AI could replace the income people need to afford AI itself.
Unlike earlier automation, general AI can move across many kinds of work.
Cheaper and open models may reduce the access problem, but not eliminate it.
If AI creates most of the productivity, people may need to share in that value through dividends, UBI, or similar systems.
Basic AI access could eventually become infrastructure, much like internet access.
I want AI to reduce how much we need to work, but that only works if losing work does not also mean losing the ability to live.
I have been thinking about a strange problem with the way we talk about AI and jobs. Most conversations focus on whether AI will replace people, which jobs will disappear first, and how quickly all of this might happen. I think there is another question that is just as important. What happens if AI becomes capable enough to replace the work I do for money, but accessing that same level of AI still requires money?
This creates a fairly simple loop. I need a job because I need money to live, but I may also increasingly need AI to remain productive enough to compete for that job. If AI eventually becomes capable enough that a company no longer needs me, I lose the income I was using to pay for the technology that is now capable of doing my work. The tool that reduces my ability to earn money can therefore become the same tool I need money to access.
I do not have a problem with AI replacing work in principle. In fact, removing work that people do only because they need an income is one of the things that makes AI exciting to me. I would happily spend less of my life doing repetitive work if that gave me more time to build things, learn, travel, experiment with ideas, change direction when I want to, and generally live a more iterative life. My problem begins when we assume that removing the work automatically gives people that freedom, because in our current economy the work and the income attached to it are still what allow most people to live.
The way I would put it is this:
We have dealt with automation before, so it is reasonable to look at history and say that we will adapt again. Factory machines replaced certain kinds of physical labour, ATMs automated parts of banking, self checkout machines reduced some retail tasks, and software removed a huge amount of administrative work. Those changes disrupted jobs, but they generally automated particular tasks while leaving plenty of other work that humans could move into. When one skill became less valuable, learning another skill was often a reasonable response.
AI makes that response more complicated because the technology is not limited to one narrow category of work. The same general systems are becoming capable of writing, programming, researching, designing, analysing information, creating images and video, communicating with customers, operating software, and handling increasingly complicated digital tasks. If AI replaces one kind of work and I retrain for another kind of work that AI is also rapidly learning to perform, then telling people to simply reskill starts becoming less convincing.
This does not mean that I think every job is going to disappear. There will almost certainly be new jobs created around AI, just as previous technological changes created work that would have sounded strange a generation earlier. Humans will still be needed to verify outputs, make decisions, provide taste and judgment, understand what people actually want, manage relationships, and take responsibility when something goes wrong. There is a meaningful difference between producing an answer and being responsible for what happens because of that answer, and I do not expect that difference to disappear quickly.
What I am less confident about is whether those new jobs will appear quickly enough, in large enough numbers, and with low enough barriers for everyone whose existing work becomes less valuable. A company might once have needed ten people to produce a certain amount of work and later need four people using AI to produce twice as much. Those four people may become dramatically more productive and may even earn more because of it, while the company also becomes more profitable. That can be a genuine success for the technology without answering what happens to the other six people.
There is a strong counterargument to my concern, which is that AI itself is becoming cheaper. The amount of intelligence you can access for a certain amount of money has improved enormously in a very short period of time. Capabilities that once required expensive frontier models gradually move into cheaper models, free tiers, smaller systems, and eventually hardware that ordinary people can own. There is no reason to assume that today's expensive AI subscription will remain expensive forever, just as many technologies that were once limited to wealthy people eventually became ordinary consumer products.
Open models make this even more important. Models from Meta, Mistral, DeepSeek and others have made it possible to access increasingly capable AI without always paying for the most expensive commercial service. Smaller models are also becoming useful enough that many everyday tasks do not require the largest model available. Better hardware, more efficient software, competition between companies, and open development can continue pushing the cost of useful intelligence downward, which could substantially reduce the affordability problem I am describing.
I think that is a legitimate reason to be optimistic, but cheap access and equal access are not necessarily the same thing. If I can use a free model that helps me write emails and summarise documents while someone else can afford an AI system capable of independently researching markets, writing and testing software, managing business processes, creating media, and completing hours of professional work, we technically both have access to AI but we do not have access to the same economic capability. The important question is therefore not simply whether everyone can use some form of AI, but whether the difference between what people can afford becomes large enough to affect their ability to earn money in the first place.
This is where the possibility of a two-tier society starts to concern me. People who can afford the best systems can use those systems to become more productive, which can help them earn more money, which then makes it easier for them to afford even better systems and more computing power. Someone who cannot afford the same tools may become less competitive even if they are equally intelligent and equally willing to work. The technology itself can then amplify an economic advantage that already existed.
The issue becomes even larger when we think about who actually owns the technology. If a relatively small number of companies own the models, chips, data centres and infrastructure responsible for a growing amount of economic production, then AI could create enormous amounts of wealth while requiring fewer people to participate directly in creating it. At that point, unemployment is only part of the question because the larger question becomes how ownership of productive technology is distributed and who receives the wealth it creates.
