In This Article
I don’t know whether AI could kill humanity. Nobody knows. What I am asking is whether it could fundamentally change the conditions under which today’s human society exists.
Former Anthropic researcher Jacob Coxon recently appeared on CNN to explain why he left one of the world’s leading artificial intelligence companies.
His warning was dramatic. Coxon believes that increasingly capable AI systems could eventually escape meaningful human control and potentially become an existential threat to humanity. His concerns include autonomous systems, recursive self-improvement and the possibility that future AI could acquire capabilities far beyond what humans can safely control. His resignation attracted unusual attention partly because he reportedly left Anthropic only about two months before a significant equity vesting event. He was willing to walk away from that financial benefit because he believed the direction of AI development was dangerous.
I watched the interview and found myself thinking about something else. I don’t know whether AI will ever develop a biological weapon, I don’t know whether a future superintelligence will attempt to manipulate humans, replicate itself, escape containment or eliminate humanity, I don’t know whether the more extreme AI extinction scenarios being discussed today have a 1%, 10% or 50% probability, and I don’t think anybody honestly knows, but there is another AI danger that requires no hostile superintelligence, no consciousness, no biological weapon and no science-fiction scenario.
And it is already visible.
What happens if AI simply becomes better and cheaper at doing human work than humans are?
And then another question immediately follows: What happens if 20%, 30% or eventually more of the population no longer has economically necessary work?
Because those people are not only employees. They are also customers, tax payers, they pay mortgages, insurance and electricity bills, they rent homes, they buy food and cars, they go to restaurants, they travel, they raise children. They keep banks, retailers, manufacturers and governments functioning.
So if AI replaces them at scale, the question is not simply whether companies can save money.
The question becomes: Who will buy everything the automated economy produces?
That may be one of the most important unanswered questions in the entire AI debate.
The AI apocalypse social media keep discussing may not be the most immediate one
The current AI safety debate has increasingly focused on existential risk.
Jacob Coxon is one of the latest people to bring that debate into mainstream television. His warnings have been accompanied by concerns from other AI researchers about whether future systems could become uncontrollable. One current Anthropic researcher, Evan Hubinger, has publicly discussed a greater-than-10% probability of an AI catastrophe within the next decade, although such probabilities are inherently speculative rather than measurable forecasts.
These concerns deserve serious consideration.
But there is an important distinction between: “AI could eventually become capable of destroying humanity.”,
and: “AI could gradually change the economic structure that humanity depends upon.”
The second scenario does not require AI to hate us. It doesn’t require AI to become conscious. It doesn’t require AI to develop a survival instinct. It doesn’t require AI to decide that humans are inefficient.
Humans can make the decision for it. And they do it!
A company can simply ask: How can we reduce operating costs by 30%?
AI will provide the answer, and the human greed will do the rest of the job.
The 10,000-employee thought experiment
Imagine a large company employing 10,000 people.
Those employees collectively receive hundreds of millions of euros or dollars in salaries.
They pay taxes, pay mortgages, buy products, support families, spend money in other businesses. Those other businesses employ other people.
The government collects taxes from all of this economic activity, banks receive mortgage payments and interest, retailers receive consumer spending, restaurants receive customers, manufacturers sell products.
The economy is a circular system, and then increasingly capable AI agents arrive.
The company discovers that it can automate a large portion of the work. Probably not everything, few humans remain necessary for oversight.
Probably the company needs only 2,000 employees instead of 10,000, and they lays off 8,000 people.
The immediate financial result looks fantastic. Labour costs fall, margins rise, profit increases, shareholders are pleased, and CEO is praised for “AI transformation.”
The stock price may rise. From the company’s perspective, the decision can be perfectly rational.
But now look at the other side of the spreadsheet.
Those 8,000 people are no longer receiving salaries, they pay less income tax, they contribute less to social security, they spend less money, some stop paying mortgages, some postpone buying cars, some stop travelling, some stop eating at restaurants, some reduce insurance coverage, some move into cheaper housing, some stop buying things they previously considered normal, some need unemployment benefits or other government assistance.
