The J-curve: why AI is not yet shifting the statistics

It is worth returning to an article that appeared in the American Economic Journal: Macroeconomics. In it, Erik Brynjolfsson, Daniel Rock and Chad Syverson described a phenomenon that reads more pointedly today than in 2021, when the text was published. They called it the Productivity J-curve. On the surface, this is a matter of macroeconomics – in fact, it is a key to understanding what is happening to an economy in which technology is developing faster than the institutions that measure it.The starting point is simple. So-called general-purpose technologies – electricity, computers, and today AI – are not enough on their own. They require significant complementary investments from organisations: training, new processes, the redesign of roles and ways of working, new systems, new competencies.

These investments are largely intangible. And that means national statistics see them poorly, if at all.Hence the paradox that the authors call the J-curve. In the early phase of a technology, organisations are already spending a great deal – but the effects are not yet visible. To be more precise: the effects exist, but not in the form in which they are traditionally measured. An illusion of stagnation appears. Later, when the complementary investments begin to “work”, productivity rises faster than current expenditure would suggest – because we inherit value produced earlier in a form invisible to the accounts.

Brynjolfsson, Rock and Syverson illustrate this with US data. Adjusting the statistics for intangible investments related to computer hardware and software, total factor productivity (TFP) at the end of 2017 turns out to be as much as 15.9% higher than the official estimates suggest. In other words, some of the growth that “appears” only in later years in fact arose much earlier – only in a form we cannot easily grasp.And this is a thought worth lingering on. If the J-curve was true for computers and the Internet, it is most likely true for AI as well. This means that today’s disappointment – “we invest so much in AI, yet productivity stands still” – may simply be a picture of the bottom of the curve. The time between the deployment of a technology and its visible effects in the economy has never been short. It will not be short now either.

Something uncomfortable for decision-makers follows from this. Leaders who judge AI investments by short-term productivity metrics may pull back at the very moment when the most has already been invested. Whether an organisation moves through the J-curve depends less on the technology itself and more on whether it can afford patience with things that cannot yet be measured.Perhaps it is worth accepting that the invisibility of effects is not proof of their absence. Sometimes the opposite is true: it is proof that something is happening which we cannot yet see.

Source: https://www.aeaweb.org/articles?id=10.1257%2Fmac.20180386

This post is part of the project “People and Algorithms in Organisations: Competences to Work in the Digital Environment” (DIGIT_People and algorithms), funded by the NAWA – Narodowa Agencja Wymiany Akademickiej)