We have been here before. In the 1960s, computers were too enormous and expensive to be used by anyone but the largest government agencies and businesses. Yet so great were concerns about “automation” that US President Lyndon Johnson launched an inquiry into the danger that computer-based technologies might “eliminate all but a few jobs.” It wasn’t to be. By the 1970s, there was no sign of a productivity surge, and fears of mass technological unemployment subsided.
Personal and business computer use then soared in the 1980s; but by 1990, as the economist Robert Solow famously observed, Information Technology (IT) was “everywhere but in the productivity statistics.” With mobile phones, the internet, ever-expanding hardware capacity, and growing software capabilities promising a new connectivity-based productivity revolution, everyone at the World Economic Forum in 2000 wanted to hear from the leaders of “Information and Communications Technology” (ICT). Cisco CEO John Chambers predicted that ICT would enable the US economy to grow by 5% per year for the foreseeable future, and that “The internet will form half of gross domestic product by 2010.”
Then there was “big data,” “the digital economy,” “machine learning,” and now artificial intelligence. None, so far, has had any measurable impact on medium-term growth rates. A case can be made that generative AI, owing to its self-learning capability, represents more than just another stage of technological development. But there are still two reasons why it, too, may not show up in growth data.
First, a large and probably growing share of economic activity involves a zero-sum struggle for competitive advantage with no positive impact on either measured growth or human welfare. Using basic internet search, and now sophisticated large language models, lawyers are increasingly able to analyze every possible precedent before presenting their arguments. But if the opposing law firm can do the same, the result is an arms race in which neither party has a durable advantage.
For at least two decades, experts have warned that after the steady decline of manufacturing jobs, professional services such as the law would be next in line for automation. But employment and pay in the field of commercial law continue to grow.
Similarly, marketing departments can use AI to produce ever more targeted and effective communications to influence consumer choice. If their competitors are doing the same, however, there is no benefit to end consumers, and no boost to GDP.
Conversely, AI will almost certainly deliver huge human-welfare benefits almost for free. The late Martin Feldstein, writing in 2017, correctly observed this phenomenon at work in the previous three decades of remarkable IT and ICT progress. By then, smartphones boasted many thousands of times more processing power and memory than the biggest computers of the 1960s, enabling vastly more communication, data storage, video and image sharing, and so forth. Yet the share of GDP accounted for by the telecoms sector had hardly changed, leading Feldstein to conclude that “low growth estimates fail to reflect the remarkable innovations in everything from health care to Internet services to video entertainment that have made life better during these years.”
Likewise, Google DeepMind’s AlphaFold Protein Structure Database (which predicts a protein structure from its amino acid sequence) is set to accelerate drug discovery while slashing the cost of research. But once drugs come off patent, their prices fall toward their marginal cost of production, and their contribution to measured GDP collapses.
If, by 2070, an AI-enabled acceleration of knowledge acquisition has furnished us with a wonder drug that gives everyone a 100-year life of perfect health, and which is produced in wholly automated factories powered by cheap nuclear fusion, it will count for almost nothing in global GDP. The more powerful a technology, the more rapidly it disappears from measured GDP.
At the same time, AI will have massive potential to exacerbate the harms to human welfare that previous generations of ICT have already produced. Deepfake capabilities are already driving an explosion of online scams, and AI-powered social-media algorithms are deepening political polarization and probably contributing to what social psychologist Jonathan Haidt sees as an epidemic of mental illness among young people. Yet none of these negatives show up in measured GDP, either.
For good or ill – or merely as an enabler of ever more intense zero-sum competition – AI will have a pervasive and perhaps transformative impact on society. But the hope that it will unleash a sustained increase in measured productivity and GDP growth is probably a delusion.