And now ask: Who actually won? The company? Initially, probably. The shareholders? Probably.
The former employees? Obviously not. The government? It has lost part of its tax base and potentially gained additional welfare costs. The bank? It now faces greater credit risk. The retailer and the restaurant? They have fewer customers. The manufacturer? It may have cheaper production but fewer buyers. The economy? It may have higher productivity but weaker purchasing power.
This is the paradox that receives surprisingly little or none attention.
Overnight, whole economy remains on shaky legs.
The employee is also the customer
A worker is not simply a cost on a company’s balance sheet. A worker is also part of the company’s market. This is one of the most important things to remember when discussing mass automation.
If a factory employs 10,000 people and pays them enough to purchase the products of thousands of other companies, those wages are not simply an expense.
They are part of the economic circulation that creates demand. Remove enough wages and you eventually remove demand.
That creates a strange feedback loop: Automation → fewer wages → less consumption → weaker demand → lower revenues → more cost pressure → more automation.
This does not automatically mean the economy collapses. Economies are far more complicated than a simple circular diagram.
Productivity can create new industries, prices can fall, new jobs can emerge, investment can increase, new forms of work can appear.
Governments can redistribute income, companies can invest their additional profits, and international demand can compensate for domestic weakness.
All of those things are possible.
But there is a threshold beyond which the traditional assumption becomes difficult: People will always find another job. What if they don’t?
“People will just do different jobs”
This is the standard response to technological unemployment, and historically, it has often been correct.
Agricultural mechanisation destroyed enormous numbers of agricultural jobs, factories destroyed some crafts and created other industrial jobs, computers destroyed some administrative work while creating enormous new industries, the internet eliminated some professions while creating others.
So why should AI be different?
Because AI is not only another machine replacing physical labour.
It is a machine increasingly capable of performing cognitive labour.
It can write, analyse, research, code, translate, design, perform customer service, conduct financial analysis, create marketing materials, perform parts of legal research, analyse documents, perform increasingly sophisticated forms of knowledge work, and not only.
And when AI is combined with robotics, the boundary between cognitive and physical automation becomes much less meaningful.
A tractor replaced part of the farmer’s physical labour. An AI system can potentially replace parts of the farmer’s planning, administration, marketing and analysis. A robot can eventually handle the physical component. The technological direction is therefore different.
It attacks both sides of the labour equation.
SEO is a small but very visible example
I have spent decades around SEO. Classic SEO contains an enormous amount of repeatable cognitive work.
Keyword research, SERP analysis, competitor analysis, content briefs, content production, internal linking, technical audits, schema analysis, reporting, content optimisation… Many of these activities can already be performed substantially faster with AI. And cheaper than one senior expert.
And companies do not necessarily ask: Can AI completely replace an experienced SEO professional?
They ask a different question: Can one SEO professional using AI do the work previously performed by five people?
That is a much more realistic economic question.
And in many situations, the answer is increasingly yes. The consequence is not necessarily that SEO disappears.
The consequence is that the economic value of performing individual SEO tasks decreases.
The same process is happening across many knowledge professions. That distinction matters.
AI does not have to eliminate an occupation to reduce the number of people required to perform it.
If five people can become one highly augmented worker, four positions have effectively disappeared even though the occupation still exists.
This is where the labour statistics become more complicated
It would be irresponsible to look at every recent corporate layoff and say: “AI did it.” That’s not supported by the evidence.
Companies lay people off for many reasons: economic downturns, over-hiring, restructuring, mergers, changing business strategies, shareholder pressure, interest rates, declining demand, outsourcing, automation, AI, and others.
And even Anthropic’s own research makes an important distinction between what AI systems theoretically can do and what businesses are actually using them to do today.
Anthropic’s labour-market research says AI’s observed exposure is still only a fraction of its theoretical capability, while occupations with greater observed exposure are projected by the U.S. Bureau of Labor Statistics to experience weaker employment growth.
That is much more interesting than claiming that AI has already destroyed the labour market. The evidence does not yet show that.
