A recent discussion by Anthropic about artificial intelligence and economic growth caught my attention. In one of its extreme scenarios, US GDP could reach US$44.4 trillion by 2030, measured at 2025 prices. At the same time, labour's share of GDP could fall from about 60% to 45.2%. These figures come from a model built on specific assumptions. They are not a firm prediction of the future, and the model leaves out important factors, including policy responses.
Source: https://www.anthropic.com/institute/econ-scenarios
The discussion nevertheless raises a question that deserves serious attention. If AI greatly increases society's productive capacity while reducing the amount of labour people contribute directly to production, how will the additional wealth reach ordinary people? If wages remain most people's main source of income, while jobs or workers' bargaining power decline, a substantial gap could open up between GDP growth and improvements in everyday life. Following this question leads to changes in the relationship between labour and capital over the past two centuries, the boundaries of ownership and entitlement to returns, and the ways people might develop themselves and judge others when earning a living becomes less dependent on paid work. These questions are connected.
In his discussion of machinery in the Grundrisse, Marx had already observed that, as large-scale industry developed, the creation of real wealth would depend increasingly on accumulated science, technology and social knowledge, rather than simply on the amount of direct labour time invested. He went on to envisage how a reduction in socially necessary labour time could release time and material resources for individuals to pursue artistic, scientific and other activities. This combined an observation about production with a reflection on the possibilities for human development.
Source: https://www.marxists.org/archive/marx/works/1857/grundrisse/ch14.htm
Looking back over the history of industrialisation, average working hours have indeed fallen over the long term in many countries. Historical estimates cited in an OECD study suggest that annual working hours in advanced industrial countries declined from around 3,000 in 1870 to between 1,500 and 2,000 by 1990. Countries did not follow the same path, however, and the decline did not continue uninterrupted in every period.
Source: https://www.oecd.org/content/dam/oecd/en/publications/reports/2001/03/trends-in-working-hours-in-oecd-countries_g17a14e3/674061356827.pdf
Several things need to be distinguished here. The labour time needed to produce an item, the hours a person actually works in a year, and the income workers ultimately receive are different measures. Suppose producing something once took eight hours and now takes only one. A company could expand output, reduce its workforce, or shorten employees' hours while maintaining their income. Greater productivity makes these possibilities available. Which one is adopted also depends on market demand and the institutions in place.
We therefore cannot describe the past two centuries simply as an uninterrupted fall in labour's income share and an uninterrupted rise in capital's share. In Australia, for example, labour's share rose during the 1960s and early 1970s before entering a long-term decline. Research published by the International Labour Organization and the OECD in 2025 found that the global labour income share fell by 1.6 percentage points between 2004 and 2024. The decline in recent decades matters, but it is not a law that has moved in the same direction without reversal for two hundred years.
Source: https://www.rba.gov.au/publications/bulletin/2019/mar/the-labour-and-capital-shares-of-income-in-australia.html
Improvements in ordinary people's incomes cannot be attributed solely to a minority obtaining highly paid positions through promotion and specialisation. Wage-setting arrangements and collective bargaining can distribute productivity gains so that people receive better pay while remaining in the same jobs. Technology, globalisation and changes in workers' bargaining power jointly influence labour's share of income.
Source: https://www.ilo.org/publications/policy-measures-address-inequalities-and-increase-labour-income-share
Still, the scarcity of professional expertise is an important part of understanding income differences. Someone who possesses knowledge that others find difficult to acquire, and can undertake work that others find difficult to perform, may command higher pay. Seen in this light, AI could disrupt precisely the path by which education, professional training and career advancement have traditionally conferred an income advantage.
If AI can perform a professional task at low cost when it previously required years of training, the expertise involved may become less scarce. Degrees and professional qualifications will still matter, but their relationship with high income may change. Someone who once possessed knowledge unavailable to others may find that many people can now access that knowledge through AI. This does not mean that all professional labour will lose its value. AI may also enable ordinary people to undertake tasks once accessible only to a small number of experts, extending what people can do. The outcome will depend on whether AI mainly helps people with their work or replaces work they previously did, and whether new demand creates enough new tasks. What concerns me is a possibility worth considering in advance: companies may increasingly expand output by deploying models and adding computing capacity, without increasing their workforce in proportion.
