What We Know about AI’s Impact on the Economy

As artificial intelligence (AI) continues its dizzying growth, the big questions for economists are how it will impact productivity and how it will affect jobs. With regard to the first question, the technological advances of the past 15 years have had very little impact on productivity as measured by statistical agencies. According to Brynjolfsson and his fellow researchers at the Massachusetts Institute of Technology, there are a number of reasons for this, including the concentration of productivity gains in a handful of industries. It’s also because it takes a long time for organizations to introduce new technologies and restructure their operations around them. General purpose technologies like AI and the electric motor and microcomputer before it don’t become transformative overnight. Before they can have their full effect, substantial complementary investments have to be made and society has to adjust. 

According to studies by Brynjolfsson, Briggs and Kodnani (2023) and others, the lag between the time a disruptive technology emerges and when it has its full impact on productivity is likely to be measured in decades, not years. That’s because the technology has to be adopted and processes reconfigured. Still, Briggs and Kodnani estimate that AI could add 1.5 percentage points to annual US productivity growth. That would be in line with previous transformative technologies. 

Of course, this assumes productivity is being measured accurately, a hotly debated topic among economists when large amounts of intangible capital are involved. Currently with AI, tangible and intangible capital are working in synergy. According to a study by McKinsey, investments in semiconductors are expected to hit US$1 trillion by 2030. The resulting increase in computing capacity should generate new AI discoveries, innovations and applications. Some even predict that investment in AI will account for 1% of business investment in the United States by the end of the decade. 

Although statisticians are well versed in the challenges of measuring intangible capital, the incorporation of this type of capital remains incomplete, and its various components are typically introduced with delays of several years, as Statistics Canada economists have explained. The fundamental challenge is capturing intangible capital formation and measuring it, including second-round and diffusion effects. However, once we do measure intangible capital more accurately, we see that it tends to grow faster than tangible capital.

This results in the J-curve phenomenon—underestimating upstream capital accumulation, then over-attributing the eventual labour productivity gains to total factor productivity, a residual that Robert Solow famously referred to in 1957 as “a measure of our ignorance.” 

This is especially important to keep in mind as we try to determine whether the recent productivity gains in the US are just a statistical blip or an actual regime shift. Getting it wrong could have major monetary policy implications, as Alan Greenspan learned in the 1990s. We should at least recognize what Brynjolfsson et al. have shown—that the difference between perceived technological progress and the statistical data could be largely due to measurement errors. 

As for AI’s impact on jobs, an International Monetary Fund study released in January found that 60% of workers in advanced economies are in occupations with high levels of AI exposure. Unlike previous technologies that disrupted the job market, AI will likely impact more highly educated knowledge workers. But that doesn’t mean AI will replace them. AI could very well boost the productivity of workers like judges and surgeons, as society wouldn’t want technology to do their jobs. Workers such as telemarketers are more likely to be replaced, however. What’s more, AI and the changes it’s ushering in will create new jobs we can’t even imagine today—a widely documented phenomenon that has proven Luddite-like doomsday predictions wrong in previous waves of industrialization.

In short, while academic research has shed some valuable light on these issues, we still don’t know much. That makes it extremely difficult to incorporate artificial intelligence into a baseline forecast. How quickly will countries and economies adopt AI? Will society accept it, for example in the medical profession and other sensitive fields? How quickly will new jobs emerge, and what skills will they require? 

Meanwhile the technology continues to develop at breakneck speed. For example, recent advances in robotics suggest AI-powered humanoid robots could soon be market-ready, meaning technology can do even more physical jobs (e.g., landscaping) that research has deemed to have lower AI exposure for now. 

Finally, it’s hard to know what the regulatory landscape will look like. AI’s strength is a double-edged sword that brings both cybersecurity risks and increased inequality and other ethical and existential issues. Will regulations be excessive, lacking or just right? Will they be standardized across the globe? Will they be able to keep pace with change? When it comes to AI’s impact on the economy, one thing’s for sure: there’s still a lot we don’t know.