What it shows is something potentially more important: The capability is arriving before the economic consequences have fully arrived.
Anthropic itself is modelling this problem
This is not merely a concern raised by AI critics.
Anthropic’s own economics team has created scenarios examining how AI could affect employment, GDP and unemployment in the United States.
The scenarios range from relatively modest impacts to highly transformative ones.
In more extreme scenarios, AI could have effects comparable to major technological transformations, with significant disruption to white-collar employment and increasing wealth flowing toward investors. One recent analysis of Anthropic’s scenarios described potential unemployment above 20% in particularly transformative cases, while stressing that these are scenarios, not predictions.
That distinction is critical. A scenario is not a forecast, but a scenario tells us something valuable:
The people building frontier AI are themselves modelling the possibility that AI could fundamentally change employment and wealth distribution.
We should probably pay attention.
Then Elon Musk says something even more radical
Elon Musk has repeatedly argued that AI and robotics could eventually make human work optional.
In January 2026, Musk said that within 10 to 20 years, work could become optional and money could become irrelevant because AI and robotics would produce enormous abundance.
At first glance, this sounds wonderful. Imagine it. Robots produce food, build houses, manages logistics, manufacture products, AI writes software, performs accounting, handles administration, performs research, machines maintain infrastructure.
Humans no longer need to spend their lives working simply to survive. That could be one of the greatest achievements in human history.
But there is an enormous unanswered question: How do humans obtain purchasing power?
If nobody works, who buys the products?
This is the question I kept returning to.
Suppose Musk’s prediction becomes broadly correct. Suppose AI and robots can perform almost every economically necessary task. Suppose human labour becomes optional.
Companies can produce extraordinary quantities of goods, production becomes incredibly cheap, productivity explodes, and the economy becomes enormously productive.
Wonderful.
Now imagine that 30% of the population has no conventional employment. Then 50%. Then probably 70%. Who buys the products?
The answer cannot simply be: “The market will figure it out.”
The market needs buyers with purchasing power. Without it, production doesn’t have sense. And purchasing power traditionally comes overwhelmingly from income.
If income from employment disappears, another mechanism must replace it, and that is not a minor detail.
And all that army of unemployed people, with their families, have to live somewhere, and eat something every day, while waiting for the “golden age” promised by Elon Musk. It is the central economic problem of the post-labour future.
The worker disappears, but the consumer cannot
This is the paradox.
A company wants to eliminate the cost of an employee, but society still needs that person to have enough income to consume.
The employee is simultaneously: a cost of production and a source of demand.
Remove too much labour income and the economy has to find another mechanism for distributing purchasing power.
Otherwise, the automated economy risks becoming extraordinarily good at producing things that fewer and fewer people can afford to buy.
That is the strange possibility: A world of unprecedented abundance with insufficient purchasing power.
Not because there aren’t enough goods, but because there isn’t enough money distributed among the people who need them.
And this is where the political question becomes unavoidable
There are several possible solutions.
Governments could redistribute part of the productivity gains.
2. They could tax capital more heavily.
3. They could introduce some form of universal basic income.
4. They could introduce a negative income tax.
5. They could provide universal basic services.
6. They could create sovereign wealth funds that own shares in productive AI infrastructure.
7. Workers could receive ownership stakes in automated companies.
8. Working hours could be dramatically reduced.
9. Instead of eliminating 30% of jobs, society could reduce everyone’s working time.
10. Probably people work two days a week instead of five.
11. Probably the concept of retirement changes completely.
12. Probably income increasingly comes from ownership rather than employment.
13. Probably several of these mechanisms operate together.
But notice what has happened.
The question is no longer primarily: “How do we build better AI?”
It becomes: “How do we distribute the economic value created by AI?”
That is a political question, economic question, governance question, and ultimately a sovereignty question.
There is an uncomfortable historical irony here
This is where comparisons with communism sometimes appear, but saying simply that “Musk is pushing communism” would be too crude.
The interesting comparison is much more precise. Classical communist theory sought to separate human access to the necessities of life from private ownership of productive labour and capital.
