The promise and peril of generative AI

CAMBRIDGE — Ever since OpenAI released its ChatGPT chatbot last year, a growing number of analysts have been predicting that generative artificial intelligence will displace millions of workers and cause widespread economic upheaval. But how exactly will generative AI affect the global economy?

Recent estimates provide an indication of the looming labour-market disruption. Goldman Sachs economists, for example, anticipate that as many as 300 million full-time jobs could be automated as a result of the latest AI breakthroughs and that two-thirds of workers in Europe and the United States could be exposed to AI-based automation. A working paper by researchers at OpenAI finds that roughly 80 per cent of the US workforce could see at least some of their tasks automated by the introduction of large language models (LLMs) such as ChatGPT. And some law firms and marketers have already begun to use generative AI tools.

But it is still unclear whether the new AIs will improve existing employees’ productivity by taking routine tasks off their hands, or simply make workers technologically redundant. To be sure, many white-collar workers would be delighted if AI tools could take on dull tasks like keeping minutes during meetings, answering routine queries or filing expense claims. But many believe, as Daron Acemoglu and Simon Johnson recently argued, that the current generative AI arms race is geared toward reducing costs by replacing workers with algorithms, rather than harnessing the power of these technologies to augment human labour.

Another possibility though is that most companies will be slow to adopt this powerful technology because of a lack of skills and know-how. This is not necessarily reassuring either. While new technologies often disrupt livelihoods and industries, they could also bring about the productivity growth needed to boost incomes and living standards. After almost two decades of extremely slow productivity growth in most advanced economies, generative AI has emerged right when we need it. But to ensure that it delivers widely shared benefits, we must heed the lessons of the previous wave of digital innovation.

Over the past 20 years, innovations like the smartphone and communication technologies like 4G and 5G wireless networks have transformed everyday life, leading to the creation of new sectors and business models. As of 2021, the average American spent roughly eight hours a day online, more than double the 2011 figure. The cloud computing and the e-commerce industries have grown rapidly, reflecting a labour market in which digital skills are increasingly a prerequisite for landing a high-paying job. Yet despite these technological advances, productivity growth has been dismal since the mid-2000s.

What explains this economic puzzle? While it is possible that digital technologies are simply not very productive, their widespread adoption suggests otherwise. A more plausible explanation is that it takes time to figure out how best to use new technologies. As a result, only a small minority of companies in the US and the United Kingdom have been able to use digital tools to boost their productivity and pull ahead.

In his 2022 book The New Goliaths, Boston University’s James Bessen explores why companies are having trouble adapting to digital technologies. The complexity of advanced software, he argues, confers an advantage on the largest and most technologically sophisticated companies, because only they have the resources and know-how needed to adopt such tools and benefit from them.

Given the massive (and costly) computing power needed to use and maintain generative AI tools, it seems inevitable that this new technology will follow a similar path. If a handful of dominant companies use deep-learning algorithms like OpenAI’s GPT-4 to build new services and products, they could enhance their market power and erect insurmountable barriers to entry.

But the true potential of these new technologies goes beyond their ability to enable a few companies to become more efficient or develop new products. To provide widespread productivity gains and create real value, generative AI models must change the way we produce things. After all, the most sustained productivity booms of the past 200 years have been the result of new technologies reshaping and rewiring our economic systems.

Consider, for example, how the introduction of interchangeable parts in the nineteenth century revolutionised manufacturing, or how Henry Ford’s assembly line extended the division of labour inside the factory during the early twentieth century. In the 1980s, the just-in-time revolution reduced the need for massive inventories, and the globalisation of supply chains during the 2000s allowed for greater specialisation. These process innovations, made possible by the emergence of new energy and communication technologies, drove economic growth by changing not just what companies produced, but how they produced.

Before the long-term usefulness of generative AI can become apparent, the hype, and the panic, must subside. Whatever its shortcomings, its introduction clearly represents an astounding technological leap. To ensure that it benefits all of us as workers, consumers, and entrepreneurs, we must provide all businesses with access to these revolutionary tools, rather than hand the key to the next great economic transformation to a few large incumbents and hope they don’t lock out everyone else.

Diane Coyle, professor of Public Policy at the University of Cambridge, is the author, most recently, of “Cogs and Monsters: What Economics Is, and What It Should Be” (Princeton University Press, 2021). Copyright: Project Syndicate, 2023. www.project-syndicate.org