Under these conditions, more of the additional wealth could flow to the owners of the relevant assets. Here, too, total capital income, capital's share of income and the rate of return on each dollar invested must not be confused. An increase in total capital income does not mean every investment becomes more profitable. A fall in labour's income share does not mean every worker's real income falls. What needs examining is how the channels through which different groups receive income, and their respective shares, change as productive capacity expands.
This brings us to the institutions governing distribution. No longer needing to work long hours to support oneself is a very different situation from losing one's job without another source of income. Both people may work fewer hours, but one gains greater freedom to organise life while the other may lose the means to live securely. Any discussion of being 'freed from labour' must make this distinction clear.
I think this discussion requires us to reconsider an assumption often taken for granted: to what extent should the returns generated by capital belong to whoever owns it?
Private investment needs to earn a return, and people who bear risks, organise production and develop technology should receive appropriate rewards. Acknowledging this does not mean that private entitlements to returns should be unlimited. Share ownership can determine how a company's distributable profits are divided among investors. Ownership percentages alone cannot settle whether society as a whole has a legitimate claim to some of the returns from that productive activity. The rules governing distribution within a company and the justification of distribution across society are questions at different levels.
Consider two people whose annual personal incomes are ten million yuan and one hundred billion yuan respectively. Must we assume that both should ultimately retain the same proportion of their incomes? I do not think that follows necessarily. This is a question about institutional design, not a claim that existing law automatically changes the nature of income once it reaches a particular amount. An amount does not suddenly turn one kind of income into another, but its scale can prompt us to reconsider the conditions that made it possible. When returns reach an enormous scale, explaining the whole outcome solely through one person's labour, talent or investment leaves a question unanswered: how much depends on conditions collectively provided by society but never individually priced?
If a return depends on accumulated knowledge, infrastructure, public education and institutional safeguards, saying 'my capital produced it' is not a complete moral account. This does not mean that higher income is inherently less legitimate. It means there can be reasonable grounds for requiring a greater social share of exceptionally high incomes. Such a requirement cannot simply be dismissed as an unfair infringement of private property.
Of course, this claim should take the form of clear, generally applicable and enforceable rules established in advance. It should not involve deciding what to do with someone's income after noticing how much they have earned. Specific tax settings still need to account for costs, risks, incentives for innovation and practical effects. These difficulties of design do not remove the need to reconsider the boundaries of entitlement to returns.
AI makes this issue more prominent. Anthropic's published disclosures about training data, for example, state that its models use publicly available information from the internet, public and private datasets, and data from other sources. Building a model also involves training, filtering and subsequent adjustments. It cannot be reduced to collecting material from the internet.
Source: https://www.anthropic.com/transparency
Equally, we cannot focus only on the money and computing power invested at the final stage while overlooking where the knowledge came from. Language, science, art, computer code and experience across countless occupations embody the accumulated contributions of many generations. In its discussion of fiscal policy for AI, the IMF also points out that recent technological breakthroughs rest on years of investment in fundamental research, including publicly funded research.
Source: https://www.imf.org/en/blogs/articles/2024/06/17/fiscal-policy-can-help-broaden-the-gains-of-ai-to-humanity
In my view, this provides an important reason for sharing AI's returns across society. Private companies deserve appropriate returns for turning dispersed knowledge into usable technology. The entry of collectively accumulated knowledge into that productive process should also be recognised in the institutions governing distribution. Social contributions should not be treated as nonexistent simply because they are difficult to register individually or convert into shareholdings. This does not establish that every AI achievement already belongs legally to all humanity, or that private capital should receive no return. Those are different questions. My argument is that mechanisms should allow society as a whole, and ultimately humanity, to share AI's returns. Private ownership should not be the final reason for excluding such claims.
How might such arrangements work? One relatively direct route would be to bring some of the additional capital income into public revenue through taxation, using it for income security, education, health care and other public services. IMF research on generative AI has already proposed strengthening taxes on capital income and improving social protection to address potentially greater wealth concentration. It also cautions that a simple tax specifically on AI technology could discourage beneficial investment and innovation.
Source: https://www.imf.org/en/blogs/articles/2024/06/17/fiscal-policy-can-help-broaden-the-gains-of-ai-to-humanity
Another possibility would be for the public to hold assets through vehicles such as public investment funds, giving people a share in the returns from growth. Ordinary people's connection to expanding production would then extend beyond selling their labour: collectively held assets could also provide a social dividend. Such arrangements would require transparent governance, clear rules about who benefits and effective risk management. Public investment cannot be treated as a tool that produces profits without costs or risks.