The AI future being described by some technology leaders could also separate survival from employment.
People would not necessarily need jobs because machines would produce what society needs.
But there is a fundamental difference: Who owns the machines? That is the question.
A future in which productive assets are collectively or broadly owned is one thing, and the future in which AI, robots, semiconductor factories, data centres, energy infrastructure and the most productive companies are owned by a tiny number of individuals and corporations is something else entirely.
Both could produce enormous abundance, but they could also produce radically different societies.
So the real political question is not: “Will AI create communism?”
It is: “Who will own the productive machinery of the AI economy, and how will everyone else participate in its wealth?”
That is a much more serious question.
The billionaire problem
Imagine that AI makes a company worth ten times more, and the shareholders become enormously wealthy.
Now imagine thousands of companies doing the same.
The people who own AI systems, robotics companies, semiconductor infrastructure, energy infrastructure and financial capital become extraordinarily wealthy.
That could produce a concentration of productive power unlike anything seen before.
But there is a strange limit to extreme wealth.
A billionaire cannot personally consume one billion people’s worth of food he cannot sleep in one billion houses, he cannot drive one billion cars simultaneously, he cannot eat one billion restaurant meals, he cannot personally use all the products that his wealth allows him to purchase.
So if productivity becomes extremely high while purchasing power becomes concentrated among a very small group, the economy faces an unusual problem.
The ability to produce things can grow much faster than the ability of ordinary people to buy them.
That is why distribution is not simply a moral question, but it becomes an economic question.
The government has a problem too
Imagine a future in which a substantial proportion of workers no longer earn conventional salaries.
Governments lose: income tax, payroll tax, social contributions, consumption taxes generated by wages, taxes associated with housing transactions, various employment-related revenues
At the same time, governments may have to spend more on: unemployment, welfare, housing, healthcare, retraining, social support, infrastructure
The tax base changes. That creates another fundamental question: What exactly should governments tax in a highly automated economy?
Labour? There is less of it.
Consumption? People have less income.
Corporate profits? Probably.
Capital? Possibly.
AI infrastructure and compute? Perhaps.
Land and energy? Perhaps.
This is not a technical problem that can be solved by writing a better tax form. It is a structural transformation.
Banks are part of the equation too
Consider a mortgage. A bank lends €300,000 to a person because that person has a reliable income, and that person spends decades repaying the loan.
Now imagine that AI eliminates the person’s occupation. The person does not immediately disappear, but the income disappears. Then the mortgage becomes difficult to service. Multiply that by millions of households.
The result could affect: mortgage defaults → property prices → bank balance sheets → lending → investment → construction → employment
Again, none of this is inevitable.
Banks have capital buffers, governments can intervene, people can find new work, interest rates can change, property markets can adjust.
But the point is that mass automation would not affect only employees. It would affect the entire financial architecture built around human income.
And there is an even more basic problem
People need food, water, shelter, energy.
A person can survive much longer without food than without water, but there is no universal “40 days without food and seven days without water” rule. Survival varies enormously depending on health, temperature, exertion and access to fluids. But the exact number is not the point.
The point is that human beings have physical requirements regardless of how sophisticated the digital economy becomes. AI does not remove biology, and this leads to another paradox.
AI is not weightless
From a laptop, AI looks almost magical. You type a question, in seconds answer appears. It feels like pure information.
But behind that answer is an enormous physical infrastructure: data centres, semiconductors, electricity, cooling, land, construction, networks, mining, manufacturing, water. AI has a physical body, and that body is getting larger.
Data centres can consume enormous quantities of electricity and, depending on their cooling systems and location, substantial amounts of water. The water question is particularly important in regions already facing scarcity.
So while AI promises a world where humans no longer need to work as much, the machines themselves require enormous physical resources, which creates another question: What happens when AI infrastructure competes with humans for scarce resources?
Water is an obvious example. Energy is another. Land is another. Semiconductors are another.
The AI economy is not detached from the physical world, it is becoming deeply embedded in it.
The danger is not necessarily AI versus humanity
This is where I disagree with the most dramatic version of the AI-extinction narrative.
