AI-driven growth: Navigating the path to new markets

Projections of the economic impact that can reasonably be expected from AI range widely. Nobel laureate Daron Acemoglu estimates a cumulative GDP uplift in the US of +1.1% on a 10-year timeframe, whereas Goldman Sachs quantifies a 15% boost from full adoption of the technology.

The chasm between these two estimates adds up to about $4 trillion at current US GDP levels. Acemoglu’s estimate focuses on only those tasks that are cost-effective to automate today; Goldman anticipates more widespread adoption over time and incorporates gains from the reallocation of workers and the creation of new tasks in the economy. The dynamic reallocation of people into new roles can be a powerful growth driver. MIT economist David Autor and colleagues found that 85% of employment growth over the last 80 years is explained by the technology-driven creation of new positions. But is it reasonable to bank on the emergence of these new tasks going forward?

The waves of digitalization so far provide one reason for optimism: tech often creates new markets. I should know: the company I work for, Google, derives almost the entirety of its revenues from markets that did not exist 30 years ago (search, video streaming and cloud). These services now support hundreds of billions of dollars in economic value annually for US businesses, nonprofits, publishers, creators and developers. This is by no means an isolated example. While flea markets have existed for centuries, eBay’s digitalization of the concept enabled the matching of demand and supply on a completely different level. The same is true for Amazon’s 'everything store's' contributions to retail in general. Uber revolutionized transportation and delivery. The list goes on.

How did this come about? The world is full of 'latent markets': situations where there is demand for a product or service but no supply or the converse. These markets fail to form for several reasons, including contracting difficulties. Pre-Uber there was no practical way to write an instantaneous contract for a stranger to come pick you up in the middle of nowhere and take you to your destination. The ability to write this new contract on the fly is an example of what Google’s first Chief Economist, and my mentor, Hal Varian called “computer-mediated transactions.” Uber implemented this through 'combinatorial innovation' relying on mobile technology, payments, digital maps and a reviews platform - technological components that existed separately, but together gave rise to a new market.

AI is well-positioned to reveal more latent markets. Economists Ajay Agrawal, Joshua Gans and Avi Goldfarb characterize recent developments in AI as advances in the statistics of prediction. Prediction is ubiquitous across the economy: most business decisions can be thought of as a combination of prediction and judgment. Think of quantitative areas, such as inventory management, logistics optimization, predictive maintenance or finance - where AI has been making rapid inroads. Surveys indicate that around 70% of financial services firms now rely on AI for activities such as cash flow predictions, credit scoring and fraud detection. Importantly, the promise of AI is not limited to advanced economies. Lenders in dozens of developing countries are offering small loans to millions of people based on Columbia economist Daniel Björkegren’s idea of using machine learning to estimate the likelihood of repayment from cellphone network data.

But these kinds of overtly quantitative applications are only part of the story. Recent AI breakthroughs have revolutionized how computers engage with language, a core component of most products and services. Chatbots powered by large language models already boast hundreds of millions of monthly users. Personalized AI tutors, such as Khanmigo, promise to revolutionize learning. At work, AI powers a host of new use cases from the generation of first drafts to meeting summarization. While these point solutions already deliver value, they foreshadow vastly larger gains from a systemic rethink of work processes, much like replacing steam engines with electrical power was only a preamble to the broader reorganization of production. The way people engage with information is also changing: Google recently shared that AI overviews in 'Search' are resulting in people “asking longer and more complex questions” than before.

These developments offer tantalizing suggestions of new ways for supply to meet demand, following the familiar pattern of combinatorial innovation from past digitalization. This helps build some intuition supporting the immense potential of AI quantitatively captured in the projections of Goldman Sachs. At the same time, Acemoglu’s more conservative outlook helps underscore that the promise of AI will not be realized automatically. Sound policy will make the difference between success and a missed or delayed opportunity. As neatly expressed by James Manyika and Michael Spence, now is the time for a positive vision of what AI can do - and effective measures to turn that vision into reality.