the economics are much harder than putting another AI model on a server.
Current evidence already points in this direction. The ILO finds that clerical occupations have the highest generative-AI exposure, with growing exposure among financial analysts, programmers, investment advisers and other professional occupations. About 34% of employment in high-income countries has some GenAI exposure, versus around 11% in low-income countries. ([Internationale Arbeidsorganisatie][2])
So your plumber may actually become relatively more valuable than the junior accountant.
But there are really two different kinds of inequality
This is where I think the story becomes interesting.
Inequality between workers could fall. Some of today’s extraordinary wage premiums for cognitive work may shrink. Imagine that AI allows one lawyer plus AI to do what five junior lawyers previously did. The senior lawyer remains valuable, but fewer junior lawyers are needed. The wage structure between university graduates and skilled trades could narrow.
At the same time, inequality between labour and capital could rise dramatically.
Suppose a company previously employed:
1 owner + 100 knowledge workers.
With powerful AI it might employ:
1 owner + 20 people + AI systems.
Even if the remaining employees are well paid, a much larger fraction of the value created can accrue to shareholders, AI-platform owners and whoever owns the complementary intellectual property, data and distribution.
So you could get the apparently paradoxical outcome:
Less inequality between plumbers and accountants, but more inequality between both of them and capital owners.
That distinction is crucial.
And there may be an even bigger divide inside the white-collar world
I don’t think AI will simply eliminate white-collar work. It will probably split it.
At one end will be people whose main economic contribution is producing routine cognitive output: ordinary reports, summaries, coding, bookkeeping, standard contracts, translations, basic analyses. Their bargaining position may weaken considerably.
At the other end will be people who decide what should be done, take responsibility, persuade other humans, build relationships, exercise judgment and combine AI outputs into something useful.
That connects remarkably well with our earlier discussion about “the why” and “the how.”
AI is becoming extraordinarily good at the how. Therefore human economic value may migrate toward determining the why: deciding which problem deserves attention, what outcome matters, whom to trust, which trade-off society should make and whether the AI’s technically excellent answer is actually sensible.
That could make curiosity, judgment and responsibility economically much more important than producing intellectual output itself.
### There is another group that may do surprisingly well: human-intensive services
Nursing, elderly care, childcare, hospitality, therapy, teaching and many personal services have something economists call a productivity problem: one nurse can only look after so many patients.
Historically that has held their wages down because their productivity didn’t rise as quickly as manufacturing or software.
But AI could turn that argument around.
If AI makes the rest of the economy enormously productive, societies may become able—and forced—to spend more of their income on things that remain inherently human.
A future economy might therefore employ fewer accountants and administrative staff but many more carers, nurses, craftsmen, coaches and other people whose scarcity comes precisely from being human and physically present.
That would be quite a reversal of the past 40 years.
So my base-case prediction would be
Over the next 10–20 years in wealthy economies, I would expect:
| Group | Likely relative position |
| Routine clerical workers | ↓ strongly |
| Junior knowledge workers | ↓ |
| Average programmers/analysts | ↓ or transformed |
| Elite experts + AI | ↑ |
| Managers who mainly process information | ↓ |
| Skilled trades | ↑ |
| Nurses/care workers | ↑ relative scarcity |
| Highly interpersonal professions | ↑ |
| AI entrepreneurs/platform owners | ↑↑↑ |
| Owners of capital generally | ↑ |
And therefore my answer to your actual question is:
Without policy changes, AI will probably increase wealth inequality even if it reduces some wage inequality.
That is because the dominant distributional question isn’t ultimately “which occupations does AI replace?” It is “who owns the machines?”
Acemoglu’s broader argument is very relevant here. He emphasizes that technology doesn’t mechanically determine distribution: institutions, taxation, corporate choices, worker bargaining power and the direction in which innovation is steered matter enormously. His proposed “working-class liberalism” is essentially an attempt to reconnect technological progress with broadly shared prosperity. ([shapingwork.mit.edu][3])
And I think AI makes that question even sharper than the industrial automation he describes.
There is a potentially rather ironic historical sequence:
Industrialization: machines complement workers → mass middle class grows.
Automation/globalization: machines/offshoring replace many blue-collar workers → educated white-collar class pulls ahead.
AI: machines begin replacing parts of educated white-collar work → education premium shrinks.
Advanced AI: capital becomes capable of performing much cognitive and eventually physical labour → the fundamental divide becomes capital owners versus everyone who mainly owns their labour.
That final stage, rather than plumbers versus professors, is the inequality issue I would worry about most.
And it gives your observation an interesting political implication too. If AI erodes the economic security and status of the educated professional class in the way automation previously affected industrial workers, **the political shock could be enormous**. Suddenly the people who designed and defended the post-industrial liberal order may themselves experience technological displacement. That would be a fascinating extension of Acemoglu’s argument—and perhaps an even bigger challenge for liberal democracy than the one his book describes.
[1]: https://www.penguinrandomhouse.com/books/815318/what-happened-to-liberal-democracy-by-daron-acemoglu/?utm_source=chatgpt.com “What Happened to Liberal Democracy? by Daron Acemoglu: 9798217182169 | PenguinRandomHouse.com: Books”
[2]: https://www.ilo.org/publications/generative-ai-and-jobs-refined-global-index-occupational-exposure?utm_source=chatgpt.com “Generative AI and Jobs: A Refined Global Index of Occupational Exposure | International Labour Organization”
[3]: https://shapingwork.mit.edu/what-happened-to-liberal-democracy/?utm_source=chatgpt.com “What Happened to Liberal Democracy? – MIT Stone Center on Inequality and Shaping the Future of Work”