The story often looks like this: AI becomes superintelligent → AI becomes autonomous → AI develops goals → AI decides humans are a problem → AI develops biological weapons → humanity disappears.
Perhaps something like that could happen. I don’t know, and I don’t think we should pretend that we know.
But there is another pathway, much more realistic: AI becomes extraordinarily capable → companies automate → labour demand falls → income becomes concentrated → purchasing power falls for displaced workers → tax systems weaken → governments struggle to redistribute the gains → social instability increases.
No evil AI required, no consciousness required, no hatred required, no biological weapon required. Just economics.
And this may actually be more difficult to control
A hostile AI can theoretically be contained. A company making a bad decision can theoretically be regulated.
But what do you regulate when every individual company has a rational reason to automate?
Imagine Company A decides to preserve 10,000 jobs. Company B replaces 10,000 employees with AI and cuts costs by 40%. Company B now has a massive competitive advantage. Company A’s shareholders demand similar efficiencies. So Company A automates too.
Then Company C does the same. Then Company D.
Nobody needs to coordinate, nobody needs to intend harm. The system itself creates the pressure, and this is a classic collective-action problem.
Each company can behave rationally while the aggregate result becomes socially destructive.
This is why “AI will create more jobs” is not enough
Probably it will. Probably entirely new professions will appear, probably humans will become more creative, probably people will work alongside AI, probably productivity gains will create such abundance that new markets emerge. All of that is possible.
But there is a difference between: > AI creates new jobs, and: AI creates enough new economically valuable jobs to compensate for the jobs it eliminates.
Those are not the same statement.
And if AI eventually becomes capable of performing most economically valuable cognitive work, the old historical assumption becomes much harder to rely on.
Previous technological revolutions created new categories of human work partly because machines remained limited.
What happens when the machine itself can participate in the creation of the next machine?
The recursive problem
This is also where Coxon’s concerns become relevant. One of his major fears involves recursive self-improvement: systems becoming increasingly capable of contributing to the development of their successors, whether that ultimately produces an uncontrollable superintelligence remains uncertain.
But economically, even a much less dramatic version matters.
Imagine AI helps engineers build a better AI, the better AI helps engineers build an even better AI, that system helps automate more work, the automation produces more capital, the capital funds more AI infrastructure, the infrastructure produces more capable AI.
This can create a positive feedback loop: AI capability → productivity → capital → infrastructure → better AI → greater productivity.
If that loop becomes sufficiently strong, human labour may not remain the central source of economic growth. That would be a historic break.
We should not pretend that this has already happened
This is important.
The evidence today does not justify saying: “AI has already replaced humanity.” It hasn’t.
Most people still work, most companies still depend heavily on humans, AI adoption is uneven, many AI systems still require substantial human supervision, many companies are experimenting rather than replacing employees, and current labour data do not show anything remotely resembling 30%, 50% or 70% unemployment caused by AI.
The transformation is real, but the endpoint is not known. That distinction is essential. If we exaggerate the evidence, we weaken the argument.
The strongest argument is not: “The apocalypse has arrived.”
It is: “The direction of travel is changing faster than the institutions built around human labour can adapt.”
The same problem exists at the individual level
And there is another uncomfortable part of this story. People like me are not outside this transformation.
I have spent decades working with information, search, SEO and knowledge systems.
AI can perform parts of that work. It can research, analyse, write, identify patterns, optimise, and automate. The threat isn’t only theoretical.
Anyone whose economic value is based primarily on performing repeatable cognitive tasks has a reason to pay attention.
That includes SEO specialists, writers, analysts, customer-service workers, designers, programmers, researchers, accountants, administrators, consultants, translators, marketing professionals, lawyers, financial analysts, and eventually many other professions.
The rational response isn’t necessarily to fight AI.
It may be to move toward the areas where human judgement, responsibility, ownership, governance, trust and context remain valuable, but that transition will not be equally easy for everyone.
Some people will adapt quickly, some will struggle, some will not be able to adapt at all. And that is where the human cost enters the discussion.
