What will AI mean for economic inequality? [MIT Tech Review]
View Article on MIT Tech Review
Prominent AI researchers expect the arrival of artificial general intelligence anywhere between “the next couple of years” and “possibly never.” At the same time, leading economists disagree about the potential impact of AI: Some anticipate a future of perpetually accelerating productivity, while others project more modest gains. But most experts agree that technological advancement, however buoyant, is no guarantee that everyone benefits.
And unfortunately, even though some of the most notable AI R&D efforts declare that making sure everyone benefits is a key goal or guiding principle, ensuring that AI helps create a more inclusive future remains one of the least invested-in areas of AI governance. This might seem natural given the state of the field: The impact AI will have on labor and inequality is still highly uncertain, making it difficult to design interventions. But we know at least some of the factors that will influence the interplay between AI and inequality over the next few decades. Paying attention to those can help us make the idea that AI will benefit everyone into more than just a pipe dream.
Because they’re largely driven by the private sector, AI development and use are heavily influenced by the incentive structures of the world’s economies. And if there is something important that can be predicted with reasonable certainty about those economies, it is their future demographic composition. There is a stark divide between higher-income countries, whose populations are aging rapidly and will shrink without migration, and low- and lower-middle income countries, which will continue to grow for the rest of the century thanks to the excess of births over deaths.
What does this have to do with AI? AI development is concentrated in the aging countries, and thus it will follow the path set by the realities, needs, and incentives in those places. Aging countries are seeing the ratio of working-age people to retirees collapse, making it more difficult to sustain pension schemes and contain health-care costs. Countries looking to maintain their retirees’ living standards and their overall economic dynamism will seek ways to expand their effective labor force, be that with humans or with artificial agents. Limited (and likely highly unpopular) gains could come from increasing the retirement age. More sizable gains could come from immigration. But keeping the ratio of the working-age to retiree populations constant would require a significant increase in immigration to the higher-income countries. Widespread anti-immigration sentiment makes that seem unlikely, though opinions could change relatively quickly when people are faced with the prospect of diminishing pensions and rising health-care costs.
If overly restrictive immigration policies do not relax in rich countries, we will likely see the economic incentives to fill labor gaps with AI go into overdrive over the next few decades. It might seem on the surface that this won’t exacerbate inequality if there are fewer people than available jobs. But if the trend is associated with an uneven distribution of gains and losses, increasingly precarious employment, excessive surveillance of workers, and digitization of their know-how without adequate compensation, we should expect a spike in inequality.
And even if the efforts to replace labor with AI unfold incredibly well for the populations of rich countries, they might dramatically deepen inequality between countries. For the rest of the 21st century, lower-income countries will continue to have young, growing populations in need not of labor-replacing tech, but of gainful employment. The problem is that machines invented to fill in for missing workers in countries with labor shortages often quickly spread even to countries where unemployment is in the double digits and the majority of the working population is employed by unregistered informal businesses. That is how we find self-service kiosks in South African restaurants and Indian airports, replacing formal-sector jobs in these and many more countries struggling to create enough of them.
In such a world, many beneficial applications of AI could remain relatively underdeveloped compared with the merely labor-saving ones. For example, efforts to develop AI for climate-change resilience, early prediction of natural disasters, or affordable personalized tutoring might end up taking a back seat to projects geared to cutting labor costs in retail, hospitality, and transportation. Deliberate, large-scale efforts by governments, development banks, and philanthropies will be needed to make sure AI is used to help address the needs of poorer countries, not only richer ones. The budgets for such efforts are currently quite small, leaving AI on its default path—which is far from inclusive.
But default is not destiny. We could choose to channel more public R&D efforts toward pressing global challenges like accelerating the green transition and improving educational outcomes. We could invest more in creating and supporting AI development hubs in lower-income countries. Policy choices that allow for greater labor mobility would help create a more balanced distribution of the working-age population between countries and relieve the economic pressures that would drive commercial AI to displace jobs. If we do none of that, distorted incentives will continue to shape this powerful technology, leading to profound negative consequences not only for lower-income countries but for everyone.
Katya Klinova is the head of data and AI at UN Global Pulse, the secretary-general’s innovation lab. The views represented here are her own.