AI could give humanity more freedom than any technology in history. Instead, I fear we're building an economic system that will make millions of people poorer before we're forced to share the benefits.
I've been thinking a lot about human potential recently.
About the extraordinary things we're capable of achieving, the opportunities that allow some people to realise their potential, and the countless brilliant minds throughout history who never had the chance.
And inevitably, those thoughts have led me back to artificial intelligence.
AI represents an extraordinary opportunity for humanity. We are developing technology that could dramatically reduce the amount of human effort required to produce goods, deliver services, discover new medicines, design infrastructure and solve problems.
In theory, that should be wonderful.
But I have a growing concern that the economic structures we've built around employment, wealth and ownership are fundamentally incompatible with the future we're creating.
I don't think AI is necessarily going to make humanity poorer. I think it could make humanity extraordinarily wealthy while making an enormous number of individual humans significantly poorer.
And I increasingly suspect that meaningful economic reform won't happen until the consequences become catastrophic.
The promise of working less
Let's start with a simple example.
Imagine I currently work five days a week, earning a comfortable salary as a technology professional.
Now imagine AI advances to the point where I can achieve exactly the same output in two days.
That's a 60% reduction in the human effort required to do my job.
Wonderful!
I could spend three additional days every week with my family. I could explore scientific ideas, develop experimental projects, contribute to open-source software, or simply enjoy being alive.
The company I work for still receives the same value. I'm still delivering everything it previously paid me to deliver.
Surely this is what technological progress is supposed to achieve?
Except I don't believe that's how it will play out.
Under existing economic incentives, the business is more likely to ask why it should continue paying me for five days when it only needs two.
Or why it should employ five people when AI allows one person to produce the same output.
The productivity gains don't automatically translate into greater freedom for employees. They create an opportunity for businesses to reduce labour costs.
And the financial benefits of those reductions are disproportionately likely to accrue to the owners of the businesses and technologies involved.
We could build a world where nobody needs to work five days a week, yet end up with a world where millions of people desperately need five days of paid work they can no longer find.
That's an extraordinary contradiction.
The illusion of rising salaries
There's another problem with how we might measure this transition.
Imagine a company employs 1,000 people, each earning £50,000 a year.
Its annual salary bill is £50 million.
Now introduce AI automation.
The company discovers it can maintain the same output with just 200 highly skilled employees. It makes 800 people redundant and doubles the salaries of the remaining workforce.
The result?
| Metric | Before AI | After AI |
|---|---|---|
| Employees | 1,000 | 200 |
| Average salary | £50,000 | £100,000 |
| Total annual salaries | £50 million | £20 million |
| Reduction in salary costs | — | £30 million |
Average salaries have doubled!
What fantastic economic progress.
Except 800 people have lost their jobs, and £30 million a year has disappeared from the company's direct contribution to household employment income.
Of course, not all that saving necessarily becomes profit. AI systems cost money, some workers will find other jobs, and some savings may be passed to customers.
But the example illustrates something important.
Rising average salaries tell us very little about what happens to the people who no longer receive a salary at all.
We could see rising wages among employed professionals, increasing productivity, and extraordinary corporate profitability while simultaneously experiencing declining household incomes across large sections of society.
If we measure the wrong things, we could celebrate economic success while millions of people experience economic failure.
We need to be looking at employment, real household disposable income, total labour compensation, working hours, wealth concentration and the proportion of economic output reaching ordinary people.
Not simply the salaries of those fortunate enough to remain employed.
The economic starvation loop
A while ago, I described something I called the AI Starvation Loop.
It concerned the relationship between artificial intelligence and the internet.
AI systems consume information created by publishers, journalists, authors and other content creators.
Increasingly, they answer people's questions directly, removing the need to visit the original websites.
Fewer visitors mean less advertising revenue, fewer subscriptions and less income for the people producing the content.
Eventually, some publishers can no longer afford to produce it.
The AI systems then begin to undermine the very information ecosystem they depend upon.
The technology becomes increasingly capable of consuming knowledge while simultaneously weakening the economic mechanisms that create new knowledge.
Unless the value flows back to the source, the system risks starving itself.
I've realised that almost exactly the same mechanism could apply to the wider economy.
AI could create an Economic Starvation Loop.
- Automation increases productivity. Companies can produce the same goods and services with fewer employees.
- Employment income declines. Workers lose jobs, hours or bargaining power, while an increasing share of the benefits flows towards owners of capital and technology.
- Consumer purchasing power erodes. Fewer households can afford the goods and services businesses are producing.
- Businesses experience declining demand. Companies become increasingly efficient at producing things their former customers can no longer afford.
- Profits come under pressure. The apparent benefits of automation are undermined by the deterioration of the wider consumer economy.