AI can hurt people without physically attacking them
This distinction matters.
Calling unemployment “killing people” literally would be imprecise, but economic displacement can have very serious human consequences.
Loss of employment can lead to: loss of income, housing insecurity, debt, family stress, loss of healthcare access in systems tied to employment, reduced nutrition, reduced education opportunities, mental distress, social isolation, declining life prospects. The effects compound.
A person who loses a job does not necessarily die, but a society that permanently removes economic security from a large percentage of its population can experience consequences far beyond the employment statistics.
The human body still needs food and water, the human family still needs housing, children still need education, people still need dignity, and humans still need a reason to believe that the future contains a place for them.
Probably the biggest risk is not unemployment
It is irrelevance. A person can survive unemployment. What is much harder is the feeling that society no longer needs what they can do.
For centuries, economic participation has provided more than money.
It provided: identity, status, structure, social connection, purpose, independence, dignity
If machines perform most economically valuable work, society will have to solve more than the income problem. It will have to solve the meaning problem.
What does a human life look like when employment is no longer necessary?
That could become one of the greatest philosophical questions of the century.
So what would a successful AI transition look like?
There is a much more optimistic version of this story.
Imagine AI genuinely produces enormous abundance, healthcare becomes cheaper, education becomes personalised and widely available, food production becomes more efficient, construction becomes cheaper, energy becomes more abundant, transportation becomes safer, people work fewer hours, parents have more time with children, older people receive better care, scientific research accelerates, people spend more time creating, learning, travelling, caring and exploring.
That would be extraordinary.
But it requires one thing that technology alone cannot provide: Distribution.
Productivity is not prosperity unless the benefits reach society. A machine can produce a thousand times more than a human, but that does not automatically mean the human receives a thousand times more. Ownership determines who captures the productivity, governance determines how much is redistributed, markets determine who can purchase the output, politics determines what rules apply, and society determines what it considers acceptable.
Would this outcome produce the better opportunities and equality for humans? I don’t know! Maybe you have the answer.
The question of ownership may become more important than the question of employment
For most of modern history, the formula has been relatively simple: Work → income → consumption
In a highly automated economy, another formula may become more important: Ownership → income → consumption
If that happens, the biggest economic question may no longer be: “What job do you have?”
It may become: “What do you own?” Do you own shares? Do you own productive infrastructure? Do you own intellectual property? Do you own land? Do you own energy production? Do you own AI systems? Do you own data? Do you own access to compute? Do you own a share of the automated economy?
That would represent a profound transformation of capitalism.
And this brings us back to AI sovereignty
This is also why the discussion about AI sovereignty cannot be reduced to where a model is hosted.
Sovereignty is ultimately about control over critical productive and informational infrastructure.
1. Who owns the models?
2. Who owns the data centres?
3. Who controls the chips?
4. Who controls the energy?
5. Who controls the information?
6. Who controls the knowledge layer?
7. Who controls access?
8. Who captures the economic value?
9. Who sets the rules?
10. And who remains dependent on everyone else?
These questions matter at the level of companies, at the level of individuals, at the level of countries, and they will matter even more if AI becomes the primary productive infrastructure of the global economy.
Jacob Coxon may be right about the danger, but perhaps not about the danger we should fear most
I don’t know whether Coxon’s extinction scenario will happen, or whether AI will develop biological weapons, or whether an artificial superintelligence will become impossible to control, or whether humanity will face an existential AI event before 2030. Those are enormous claims, and certainty would be dishonest.
But something else is already happening.
AI is becoming capable of performing increasing amounts of economically valuable human work.
Companies have a financial incentive to automate.
Investors reward productivity.
AI companies are racing to increase capability.
Robotics is advancing alongside AI.
Governments are struggling to build policy fast enough.
And even the leaders of AI companies are increasingly acknowledging that the pace of development creates risks that society may not be prepared to manage.
On September 12, 2026, Anthropic CEO Dario Amodei publicly called for AI companies to slow the development of increasingly powerful models, arguing that progress needs to be accompanied by more time for risk alignment and independent evaluation.
