The Nobel-Prize winning economist Robert Solow noted in the late 1980s that “you can see the computer age everywhere but in the productivity statistics.” True then, true now.
Artificial intelligence (AI) technologies are the new computer age. Everyone is talking about them, new models are being developed and released at lightning speed, and millions if not billions of users are experimenting with them. Yet productivity in Canada and elsewhere continues to languish.
While there’s been much focus on the extent to which productivity growth could be boosted, there’s been much less on how soon such gains could occur, and still less on whether any of this can be reliably measured. These questions have critical implications on the development of policy to support the constructive diffusion of this technology.
A look back to the information and communications technology (ICT) revolution in the 1990s is instructive.
Although firms invested heavily then in ICT equipment, it took years for that to translate into higher productivity. And for good reason. The most important gains came from the technology’s use, not its production, and its use required not only substantial investments in ICT but also complementary investments in how firms were organized, staff training and so on — intangibles that are difficult to measure and come with adjustment costs that must be overcome before gains are realized. In the near term, those costs are highly visible and can make it appear that investments are not paying off.
Economists, the old saying goes, are always the last to know when something has happened, partly because of the time lag in gathering data. That was certainly true of the 1990s boom. The difficulty in measuring ICT impacts (in part, because implementation affected how to measure both output and prices) meant that the gains were likely accruing before the data confirmed them. In fact, the United States was posting substantial productivity gains that were only understood by policy makers later when data revisions of official data were released. In other words, the gains were there but didn’t show up in official productivity data until years later.
Why does this matter?
First, this state of affairs creates a thorny issue for macroeconomic policy. The apparent failure of investments to pay off can be mistaken for supply-side constraints in the economy that could generate inflation. Such a backdrop can lead to tighter monetary and fiscal policy that, in turn, chokes off growth and limits the economy’s ability to absorb the new technologies.
Second, this perception can become a damaging self-fulfilling prophecy: when investment doesn’t seem to pay off, further investment is discouraged.
Of course, we are not living in the 1990s. Policy makers and firms have learned from their experiences. Statisticians have substantially improved their ability to measure intangibles. The risk of missing productivity gains as they occur is therefore lower than it was three decades ago (although not eliminated). Moreover, firms’ capacity to rapidly scale AI is much greater than in previous tech booms: firms and users, for example, are able to experiment more quickly than was previously the case.
Nevertheless, we can expect a rocky road ahead. With any new general-purpose technology, there are unknowns. Past experience can be a guide but never a prophecy, and the debate on how and when AI might boost productivity is far from settled.
Further, AI adoption presents its own challenges. A recent survey points to a range of issues, including accessing financing, hiring new skilled workers and upskilling the existing workforce. These challenges existed with ICT adoption as well. But the survey also points to issues around the privacy and security of data. If anything, those risks have been understated. Addressing them will be fundamental to the speed with which AI can, and should be, diffused, and thus to its impact on productivity.
To this end, updated privacy legislation that sets the appropriate ground rules on personal data use and monetization is fundamental to the diffusion of AI. But privacy legislation in Canada (and the regulatory framework for data and AI) continues to languish years after it was first introduced in Canada’s Parliament. Against this backdrop, Canada has announced substantial investments in computing, including quantum. Although this is a welcome development, the focus is on something tangible — compute power — and what we learned from the 1990s is that a greater focus is required on investment in the intangibles that make these tangibles more effective. That is surely even more important today in a world where intangibles are a critical driver of economic growth.
