3/3: Will Artificial Intelligence Take Your Job?
Media sensationalism says 'YES' - History and economics say, 'NO'.
Growth Without Prosperity ~ Part III
This is the third and final part in a series of three short essays. Essay 1 explains the disease. Essay 2 shows the symptoms mutating into something more extreme. Essay 3 provides a counter argument to balance the debate.
Part I, Why Workers Keep Losing Even When the Economy Grows
Labour loses bargaining power.
The economy becomes structurally imbalanced.
Part II, AI Is Breaking the Link Between Growth and Prosperity
GDP disconnects from prosperity.
The economy begins to “boom” while society weakens underneath it.
Part III, Will Artificial Intelligence Take Your Job?
Is AI eating the world?
Or does AI create a bigger pie to feed more mouths?
Feeling Threatened By Artificial Intelligence?
One of the most common assumptions about artificial intelligence is also one of the oldest mistakes in economics.
People see a machine doing a task that previously required a human and conclude that the human will no longer be needed. It sounds logical. History suggests otherwise.
Every major leap in productivity has sparked predictions of mass unemployment. Every generation has found compelling reasons why “this time is different.”
Yet the same pattern keeps repeating. Technology automates specific tasks, lowers costs, expands access and creates far more demand than existed before.
The spreadsheet is a perfect example.
When Dan Bricklin released VisiCalc for the Apple II in 1979, accountants immediately grasped its significance. Bricklin later recalled demonstrating what the software could do almost instantly. Financial professionals would stare at the screen in disbelief before saying, “I spent all week doing that!”
The prevailing view was that computers would dramatically reduce the need for accountants. If calculations that once took days could be completed in minutes, why would companies continue employing so many people to do them?
But calculations were never the scarce resource.
The real constraint was the cost of analysis. Before spreadsheets, modelling different scenarios, updating forecasts or testing assumptions was slow and expensive. Once software removed that bottleneck, businesses didn’t perform the same amount of financial analysis with fewer people. They performed vastly more analysis than had ever been practical before.
The profession expanded rather than contracted. The United States went from roughly 340,000 accountants and accounting clerks in 1980 to around 1.4 million accountants and auditors today. Entire categories of work, from financial planning departments, quantitative analysts on trading floors, leveraged buyout modelling and sophisticated corporate planning, flourished because spreadsheets made them economically viable.
Desktop publishing followed almost exactly the same trajectory.
When PageMaker arrived for the Macintosh in 1985, many believed it would destroy professional design. Acclaimed designer Massimo Vignelli famously described desktop publishing as “a disaster of mega proportions,” arguing that giving everyone publishing tools would produce little more than visual chaos.
He was partly right.
The barriers to entry collapsed. Amateurs flooded in. Traditional typesetting jobs declined and long established unions faded away.
But what happened next was unexpected. As design became cheaper, demand exploded. Small businesses could suddenly afford marketing materials. Local organisations could produce newsletters. Startups could build brands. Millions of projects that would never previously have justified hiring a designer became worthwhile.
The profession didn’t disappear. It adapted to a much larger market. Today there are substantially more graphic designers than before desktop publishing arrived, even though the tools available to each designer are incomparably more powerful.
Radiology offers perhaps the most relevant modern comparison.
In 2016, Geoffrey Hinton (father of Artificial Intelligence and Neural Networks) argued that training radiologists would soon become unnecessary because deep learning would outperform them. His prediction became one of the defining soundbites of the AI revolution.
Almost a decade later, demand for radiologists remains exceptionally strong. Residency positions continue to increase. Leading medical centres have expanded their radiology departments. Compensation has remained among the highest in medicine. Hinton himself later acknowledged that his prediction had been too broad.
The reason is straightforward.
Reading scans is only one component of a radiologist’s role. The value lies in integrating information, exercising clinical judgement, communicating uncertainty and making decisions in complex real world settings. AI can accelerate those processes, but it also makes imaging cheaper, faster and more widely available, increasing the number of scans that clinicians can order and specialists must oversee.
Today’s AI systems are remarkably capable, but they remain highly specialised.
Whether it’s AlphaZero mastering chess and Go or ChatGPT generating text, each system excels within the domain it has been trained for.
A chess engine can learn the geometry of a 64-square board and calculate the probabilities behind millions of possible positions. A Go engine can do the same on a 19×19 grid. Yet neither understands strategy in a transferable sense. Success in one game provides no meaningful advantage in the other. Each new problem requires new training, new data, and a new framework.
That’s because modern AI doesn’t reason about the world in the way humans do. It doesn’t build a flexible mental model that can be applied across different contexts. Instead, it identifies patterns within a defined environment and optimises for a specific objective. The result is extraordinarily powerful within its domain, but surprisingly fragile outside it.
This distinction matters in the workplace.
Most valuable human work isn’t the repeated execution of a single task. It’s multi-disciplinary thinking, constructing mental models, navigating ambiguity, exercising judgement, and adapting when circumstances change. People regularly transfer lessons learned in one context to another. We improvise when rules break down. We make decisions despite incomplete information.
That doesn’t mean AI poses no threat to jobs. It clearly does.
AI will automate a growing number of routine cognitive tasks, just as machines automated many forms of physical labour. The fact that AI lacks human-level general intelligence doesn’t make workers immune from disruption. History shows that replacement doesn’t require human equivalence.
The more likely outcome is reconfiguration rather than outright substitution. AI will increasingly handle the predictable layers of many professions, while humans focus on exceptions, oversight, judgement, and adaptation. In some industries that may reduce demand for certain mid-level roles even as productivity rises.