That is not a fringe activist speaking. That is the CEO of one of the companies at the centre of the frontier AI race.
Are we asking the wrong question?
The question dominating headlines these days is: Will AI kill humanity?
Maybe! Maybe not!
But there is another question that deserves just as much attention: What happens if AI becomes so productive that human labour is no longer economically necessary?
And another one: Who owns the machines?
And another one: Who receives the purchasing power?
And another one: Who pays the taxes?
And another one: Who pays the mortgages?
And another one: Who buys the products?
And another one: Who controls the infrastructure?
And another one: What is the role of a human being in an economy that no longer needs human labour?
And probably 100 more questions I have at this moment in my head.
These questions are connected, and they are not separate AI problems. They are one single problem.
The machines don’t have to destroy us
There is an irony in the current AI debate.
We spend enormous amounts of time imagining an AI system becoming conscious, developing hostile intentions and deciding that humanity must be eliminated. That may be the most dramatic scenario.
But probably the more immediate danger is considerably less cinematic, where machine doesn’t have to hate us, doesn’t have to want anything, doesn’t have to understand what it is doing. It only has to become cheaper, faster and more capable than us at enough economically valuable tasks.
Humans will then make the decisions.
Companies will automate because competitors automate, investors will demand productivity, governments will struggle to replace lost tax revenues, banks will continue demanding mortgage payments, consumers will need purchasing power. and the owners of the most productive systems will capture an increasing share of the economic value.
No villain is required. No evil machine is required. No biological weapon is required.
The system can arrive there through millions of rational decisions.
The real AI race may therefore not be between humans and machines
It may be between two possible futures.
In the first: AI increases productivity, wealth becomes concentrated, human labour loses value, purchasing power becomes concentrated, governments react too slowly and society becomes increasingly divided between those who own the machines and those who depend on them.
In the second: AI increases productivity, working hours fall, ownership becomes broader, productivity gains are distributed, essential services become cheaper, people retain economic security and humans use their additional time for things that machines cannot meaningfully replace.
Both futures are technologically possible, and neither is predetermined.
And that is perhaps the most important point. AI itself does not decide which future we get. Humans do. We do! Today!
The danger is not only what AI can become
Danger is in what humans decide to do with what AI can already do.
We should absolutely continue researching alignment. We should take autonomous AI agents seriously. We should investigate cybersecurity risks. We should develop safeguards around biological and chemical misuse. We should build independent evaluation. We should discuss international governance. And we should listen when people working inside frontier AI companies raise serious concerns. But we should also look at the danger directly in front of us.
The danger of an economic system where machines can produce more while humans have less purchasing power, the danger of wealth becoming concentrated among those who own the productive infrastructure, the danger of governments losing tax revenue faster than they can redesign taxation, the danger of financial systems built around stable employment facing a world of unstable employment, the danger of communities losing not only jobs, but purpose and economic relevance, and the danger of building a machine economy before deciding what humans are supposed to receive from it.
Because there is one question nobody can avoid forever
Suppose Musk is right.
Suppose, within a couple of decades, AI and robots really can perform almost all economically necessary work.
Suppose food becomes extraordinarily cheap, manufacturing becomes almost entirely automated, software is generated by AI, transportation is autonomous, construction is robotic, scientific research is largely AI-assisted, human labour becomes optional.
Then we have achieved something humanity has dreamed about for thousands of years. Machines work. Humans don’t have to.
But then comes the question that determines whether this becomes paradise or disaster: Who owns the machines?
And immediately after that: How does everyone else obtain the purchasing power to participate in the abundance those machines create?
If we answer those questions intelligently, AI could become one of the greatest achievements in human history.
If we don’t, we may create the most productive economy humanity has ever seen… while simultaneously creating a society in which millions or billions of people can no longer afford to participate in it.
And probably that is the AI danger we should be discussing more urgently, not whether the machine will eventually decide to kill us, but whether we will build an economy that no longer needs us, without first deciding how humans are supposed to live in it.
Decision is on us. Today. Not tomorrow.
