As artificial intelligence begins reshaping the labour market, the real risk may lie in the turbulent gap between jobs lost and jobs created. The uncomfortable question is whether our economic safety nets are built for disruption at this speed and scale.
Somewhere around the start of last year, I began to feel anxious about artificial intelligence. Not in a Terminator-style, the-robots-are-coming kind of way, although Anthropic co-founder Dario Amodei’s prediction that AI models will one day build and control their own robots did make me feel a bit… queasy.
No, my anxiety comes from a much more practical place. I’m anxious because I struggle to understand what happens when we have more people than we have jobs for.
When I read AI-related thoughtpieces online, there is a lot of talk about the smaller workforce of the future, with parallels drawn to those shrinking birthrates that are currently alarming statisticians worldwide. It makes sense in the (very) long run: a smaller global population would equal a smaller human workforce, and the gaps left by people could be filled in by AI.
Seems like a neat solution, right?
But what I don’t quite understand is what happens now, in the year 2026, when our world population stands at 8.3 billion and the global workforce is at 3.7 billion. The clever people at McKinsey are constantly running models to try figure this out, with their latest research suggesting that between 400 million and 800 million individuals could be displaced by AI-ification in the next 4 years.
The awkward middle of automation
Of course, this isn’t the first time in human history that invention has led to job loss. We could look back at the Industrial Revolution, for example, or more recently, at the invention and widespread adoption of the personal computer.
In the US alone, about 3.5 million jobs were lost to the adoption of the computer and the internet in the 1980s and 90s. For context, the US population at the time was around 250 million, and 63% of Americans – around 160 million people – were employed. So, roughly 2.2% of American jobs were lost to new technology in the 90s.
A somewhat stark comparison to the roughly 15% of jobs that McKinsey predicts will be felled by AI, isn’t it?
But on the flipside of that coin, an estimated 19 million US jobs have been created since the 90s as a direct result of computer technology. Here we are just a few decades after the big disruption and about 10% of the civilian labour force is in an occupation that’s a direct result of the introduction of the computer.
History therefore teaches us that the computer and internet did indeed create more jobs. And how many jobs AI will create is certainly one question. But perhaps a more important question is: who will these new jobs be for?
History is far murkier on whether those who are displaced are the ones who get to claim those shiny new jobs. And with what we are currently observing in AI, this feels very different to someone learning to point and click a mouse.
The new roles that AI is expected to generate – things like prompt engineers, data annotators, AI auditors, model trainers, and robotics technicians – tend to sit on the more technical end of the skills spectrum. They reward digital fluency, adaptability, and in many cases, formal training. If you held a position as a delivery driver, a call centre agent, a retail cashier, or a bank teller – all roles that are earmarked for AI replacement – then the distance between your current skill set and the jobs being forecast is not necessarily a short one.
Entry-level white collar roles are at serious risk, as are junior professional roles that require training from senior staff.
Reskilling sounds good in theory, but in practice it can be a slow and expensive process. It also assumes people have the time, stability, and financial buffer to retrain themselves while their income is under pressure – and we are talking about a generation that can barely even afford to have children, let alone build up a balance sheet to withstand a crisis.
Labour markets do not rebalance overnight. Even if AI ultimately creates more jobs than it destroys (an outcome that is predicted, but not guaranteed), there may well be long and uncomfortable gaps in between. In those gaps, we could see mid-career workers struggle to pivot, while younger workers flood into new sectors faster than older ones can adapt.
The transition period, in other words, is where most of the real human pain tends to sit. In considering how to address this pain, we can refer to a writer named Paine. Nominative determinism never lets us down.
The original “disruption dividend”
Once upon a time (in 1795, to be precise) a political thinker named Thomas Paine published a pamphlet titled Agrarian Justice. Paine wrote his pamphlet because he was bothered by the rise of private land ownership. While he accepted that enclosing land was a logical step in the development of agriculture, he worried about what was being lost in the process: rivers, fields and orchards that had once functioned as shared sources of food were now disappearing behind fences and property lines.
In his writing, he shared his belief that those who benefited from private land ownership had some obligation to compensate the wider public, from whom the ability to sustain themselves – to hunt, fish, gather or farm on open land – had effectively been taken.