- Businesses seek further efficiencies. Automation, consolidation and cost reduction become responses to weakening demand, potentially reinforcing the original problem.
This wouldn't necessarily happen evenly or everywhere. Some industries would flourish while others declined. Some workers would benefit enormously. Lower production costs might make many goods cheaper.
But if the overall effect is to concentrate purchasing power faster than it is redistributed, the system could progressively undermine its own foundations.
The internet starvation loop risks destroying the producers of information. The economic starvation loop risks destroying the consumers of production.
Both emerge from a similar underlying problem: extracting value from an ecosystem without adequately sustaining the people who make that ecosystem viable.
The problem with wealth flowing upwards
One of the fundamental assumptions behind modern economic progress is that increasing productivity eventually improves living standards.
And historically, technological advances have indeed contributed enormously to prosperity.
But the distribution of those gains matters.
A company reducing its costs doesn't automatically mean its customers become wealthier.
A business becoming more profitable doesn't automatically mean the communities around it become more prosperous.
And a country becoming more productive doesn't automatically mean its citizens enjoy greater economic security.
Wealth isn't a fixed quantity. Economic activity creates new value, and I don't believe there's some absolute ceiling on how much wealth society can generate.
But there may be a practical limit to how much purchasing power can become concentrated before the wider economy starts experiencing serious problems.
Businesses need customers.
Customers need purchasing power.
And for most people, purchasing power currently comes primarily from employment.
What happens when businesses no longer need to employ the very people they depend upon to buy their products?
Some wealth will continue circulating through investment, consumption, taxation and new industries. That matters, and the economy is more complicated than a simple transfer of wages into profits.
But I worry that those mechanisms won't redistribute enough purchasing power, quickly enough, to compensate for the scale of disruption AI could create.
And there is an especially uncomfortable possibility.
The wealthiest businesses might continue to prosper for quite some time by selling to other businesses, governments and affluent consumers.
Which means the suffering of ordinary households might not immediately create enough economic pressure to force change.
Governments will inherit the problem
As people lose employment income, demand for social support is likely to increase.
Governments may face rising expenditure on unemployment assistance, housing support, healthcare and other public services.
At the same time, if labour income falls substantially, revenues from income tax and employment-related contributions could weaken.
Governments would need to find replacement revenue.
In a world where AI produces enormous economic value, the obvious candidates include some combination of corporation taxes, capital gains taxes, wealth taxes and other mechanisms for taxing the ownership and returns of productive assets.
But this introduces another problem.
Capital is international.
Large businesses can choose where to invest, where to establish operations and, within legal constraints, how to structure their profits.
Countries compete to attract those businesses.
A government that significantly increases taxes on corporate profits or accumulated wealth may fear losing investment to another country offering more favourable terms.
And so governments find themselves competing against one another to retain the very sources of wealth they increasingly need to tax.
International tax agreements demonstrate that cooperation is possible. There are policy solutions available.
But agreement is difficult, implementation is slow, and the incentives to break ranks can be powerful.
Meanwhile, governments may find it politically or administratively easier to collect revenue through taxes on employment and consumption.
Those approaches can place additional pressure on households already struggling with reduced purchasing power, depending on how the taxes are structured.
The result could be a system increasingly dependent on public support to maintain household spending, while struggling to tax the economic gains that could fund that support.
Public transfers aren't inherently unsustainable. A sufficiently productive economy can support substantial redistribution.
The danger is an extraordinarily productive economy whose governments cannot capture and distribute enough of that productivity to keep society economically healthy.
Why I think catastrophe may come before reform
This is the part where I become considerably less optimistic.
I believe there are ways to avoid the outcome I'm describing.
We could reduce working hours without proportionately reducing salaries.
We could expand employee ownership and profit-sharing.
We could develop better systems of taxation, international economic coordination and income security.
We could deliberately distribute the benefits of automation more widely.
None of these ideas requires a collapse of civilisation.
But I don't believe they'll be implemented quickly enough, or at a sufficient scale, to prevent enormous economic disruption.
Not because every business leader is malicious, or every politician is corrupt.
Because the incentives are misaligned.
For an individual company, reducing costs through automation may be economically rational.
For an executive whose rewards are tied to financial performance, pursuing those savings may be personally rewarding.
For a government, maintaining corporate investment and competitive tax conditions can be politically and economically attractive.
Each participant can act rationally according to the incentives immediately in front of them, while their collective decisions progressively undermine the wider system.
Why would a company voluntarily surrender a substantial proportion of its automation-driven profits when competitors aren't required to?
Why would a government introduce aggressive wealth redistribution if doing so risks losing internationally mobile businesses?
Why would wealthy asset owners support significant changes to arrangements from which they are benefiting?