Society's gains need not all take the form of additional money in individual bank accounts. People also share in greater productive capacity when education, health care and aged care become more accessible and the cost of meeting basic needs falls. Yet saying 'goods will become cheaper' cannot replace a discussion of distribution. If someone loses their main source of income, lower prices for some goods may still be insufficient to support them.
The common direction of these mechanisms is to make people's entitlement to share in the fruits of production less dependent on how much labour they can currently sell in the market. Social protection should not be understood only as the successful looking after the unsuccessful. If productive capacity already incorporates widespread social contributions, sharing some of its returns can be understood as an entitlement that needs institutional expression.
Extending this reasoning to humanity as a whole also requires us to confront differences between countries. Knowledge has not accumulated within the boundaries of a single nation, while taxation and public services are largely organised by national governments. Enabling societies with little capital, computing capacity or ownership of major AI companies to benefit would require further international arrangements. 'Benefits for all humanity' cannot simply mean redistribution within a few technologically leading countries.
Only as these distributive conditions develop might reduced labour requirements begin to change how people live. Here, too, being freed from labour needs a precise meaning. People will still care for their families, investigate questions, create works and participate in their communities. All these activities require effort. What may diminish is dependence on paid work to make a living, rather than the need to undertake anything demanding.
What interests me more is whether the standards by which people judge one another might change under these conditions. Today, when assessing someone's social position, we often ask which university they attended, what they do for a living and what position they hold. Much of the importance attached to this information comes from its connection to income, security and future opportunities. If AI reduces the market scarcity of some professional abilities, while society can provide reasonably dependable living conditions through other channels, degrees and occupations may become less important as signals of income advantage. Knowledge itself need not lose value. Learning could become more closely connected with understanding the world, exercising judgement and developing oneself, without always revolving around competition for employment.
Under these conditions, philosophical ability, artistic ability, critical thinking, emotional understanding and personal cultivation might have more room in our assessments of one another. Whether someone can think a problem through, make a serious effort to understand others, develop their own aesthetic sensibility and sustain meaningful interests could have a more direct bearing on their relationships.
This change does not require us to assume that AI will never surpass people in these abilities. Someone may learn the piano because they want to understand, express and perform music themselves. Someone may study philosophy because they want a clearer understanding of their life and the world. Even when these activities do not produce a higher income, their significance for personal development need not diminish.
Following this line of thought, we can also make a conjecture about partner choice. If occupation and income no longer determine material security as directly as before, people might place greater weight on the quality of a relationship, emotional capacities, intellectual exchange, artistic interests and attitudes to life when choosing a partner. Attraction, choice and comparison would remain, but the qualities people value, and the weight they give them, could change.
This possibility is compatible with an evolutionary perspective, but it is not an inevitable conclusion that follows from evolutionary theory alone. Changes in partner preferences are, in the first instance, changes in culture, institutions and living conditions. Whether they would also produce long-term biological changes requires separate evidence. We cannot equate the popularity of a particular profession, or a future preference for a certain kind of personal cultivation, with biological 'improvement' in human beings.
Nor should we forget the distributive conditions envisaged earlier. If earnings from work fall while capital ownership becomes more important, and most people have no way to share capital returns, partner choice and social judgements could place even greater weight on family assets and inherited wealth. Philosophy, art and personal cultivation will not automatically replace wealth as the main point of reference simply because AI exists. Greater scope for human development still depends on ordinary people having the practical conditions that make such development possible.
The institutional adjustment I envisage would therefore preserve returns on private investment and allow differences in income, while revising an understanding that treats entitlement to capital returns as absolute. The more production depends on conditions collectively accumulated over time, the stronger the case for incorporating social sharing mechanisms, so that people who have not directly invested capital can also receive a reliable share of the gains from increased productive capacity.
Ultimately, this can mean something very concrete in everyday life. An ordinary person could reduce the time spent earning a living without fearing the loss of basic security, and could continue learning, care for family members and develop interests without first having to prove how much income those activities might eventually generate. Only then would greater productive capacity provide the conditions for more people's all-round development.
Discover more from Geoffrey Chen
Subscribe to get the latest posts sent to your email.