One possible answer is some form of AI dividend. The basic idea is that if AI creates an enormous increase in economic productivity, ordinary people should have some way of sharing in that increase rather than receiving the benefits only through employment. Alaska's Permanent Fund is often used as an example of the broader principle because part of the wealth generated from the state's natural resources is invested and residents receive a dividend. Sam Altman has also written about ways technological progress and capital could be distributed more broadly as automation increases productivity. AI is obviously different from oil, but the underlying question is similar: if society possesses something extraordinarily productive, how widely should the value it creates be shared?
Universal basic income is another possibility that becomes more interesting in this context. If machines eventually perform a larger share of economically valuable work, governments may have to rely less on systems built around taxing human labour and think more seriously about taxing the profits or economic output created through automation. Some of those gains could then support a basic income or another form of social dividend, allowing people to continue participating in the economy even when there is less demand for human labour.
I do not see this as paying everyone to do nothing, which is often how the idea gets reduced in political arguments. If technology genuinely allows society to produce more goods and services with fewer hours of human labour, then insisting that everyone must continue selling the same amount of their time simply to qualify for food, housing and basic security would be a strange way to use that technological progress. The point of becoming more productive should eventually be that we need to work less to maintain a good standard of living, not that we create unnecessary work simply because our economic system does not know how to distribute income without it.
Another possibility is that access to AI itself eventually starts being treated more like basic infrastructure. Electricity is essential to participating in modern life, and internet access has moved in a similar direction as education, employment, banking and government services have become digital. Governments regulate these systems, subsidise access in some situations, and generally recognise that being completely excluded from essential infrastructure can also mean being excluded from economic participation.
AI may eventually reach that point if it becomes deeply integrated into education, healthcare, employment, business and public services. I am not suggesting that governments should give everyone unlimited access to the most expensive AI system or unlimited computing power, because that would obviously be unrealistic. I am suggesting that if a useful level of AI becomes necessary to participate meaningfully in the economy, then guaranteeing some basic level of access may eventually make more sense than treating all AI as an ordinary luxury software subscription.
The biggest problem may simply be timing. AI companies can improve their systems remarkably quickly, while governments and economic institutions usually move much more slowly. Even if capable AI eventually becomes extremely cheap and widely accessible, there could be several uncomfortable years during which jobs change faster than new systems for income, education, taxation and access can adapt. A technology becoming affordable eventually does not necessarily help someone who loses their income before that happens.
I also do not think we need to settle the argument about whether AGI has officially arrived before taking this problem seriously. There is a tendency to attach every discussion about AI and employment to a dramatic milestone where somebody declares that machines have reached human-level general intelligence, but the economy does not work according to philosophical definitions. Project Astra, for example, is Google's work on a universal AI assistant rather than an OpenAI GPT model, and Jensen Huang's comments about AGI have generally reflected optimism about how quickly AI could meet certain definitions rather than establishing some universally accepted moment when AGI officially arrived.
For the economic question, that distinction barely matters because AI does not need to become universally recognised as AGI before it affects someone's salary. It only needs to become capable enough that paying a person to perform a particular task stops making financial sense. That process can happen gradually, company by company and profession by profession, without there ever being a single morning when everyone agrees that AGI has arrived.
I am still optimistic about what AI can do for us because I do not think preserving unnecessary work should be the goal. If a machine can do something that a person spends eight hours a day doing only because they need the salary, I would rather find a way to let the machine do it and give that person more control over their time. Technology should make it easier to live, learn, build things, care for people, start businesses, explore ideas and change direction in life without every decision being determined by the need to remain employable.
The part we cannot ignore is that eliminating someone's work does not automatically eliminate their expenses. A person still needs somewhere to live, something to eat, healthcare, education, transportation and access to the tools that increasingly determine whether they can participate in the economy. If AI removes the need for some human labour while those basic needs continue to require money, then we have solved the technical problem of doing the work without solving the economic problem of how the person who used to do it lives.
I do not know whether the eventual answer will be dramatically cheaper AI, open models, new kinds of employment, an AI dividend, basic income, public access to computing resources, changes in taxation, or some combination of all of them. I suspect it will be a combination because the transition is too complicated for one policy or one technological trend to solve by itself. What matters to me is that we start treating access and distribution as part of the AI conversation rather than something we can figure out after the technology has already transformed the labour market.
I want AI to make work less necessary because I think having more control over our limited time is one of the best outcomes this technology could give us. I just do not want us to confuse being free from work with being free from the need for income. If AI replaces the work I used to earn money from, while I still need money to access the AI that has become necessary to earn money, then we have created an affordability loop that cannot work for everyone forever. Breaking that loop may eventually be just as important as building the technology that created it.
“I do not mind AI replacing the work I have to do if it gives me more time to do what I actually want with my life, but that freedom only exists as long as I can still afford to live.”