Whether AI can crack the Solow paradox is the era’s big question

It is now almost a year since large language models such as ChatGPT and Bard captured public attention. The initial excitement has now been tempered with some scepticism about these AI tools, but there is hope that they will keep improving with newer versions. In fact, as Jaspreet Bindra wrote in these pages last week: “The problem is not the technology, but our expectations of it… The moment we temper our expectations, we may see it for what it is: a near-miraculous technology that unleashes the power of language and creativity."

The possible impact of AI on economic growth should also be seen in the context of recent concerns among some economists. Many of them fear the combination of declining populations and greater difficulty in finding new ideas that can increase productivity across the economy in effect means that the world will gradually lose economic momentum, and perhaps even approach the stationary state that the very first economists such as Adam Smith, David Ricardo and John Stuart Mill had predicted.

The first wave of research on whether the use of AI tools increases labour productivity has generally been focused on specific tasks rather than a wider arc of human activities, and understandably so, given the relatively recent use of AI tools. There are now some early indications that the use of AI tools can increase productivity in many standard processes. What is less clear is whether AI can write the next great novel or come up with the next great scientific breakthrough on its own.

However, what really matters for an economy is whether the introduction of a new technology improves human productivity in a ubiquitous manner across a wide variety of human activity. What made innovations such as the wheel, steam engine, electricity, semiconductor and the internet so important to economic growth in their respective eras is that they became what are called general purpose technologies. These become pervasive, could be improved over the years, and could also be used in several downstream sectors. Generative AI has the attributes of a general-purpose technology and holds the potential to make a deep impact in most corners of the economy.

The fact that AI can lead to another wave of higher human productivity will also present policy challenges, especially in countries such as India that have a growing labour force. There are three possibilities here in terms of employment, productivity and real wages.

The first is that productivity increases with the same combinations of labour and capital used in the production of a good or service. Economists call this Hicks-neutral technical change, named after the 20th century English economist John Hicks, who provided the broad framework in a 1932 book on the theory of wages.

This sort of technical change is not the only one that is possible. Technical change can either save labour or save capital, with important implications on the payments to these two factors of production, or wages and profits. A simpler way to pose the question is by asking whether AI is more likely to increase the productivity of the labour force or the productivity of capital assets. A related problem is whether AI will help unskilled or skilled workers more. Some such as Noah Smith believe that the spread of AI tools will allow unskilled workers to compete in tasks that were once the preserve of skilled workers. For example, one finding in a recent study on the use of AI by consulting firm Boston Consulting Group was that lower performers saw the biggest gains in efficiency across 18 business tasks that the subjects of the study were given.

Salman Khan of Khan Academy is an AI optimist. In a recent TED talk, he said that AI can revolutionize education by giving every student access to a personalized tutor as well as giving every classroom teacher access to a teaching assistant. This can have profound implications in a country such as India where poor students are mostly condemned to substandard schools, regardless of whether these are run by the government or the private sector.

The upshot is that even if—and it is still an if—AI increases productivity in the macroeconomy, it may have important distributional effects that policymakers will have to grapple with in the coming years. These distributional effects will depend on the nature of technical change unleashed by AI, and whether it will help close some of the skill gaps that have been an important driver of higher inequality in many countries.

These are just straws in the wind right now. There is little evidence right now that AI is actually powering productivity changes at a macro scale. The Solow Paradox lives on in the age of AI, though that could change over the next decade if the optimistic case plays out.