Chapter 1
Mind the gap: GenAI and inequality
How will GenAI risks related to distribution of income play out?
As GenAI continues to make large strides in adoption and diffusion, one of the most pressing concerns is that its benefits may not be shared equitably. GenAI could exacerbate two types of inequalities: 1) inequality in the distribution of national income between corporate profits and labor, and 2) inequality in the distribution of income across workers and households.
A smaller slice of the GenAI pie
As GenAI technologies spur greater productivity gains, most of the increase in business output will be split between workers in the form of wages (labor share) and businesses in the form of profits (profit share). In advanced economies, this split has become increasingly unequal in recent decades with the labor share declining markedly since the late 1990s. In the US, it reached a record low of 52% of GDP in 2022 as the profit share reached record high levels of around 13%. This well-documented secular trend has been attributed to a combination of factors, including technological innovation and automation – with the International Monetary Fund (IMF) estimating that half of the decline in the labor share in advanced economies can be explained by technological change – and, more recently, industry concentration and market power3.
Like prior technological innovation, GenAI will likely exert downward pressure on the labor share. As organizations adopt and absorb the new technologies, they will most likely increasingly substitute capital for labor. This would likely lead to lower labor demand and wage growth amid reduced workers' bargaining power. Elevated market concentration, as the GenAI market becomes increasingly dominated by a small number of large businesses, will also tend to generate higher markups and result in a growing fraction of productivity gains going to corporations.
US employee compensation and before-tax corporate profits as a share of GDP
GenAI could reinforce income inequality
Rapid technological change, led by the digital revolution, has also been associated with a significant increase in income inequality across most major advanced economies over the past four decades. This widening wealth and income gap has been particularly salient in the US. Since 1980, the average income for US households in the bottom income quintile has quadrupled while the average income for those in the top quintile has increased sixfold. GenAI has the potential to perpetuate and potentially exacerbate these longstanding inequality patterns if not implemented in an inclusive way.
As we have highlighted, the GenAI-induced capital investment and productivity gains are set to provide a substantial lift to the global economy worth between $1.7t and $3.4t4. For the US economy, the real GDP boost could range between $900b and $1.7t5 over the next decade. With personal income accounting for 75% of US GDP, the lift to aggregate household income could range between $675b and $1.3t6.
If distributed equally among households, it would represent an average gain of $5,135 per household. However, the boost to income from GenAI will likely be uneven. Indeed, US households in the lowest income quintile received only about 3% percent of all income in 2022, compared with 52% for the highest income quintile. Assuming this trend holds, GenAI could increase the income gap between the top and bottom earners by nearly $33b annually in the next 10 years7.
US distribution of household income gain from GenAI by income quintile, $bn
The risk of wage polarization
The extent and nature of the labor market disruptions from GenAI will be key for the trajectory of inequality. Globally, we find that 59% of occupations have a high to moderate exposure, with 67% in advanced economies and 57% in emerging markets8. We showed that that 66% of jobs will be moderately to highly impacted by GenAI. The remaining 34% of occupations have low AI exposure but will still be affected by GenAI via some tasks9. Unlike prior waves of technological innovation, which had the strongest impact on low- and middle-wage workers in routine jobs, higher-wage workers are the most exposed to GenAI augmentation.
Research from the IMF10 shows that when GenAI is highly complementary to labor, the complementarity effect more than offset the displacement effect, particularly at the top of the income distribution, which leads to a smaller share of high-income workers facing potential job losses. With our GenAI augmentation scores generally stronger for occupations generating higher wages, earners at the top of the wage scale seeing stronger productivity gains could see a disproportionate increase in their labor income compared with low-wage earners who are less exposed to GenAI, which would lead to a rise in income inequality.
Chapter 2
Superstars and winner-takes-all effect
GenAI may further widen the digital gap by pushing smaller businesses out.
Since the 1980s, technological innovation has led to a marked increase in market concentration where a small number of large organizations capture a larger share of the profits and value added.
This has been especially apparent in the digital economy with the US high-tech digital sector dominated by a few big players, leaving little room for innovators to break in. GenAI has the potential to deepen the current divide between technological leaders and laggards.
