Blog post

2018 Won’t See a Massive Productivity Boost From AI – 2019 Might Show It

By Andrew White | January 04, 2018 | 0 Comments

ProductivityInnovationArtificial Intelligence

I first want to note five different news articles and then show how they all connect:

Looking for a Productivity Miracle (Wall Street Journal, US print edition, Dec 12)
Technology-Driven Boon is Finally Coming (Wall Street Journal, US print edition, Dec 28)
Language Looms as the Next Big Frontier in AI (Financial Times, US print edition, Dec 30-31)
Zombie Companies Stalk a Broken Monetary System (Financial Times, US print edition, Dec 27)
How Intangible Assets are Changing Investments (The Economist, US print edition, Buttonwood: Out of Touch, Dec 23)

James Mackintosh, of the WSJ, highlights (Looking for a Productivity Miracle) a long-standing issue I have been blogging about for a long time. Despite our massive investments in IT and other innovative areas, our economy (in fact most developed and to a lesser degree, emerging economies) show a distinct lack of improving productivity. In fact, in some areas and more recently, we have seen the rate of productivity actually decline. This is extremely worrying since it bodes poorly for increased standards of living, something that has been central to our business and social psyche for a long time. If we all gave up and assumed that what we were to pass onto our children was less fun and useful than what we experienced, we might not bother with the whole effort. The promise of betterment and the premise of productivity is what makes everything worthwhile.  Or so we were led to believe.

Past blog: Raising Productivity is Our Number One Task

See the BLS video that neatly explains productivity on this page.

But the data is not with us, as Mr Mackintosh explains. In the second article, Greg Ip, also of the WSJ, reports (Technology-Driven Boon is Finally Coming) on some new research by Erik Brynjolffson and others (see Artificial Intelligence and the Modern Productivity Paradox: A Clash of Expectations and Statistics) that suggests that the problem with productivity (or the lack of it) is not a problem at all. The article reports that it can take many years for an innovation to become widespread (e.g. diffused) and mature enough that many more practical innovations ‘take off’.  As such early “results” of these first innovations are not seen in the official numbers though their worth slowly creeps across the economic landscape.

Brynjolffson and his colleagues have been writing about this idea for some time. They suggest that AI is a kind of general purpose technology, much like electricity. Like electricity it took a long time before supplemental innovations emerged that leveraged it such that productivity booms did not follow easily the core innovation but were seen many years later. Thus, so the story goes, we are about to see the cycle repeated. Productivity has been slower recently since too few foundational innovations have been around, until now and AI.

I like their reasoning but I feel that their focus is too narrow. I don’t think AI is the key here, but I do think that the kind of innovations developed in the last twenty years have not delivered the value we all expected and sought.  I do believe that what Brynjolffson sees with AI is applicable at a much higher and broader level.  I will explain shortly.

In “Language Looms as the Next Big Frontier in AI” Richard Waters of the FT actually explores some of the patterns Brynjolffson and others (Gartner’s Tom Austin for one) have seen with AI.  In context to the topic of how productivity will develop, AI is a kind of platform. Alone it does little for anything or anybody, and so it needs to be applied. It is the application of the AI, and the needed management capability and educated workforce that creates the productivity-inducing opportunity that a) improves the right kind of productivity, and b) shows up in the public numbers. This combination of platform (Brynjolffson’s and economists’ ‘general purpose technology’), application, and timing of other factors such as management capability to reorganize and workforce skills, is what is needed for a productivity boon (to be recognized and measured).

What Richard Waters does is demonstrate nicely how AI needs to be applied to specific (business?) problems or opportunities in such a way that a known outcome can be improved, or an new outcome created. This is the essence of the translation or the gap between an invention (no commercial value yet) to an innovation (commercial value realized). Or, to be more granular, different kinds of innovation. I have mused in this before in my blog by suggesting that general purpose technologies are practically ‘platform’ innovations in which specific ‘product or service’ innovations are spun out or built on. Finally ‘process’ innovation concerns the method by which platforms or products and service innovations are delivered to market.

Past blog: The Ongoing Productivity Paradox and Are we out of big ideas?

Hidden in all of Brynjolffson’s (and many others) work and key here is timing. And it is timing of the availability and maturity of each part that has been against us, until now.  The timing of these combined productivity-inducing seeds is key:

• Platform (that is broad enough and mature enough to support productivity-inducing innovations)
• Products/service innovations conjoined and built on a platform
• Management capability (to exploit above points)
• Workforce skills (to exploit above points)

The lack or out-of-sequence timing and maturity of any one of the four results in a misstep and loss of apparent productivity-inducing results.  Any two or three might generate value; but when and if all four are aligned at the right time, we observe (so my theory goes) a productivity boom.

In “Zombie Companies Stalk a Broken Monetary System”, Chris Watling, also of the FT, reports on a political dimension that bears a heavy weight in the topic of why productivity is lacking and why the timing of the needed seeds have been inhibited to conjoin. Zombie business, much written about in places like Italy, but present everywhere and increasingly since the financial crisis, are a dead weight to the invisible hand of market forces and the need to reallocated capital to more productive use. Zombie businesses, while keeping short term unemployment down, are limiting our economies natural ability to repair themselves. Until and if we get the politicians out of the way, those seeds of productivity noted above will struggle for water and oxygen. Worse, public sector leaders who accept these seeds as face value, would rather justify their existence in the idea that they can determine how best, through policy and regulation, to control those seeds.