Some will, of course. Not everyone acts solely in their own financial interest.
But I'm not convinced enough will.
My fear is that the people with the greatest power to change the system won't experience sufficient pressure to do so until the system itself begins threatening their wealth.
And by that point, the consequences for millions of households could already be devastating.
I suspect meaningful reform may ultimately be driven not by fairness, but by economic self-preservation.
When consumer demand collapses, businesses fail, tax revenues deteriorate and economic instability spreads across major economies, the incentives change.
At that point, transferring more purchasing power back into society becomes essential to maintaining the markets upon which concentrated wealth depends.
The wealthy may discover that keeping everyone else economically viable is necessary to keep themselves wealthy.
That would be an extraordinary irony.
We may end up redistributing wealth not because we've collectively decided people deserve to benefit from technological progress, but because the alternative has become economically catastrophic.
And even then, I suspect the pressure would be to redistribute enough to restore economic stability, rather than fundamentally rethink how prosperity is shared.
AI makes the speed of change different
None of these concerns began with AI.
The Industrial Revolution displaced workers and transformed communities.
Mechanisation fundamentally changed agriculture.
Computers disrupted clerical work.
The internet transformed retail, media and countless other industries.
Technology has always changed the economic value of human labour.
But AI introduces a potentially important difference.
It may reduce the time available for people to adapt, while simultaneously narrowing the range of professions into which they can realistically retrain.
Historically, the familiar response to technological displacement has been education and retraining.
Learn new skills. Move into a different industry. Adapt to changing demand.
But what if the new skills you're learning are themselves being automated before you've finished acquiring them?
What if a profession that looked relatively secure three years ago is no longer secure by the time you've qualified?
And what happens when increasingly capable AI systems help accelerate the development of further AI capabilities?
I don't know how quickly any of this will happen. Nobody can confidently predict the precise pace or ultimate limits of AI-driven automation.
But the possibility is sufficiently serious that I find the familiar reassurance that technology always creates new jobs rather unsatisfying.
It might.
The question is whether those jobs emerge quickly enough, whether there are enough of them, and whether the people who lose their livelihoods can realistically access them.
An economy might eventually adapt. That doesn't mean the people living through the transition will be okay.
Progress should mean more than productivity
Perhaps what frustrates me most about this possible future is the extraordinary opportunity we could waste.
Imagine a society where AI allows us to produce everything we currently produce with a fraction of the human effort.
Imagine people working two or three days a week while maintaining a comfortable standard of living.
Imagine parents spending more time with their children.
Imagine more people having the freedom to study science, create art, build things, contribute to their communities or simply pursue their curiosity.
Imagine how much human potential could be unlocked if fewer people had to devote most of their waking lives to earning enough money to survive.
That's the future I would love AI to help create.
I'm enthusiastic about AI. I work with it constantly. I genuinely believe it represents one of the most extraordinary technological developments in human history.
I'm not frightened of machines becoming more capable.
I'm frightened that we'll use those capabilities to create extraordinary abundance, then distribute the economic benefits in a way that makes ordinary people less secure.
And I'm particularly frightened that we'll recognise the problem, understand the available solutions, and still fail to act before the consequences become severe.
The ultimate irony
The original AI Starvation Loop asks a relatively simple question:
What happens when AI destroys the economic incentives to produce the information it relies upon?
The Economic Starvation Loop asks a much larger one:
What happens when AI destroys the employment income that allows people to consume the goods and services it helps produce?
In both cases, the technology could become extraordinarily successful at extracting value while gradually undermining the ecosystem that sustains that value.
I hope I'm wrong about how this unfolds.
I hope governments cooperate early, businesses find ways to share productivity gains, and AI creates opportunities we haven't yet imagined.
There are credible reasons to believe those outcomes are possible.
But I believe it's more likely that we'll wait until economic pain becomes sufficiently widespread to threaten the prosperity of those currently benefiting most from the transition.
Only then will the incentives become powerful enough to force meaningful change.
And that could mean years, perhaps decades, of entirely avoidable hardship for millions of people.
What makes this particularly tragic is that the technology causing the disruption might also give us the productive capacity to eliminate much of that hardship.
We could have the ability to create unprecedented prosperity and still end up with unprecedented economic insecurity.
Not because we lacked the intelligence to build a better future.
But because we lacked the collective willingness to share what we built.
The greatest threat from AI may not be that machines become more capable than humans.
It may be that our ability to create wealth advances far faster than our willingness to distribute it.
And by the time we recognise that as a systemic economic failure, we may already have starved the economy of the very people it exists to serve.
These are my personal reflections on AI, economic incentives and the distribution of technological progress. The Economic Starvation Loop is a proposed scenario, not a prediction supported by a settled economic model. I hope its conclusions turn out to be wrong.