Research from the Organisation for Economic Co-operation and Development (OECD)11 has shown that technology diffusion is a highly uneven process as the productivity gap between the most productive businesses – global frontier organizations – and the rest increased significantly during the 2000s. Between 2001 and 2009, labor productivity in OECD countries grew 35% among “frontier firms,” compared with only 5% for other businesses. This widening productivity gap has also been documented in more recent research12 showing that the gap between leading and laggard organizations has increased over time, with the greatest increase in the IT sector.
The risk is that the benefits from GenAI – including higher productivity levels and stronger profitability – could accrue to a handful of “superstar” businesses that have the resources to successfully deploy GenAI solutions and applications, and develop these capabilities via access to the vital building blocks of GenAI, including:
- Large and robust datasets: The data-intensive nature of GenAI means that companies will need to make significant investment in gathering, storing and processing data. This will likely hamper the ability of new players to enter the market and reinforce the dominance of large incumbent technology businesses that have access to the largest datasets.
- Computational power: GenAI systems typically require significant computational resources to run and train sophisticated AI models, including deep learning and natural language processing models. This requires a sizable investment in physical and digital infrastructure that only the largest organizations can afford. For example, tech analysts estimate the GPT-4 model cost more than $100m to train during its initial development and requires about $700k a day to run.
- Skilled talent: Developing a GenAI model also requires a workforce with a particular and currently relatively scarce set of skills, leading frontrunner businesses to impose and maintain non-compete clauses that prevent the free movement of talent across organizations and industries.
Chapter 3
The global GenAI divide
Lack of GenAI readiness could put another wedge between the most and least advanced economies.
The deployment of GenAI is poised to provide a significant boost to the global economy. However, the benefits are likely to be unevenly distributed among economic regions and could widen global economic disparities as some economies are better equipped for AI adoption than others.
Oxford Insights’ Government AI Readiness Index, which assesses the capacity of governments to exploit the innovative potential of AI technologies, sheds light on these global gaps in AI preparedness. The index, which ranks 193 countries based on 39 indicators across three pillars (government, technology sector, and data and infrastructure) shows that there is a significant gap in AI readiness across countries and regions around the world, with low-income and developing economies with low GDP per capita levels generally lagging developed economies.
The Stanford Global AI Vibrancy Tool, which assesses countries’ level of AI advancement across various metrics related to research and development and the economy, paints a similar picture with most advanced economies better positioned to reap the AI benefits. However, China and India stand out as emerging economies ranking in the top three economies for their AI vibrancy with China leading in R&D and India in AI talent.
AI Government Readiness Index and GDP per capita by country
Overall, we find that three groups of countries stand out:
- Pioneers: The US and China are leading the AI race and will likely experience the greatest economic gains from AI. Both countries combine the strongest research activity in AI and a high number of innovations. Indeed, most AI-related patent filings are made at the patent offices in the US and China, and the two countries account for most of AI-related scientific publications. However, the US continues to lead in the production of machine learning systems, with over half of the world’s large language and multimodal models produced by American institutions in 202213.
- Front-runners: This group is best positioned to adopt and deploy AI technologies and includes a range of economies and regions such as Singapore, Japan, South Korea, Canada, Europe, Australia and India. These are typically advanced economies with strong governance, and the infrastructure and human capital required to enable these technologies. India's abundant supply of qualified and highly skilled tech talent makes it well placed to harness the potential of AI technologies.
- Laggards: This group includes the regions that are least prepared to reap the benefits from AI such as sub-Saharan Africa, some Latin American countries, and Central and South Asian countries. These countries will likely miss out on the initial wave of AI-driven productivity gains given their lack of essential infrastructure and skilled workforce to enable AI technologies and develop a broad AI ecosystem.
Such disparities in AI readiness and advancement could worsen inequalities between the most and least advanced economies. The risk is that the AI revolution could amplify the existing global digital divide – which refers to the disparity in access and use of information technologies such as the internet. Global cooperation aimed at reducing technological disparities by expanding access to AI technology and infrastructure and building digital skills will be critical to help bridge the global AI divide.
Figures in the content include those obtained from the Bureau of Economic Analysis (BEA), Bureau of Labor Statistics (BLS) and the Census Bureau.
Summary
This is the fifth installment of the EY-Parthenon macroeconomic article series on the economic impact of AI. The series aims to provide insights on the economic potential of generative AI (GenAI), including new developments and actionable insights to arm companies’ decision-makers. The fifth article in this series covers GenAI risks and challenges for the economy.