The IMF have written about zombie firms and how their maintenance seems to be impacting the efficient allocation of assets to more productive uses:

  • “Firms with weaker balance sheets are found to have reduced their investment rate in intangible assets – measured as a share of total value added – by 0.5 percentage points more than their less vulnerable counterpart
  • One channel through which the global financial crisis may have persistently weakened TFP growth is lower investment in intangible capital, such as R&D, in vulnerable firms.  Agh ion and others (2002) show that firms face credit constraints after severe downturns, R&D expenditure becomes pro-cyclical, imparting future productivity growth.”
  • “On average across business sectors in advanced economies, measured capital misallocation seems to have increased both before and after the global financial crisis (Figure 9).
  • “Growing misallocation during the pre-global-financial-crisis financial boom is consistent with results for the Spanish manufacturing sector in Gopinath and others (2015), who link the increased misallocation of capital in Southern Europe to the start rise in poorly intermediates capital flows following the inception of the euro (see also Reis 2013; Borio and others 2016; and Dias, Marques, and Richmond 2016).  The global financial crisis might have worsened capital allocation further by impeding the growth of financially constrained firms relative to their less constrained counterparts. Indeed, the divergence in TFP paths between both types of firms shown in Figure 7 was accompanied by a growing gap in their marginal revenue product of capital, as factors of production were adjusted and reallocated across firms only slowly.
  • “Possibly slowing this reallocation further has been that banks may have “evergreened” loans to weak firms to delay loan-loss recognition and the need to raise capital – particularly in continental Europe where progress toward addressing banking sector problems has been slower than in some other advanced economies such as the United States.  Together, these forces may have fostered the emergence of some “zombie firms” and thereby further increased misallocation of capital.”

From Gone with the Headwinds – Global Productivity, IMF Staff Discussion Note, G Adler, R Duvl, D Furceri, S K Celik, K Koloskova, M Poplawski-Ribeiro, 2017

Past blog: The Good, The Bad, and the Get on with It (Interesting Economist Articles This Week).

In “How Intangible Assets are Changing Investments”, The Economist reports on a fascinating book I am part-way through. The book is “Capitalism Without Capital” and coupled with another article, the analysis explores how investments in tangible asset innovation yields different financial, economic and social benefits to investments in intangibles, such as IP, processes, and software. And this is where I come full circle.

Investment in software outstripped hardware in the US somewhere in the late 1990s, or a little before it depending on whose data you use. In the UK and other developed economies it’s generally about the same time or a little later. But hardware and software are very different things, and cannot be interchanged. Hardware is basically storage, infrastructure, and compute.  Yes, hardware needs software to do work. But software is more than that. Software is used to describe what a firm does; it is used to describe a process, a decision, an analytic. Software is used to compete, to sell, to drive, to stop. In fact, software is (I believe) the linchpin to the productivity miracle and the lack thereof.

Hardware was IT’s miracle since inception and up through to the late 1980s. Increasingly software is the new genie. AI is just one example of software in action; software is the root platform innovation (general purpose technology), if you will, and AI is another platform innovation built on software (a GPT that spin out of a previous GPT), itself now being built on with specific product and service innovations (special purpose technology). But there are more examples outside of just AI: ERP or more generally packaged business applications that capture differentiated and innovative business process, over time tending to commodity; advanced analytics and big data, now IOT, which captures distributed and embedded insight into the environment and events taking place. There are many more examples built on software.

The link between capital spending and the kind of spend is also mentioned in the above mentioned IMF paper:

Feedback loop between weak investment in physical capital and productivity

  • “Private fixed investment fell sharply in advanced economies in the aftermath of the global financial crisis – and weakened more gradually in emerging market economies and low-income countries – largely as a result of weak aggregate demand (IMF 2015a).  This drop is likely to have contributed to subdued labor productivity growth not only by weakening the contribution of capital deepening, but also by affecting TFP growth itself through a slower adoption of capital-embodied new technologies.
  • New capital equipment enables some innovations to find their way into actual production (Solow 1959).  For example, in the late 1990s and early 2000s technological changes such as internet use was “embodied” in new and increasingly powerful computers.  New investment may also facilitate TFP-enhancing organizational innovations – for instance, just-in-time manufacturing and supply chain management emerged during the 1980s-1990s thanks to new information technology equipment and software.  See further discussions in Wolf (1991) and Greenwood, Hercowitz, and Krusell (1997).

From Gone with the Headwinds – Global Productivity, IMF Staff Discussion Note, G Adler, R Duvl, D Furceri, S K Celik, K Koloskova, M Poplawski-Ribeiro, 2017

As Nicholas Carr asked in 2004, Does IT Matter?  I would accept that concerning hardware IT is closer to the commodity that Mr. Carr was implying.  As such hardware is closer to being part of our overall economic operating system. Cloud computing such as IaaS is the epitome of this. But software is very different and about to come into its own. It may not drive the productivity numbers up significantly this year even in the AI space, but the forces are building. It’s just a matter of time, and picking the right horses.

And as if to prove my point, there will be all manner of winners with AI – just not enough of them that will be democratized enough such that the technology will be successfully diffused enough to drive the productivity numbers.  Here is a current winner with AI:  Big-City Murders Drop to Historic Lows.


The Gartner Blog Network provides an opportunity for Gartner analysts to test ideas and move research forward. Because the content posted by Gartner analysts on this site does not undergo our standard editorial review, all comments or opinions expressed hereunder are those of the individual contributors and do not represent the views of Gartner, Inc. or its management.

Comments are closed