Paine didn’t stop at theory. He sketched out a concrete funding model where landowners would be taxed once per generation to support those who owned no land at all. Even in the late eighteenth century, the architecture is recognisable: not charity, exactly, but an attempt to design a system-level response to the economic dislocations created by structural change.
I heard echoes of this story when I first read the news that OpenAI’s Sam Altman was funding a basic income experiment in the US. From 2020 to 2023, Altman’s OpenResearch distributed $1,000 a month to 1,000 low-income participants in Illinois and Texas, while a control group of 2,000 people received $50 monthly. All participants were earning at or below 300% of the federal poverty line, with average annual household incomes under $29,000 (or around $2,400 per month).
The findings were intriguing. At the risk of straying from the main thread of this article, I encourage you to explore OpenResearch’s full report here. What fascinates me is the parallel between these two stories, even though they are separated by two centuries.
In 1795, Thomas Paine argued that people should be compensated as they lost the ability to sustain themselves from the land. In 2021, Sam Altman wondered whether people might one day need compensation as they lose the ability to sustain themselves through their labour. Paine didn’t try to block private land ownership, because he understood that, while it was disruptive to the status quo, it was necessary in order for agriculture (and by extension, humanity) to progress.
Similarly, Altman seems to be acknowledging that artificial intelligence will bring inevitable disruption, and that this disruption is necessary in order for our species to advance to the next phase. Like Paine, he appears to be wondering whether those who “own the land” should be responsible for addressing the gap left by their inventions.
Who actually pays for AI disruption?
Each major technological leap expands what the economy can produce, and unsettles how people earn a living along the way. Artificial intelligence appears to be following that same script, only faster. The real tension is not the long-term outcome, but the messy transition period in the middle, where labour markets tend to wobble before they rebalance.
That is why the basic income conversation keeps resurfacing. Not because it is politically easy (it is not), and not because it is proven at scale (it is not that either). An unconditional universal basic income, in its pure form, has never been successfully implemented at national level. What we have instead are pilots, partial schemes (like the ones implemented during the Covid-19 pandemic) and local experiments. And while these are all useful signals, they are far from settled policy.
Of course, even if the basic income model could be made to work inside one country, a more complicated problem sits just beneath the surface. Back in 1795, the world was a whole lot less connected. Thomas Paine was writing about taxing land within national borders to support that nation’s citizens.
Today’s AI economy does not respect those boundaries.
The companies building the most powerful systems are concentrated in the United States, but the labour disruption that those systems will trigger is spreading globally. If the “owners of the land” – the AI creators – are only taxed or regulated in their home markets, the resulting support mechanisms will also remain largely domestic.
That creates an obvious imbalance. The productivity gains may cluster in the US, where the technology is developed, while labour market disruption ripples outward into economies with far less fiscal room to cushion the blow. The geography of AI profits vs. AI pain is not set to line up.
So how does the world deal with that mismatch?
One idea occasionally raised in policy circles is some form of cross-border adjustment in the form of tariffs, usage levies or access fees on advanced AI tools deployed in foreign markets. For example, if you are a South African business preparing to replace a South African employee with technology that makes profit for an American business, the government could make you pay more for that choice through tariffs. In theory, this could help recycle a portion of AI-driven value back into the regions absorbing the labour shock, as companies will weigh up the cost of AI tokens vs. the cost of a warm body in the office.
In practice, it would be technically complex, politically sensitive and highly contested in global trade forums. But solutions seem thin on the ground right now.
The deeper we move into an AI-shaped economy, the clearer it becomes that this is not just a domestic labour story, but a global one. Technological disruption rarely waits for policy to catch up. It compounds first, and the surrounding systems adjust later.
The open question is whether we start designing those adjustments while the shift is still gathering speed, or only after the imbalances have already taken hold. Society as we know it today may depend on the answer to that question.
About the author: Dominique Olivier
Dominique Olivier is the founder of human.writer, where she uses her love of storytelling and ideation to help brands solve problems.
She is a weekly columnist in Ghost Mail and collaborates with The Finance Ghost on Ghost Mail Weekender, a Sunday publication designed to help you be more interesting. She now also writes a regular column for Daily Maverick.
Dominique can be reached on LinkedIn here.
