How will artificial intelligence reshape the global economy? Some economists predict only a small boost — around a 0.9% increase in gross domestic product over the next ten years1. Others foresee a revolution that might add between US$17 trillion and $26 trillion to annual global economic output and automate up to half of today’s jobs by 20452. But even before the full impacts materialize, beliefs about our AI future affect the economy today — steering young people’s career choices, guiding government policy and driving vast investment flows into semiconductors and other components of data centres.
Why evaluating the impact of AI needs to start now
Given the high stakes, many researchers and policymakers are increasingly attempting to precisely quantify the causal impact of AI through natural experiments and randomized controlled trials. In such studies, one group gains access to an AI tool while another continues under normal conditions; other factors are held fixed. Researchers can then analyse outcomes such as productivity, satisfaction and learning.
Yet, when applied to AI, this type of evidence faces two challenges. First, by the time they are published, causal estimates of AI’s effects can be outdated. For instance, one study found that call-centre workers handled queries 15% faster when using 2020 AI tools3. Another showed that software developers with access to coding assistants in 2022–23 completed 26% more tasks than did those without such tools4. But AI capabilities are advancing at an astounding pace. For example, since ChatGPT’s release in 2022, AI tools can now correctly handle three times as many simulated customer-support chats on their own as they could before5. The better, cheaper AI of tomorrow will produce different economic effects.
Second, carefully controlled studies do not capture the wider ripple effects that accompany AI adoption. For example, the studies involving call-centre workers3 and software developers4 found that when organizational structure remained fixed, the less-experienced workers benefited most from AI assistance. But in the real world, managers might respond by reorganizing work or even replacing some of the less-experienced workers with AI systems. If they do, the effect on those individuals could be the opposite of that estimated in controlled studies. Indeed, payroll data suggest that employment of younger workers has declined since 2022, particularly in occupations that include tasks that AI excels at, such as customer service and software development6. However, researchers are still trying to understand how much of the pattern is attributable to AI technology.
Carefully controlled studies are like flashing a bright, narrow spotlight: they are only part of the illumination needed to understand how society is adapting to AI. With so much still unknown about its broader economic and social effects, popular debate often slips into speculative, science-fiction narratives of a world dominated by machine intelligence.
Social science could help to navigate these uncertainties, but it would require both imagination and grounding. Here, I describe three complementary approaches that can guide researchers working in this rapidly evolving field.
Social science fiction
One approach is to create what economist Jean Tirole calls social science fiction7 — speculation about the future that remains rooted in fundamental economic principles and behavioural theories. Rather than relying on imagination alone, this kind of analysis uses models to explore how technologies might interact with market forces.
For example, in 2019, researchers modelled how self-driving cars might reshape cities and found that the vehicles could make traffic worse8. Because passengers in self-driving cars can relax, read or watch videos, the personal cost of time spent in traffic falls. But as more people choose to travel by car, they impose greater congestion on others. Whether that leads to inefficiency will depend on whether governments implement policies such as congestion pricing to correct the ‘externality’.
... continue reading