For the foreseeable future, however, the final layer of complex work remains difficult to automate. Strategic thinking, ethical judgement, cross-disciplinary creativity, and genuine human relationships continue to resist standardisation.
The reason humans remain valuable isn’t because we can play every game. It’s because we know when the game itself needs to change.
This is the distinction many discussions about AI miss.
Technology rarely eliminates occupations wholesale. It automates individual tasks within occupations. The easier those tasks become, the more work society chooses to do.
That observation has deep roots in economics. In the nineteenth century, William Stanley Jevons noted that improvements in steam engine efficiency did not reduce Britain’s coal consumption. They increased it. More efficient engines lowered costs enough to make entirely new applications commercially attractive, causing overall demand for coal to surge.
The same dynamic appears repeatedly throughout history.
Lower the cost of computation and businesses run more models. Lower the cost of design and organisations create more content. Lower the cost of medical image interpretation and healthcare systems perform more diagnostic imaging.
Artificial intelligence is likely to follow the same path.
As the cost of producing software, research, legal drafting, marketing campaigns and creative content falls, demand for those outputs is unlikely to remain fixed. Companies will build tools they previously couldn’t justify. Entrepreneurs will launch products that were uneconomic before. Individuals will access expertise that was once available only to large enterprises.
Some jobs will certainly disappear, just as typesetters were displaced by desktop publishing. But history suggests that focusing only on the tasks machines replace misses the bigger story.
The more important effect is that lower costs expand markets. They bring in new users, enable new business models and create entirely new categories of work. Professionals spend less time on repetitive production and more time applying judgement, taste, context and accountability.
If previous technological revolutions are any guide, AI will not simply divide the existing pie differently. It will make the pie much larger.
However, whether or not AI ultimately replaces jobs is arguably the wrong question.
With reference to the two prior essays in this series, the more important question is who captures the economic value that AI creates.
Throughout history, technological progress has increased productivity, generated more wealth than the one before it, and introduced higher standards of living across the population. Think railroads, electricity, aviation and the internet.
Today looks very different. The economic challenge that flows from the introduction of AI is not that society becomes less productive. Quite the opposite. Society may become vastly more productive.
But Keynes understood that economies depend on aggregate demand. Businesses invest because consumers spend. Consumers spend because they earn income. The circular flow of money is what keeps the system functioning.
A machine can generate output. It can’t consume it.
Whether AI takes our jobs or not becomes a secondary issue. The primary issue is that the fruits of our enhanced productivity are flowing into concentrated pockets which undermines Keynsian economics entirely.
Economies require broad based consumption, which in turn depends on a more even distribution of money. Yet more and more of the benefits of production are flowing to capital, evidenced by the growing number of individuals with net worths measured in tens or even hundreds of billions of dollars.
This is not the cause of the problem. It is a symptom of it.
The deeper issue is whether the distribution mechanism that converts productivity into broad-based purchasing power continues to function.
If too much income accumulates at the top, consumption eventually becomes constrained. A billionaire does not consume one thousand times more groceries, airline tickets, or restaurant meals than a millionaire. At some point, additional wealth is invested rather than spent.
Businesses ultimately require customers, not merely investors.
An economy cannot thrive indefinitely if productive capacity grows faster than the purchasing power of the population it serves.
This is why the AI debate extends far beyond technology.
It is not a technological problem. It is an economic and political one.
The social fabric does not break because machines become productive. It breaks when large numbers of people feel disconnected from the prosperity those machines create.
History suggests societies can tolerate high levels of inequality for surprisingly long periods. What they struggle to tolerate is a system that no longer offers participation.
That is the question policymakers, investors, and business leaders will increasingly confront over the coming decade.
What do you think? Please leave a comment:
This is the third and final essay in this series, if you haven’t read the other two, please circle back.
Some time ago I was digging on the academic literature on forecasting and how to be better at making predictions. One of the reasons why we are so bad at making predictions is that we think we know better and see relationships of cause and effect, while omitting important variables, when we could reach more accurate predictions by just inferring the answer (in other words, look at similar situations that happened in the past). By following this logic, you are 100% right. Every time that technology has lowered the cost of production, demand has followed, and new jobs have been created more than making up for the losses.
On a more fundamental perspective though, and by applying the same line of reasoning, I wonder what will AI produce at a lower cost, that we will want to have lots of in the coming years and decades? Imagine that for example, you start using AI to research companies and write articles. Now we have one new article every day. Who’s going to read all of that? We are already swimming in content from hundreds of writers. Imagine that it’s 2050 and an AI is monitoring you while you watch TV, measuring every movement in your face, gathering data about what you like, then putting in front of you some content that is tailor made to your linking and your preferences and vibed into existence in real time. Does this sound utopian or dystopian to you? I think you can see where I am going. The human touch is lost, and as content becomes abundantly available, its value dilutes. The machine starts making more decisions for you. Human cognition is impacted. We are already seeing the effects. No wonder that a lot of people have started to hate AI.
If AI replaces jobs that impact cognitive workers like engineers, economists, etc, perhaps we can have more people doing science, art and philosophy. That does not sound dystopian to me. Being a cognitive worker today typically means sitting in meetings and sending emails. We can use our cognitive capacity for much more.
Alternatively, if the AI job apocalypse materializes, we might need to eventually impose a form of universal free income to keep the economy moving.
Thanks for the very good analysis of AI.