Watch Now


Commentary: Can AI and machine learning improve the economy?

Is your company determining what technologies could make your supply chain operations more productive?

A general purpose technology can become pervasive, can be improved and leads to complementary innovations. (Photo: Shutterstock)

The views expressed here are solely those of the author and do not necessarily represent the views of FreightWaves or its affiliates.

In this installment of the AI in Supply Chain series (#AIinSupplyChain), I tried to discern the outlines of an answer to the question posed in the headline above by reading three academic papers. This article distills what I consider the most important takeaways from the papers.

Although the context of the investigations that resulted in these papers looks at the economy as a whole, there are implications that are applicable at the level of an individual firm. So, if you are responsible for innovation, corporate development and strategy at your company, it’s probably worth your time to read each of them and then interpret the findings for your own firm.

‘Artificial Intelligence and the Modern Productivity Paradox’ (November 2017)

In this paper, Erik Brynjolfsson, Daniel Rock and Chad Syverson explore the paradox that while systems using artificial intelligence are advancing rapidly, measured economywide productivity has declined.

Recent optimism about AI and machine learning is driven by recent and dramatic improvements in machine perception and cognition. These skills are essential to the ways in which people get work done. So this has fueled hopes that machines will rapidly approach and possibly surpass people in their ability to do many different tasks that today are the preserve of humans.

However, productivity statistics do not yet reflect growth that is driven by the advances in AI and machine learning. If anything, the authors cite statistics to suggest that labor productivity growth fell in advanced economies starting in the mid-2000s and has not recovered to its previous levels.


Therein lies the paradox: AI and machine learning boosters predict it will transform entire swathes of the economy, yet the economic data do not point to such a transformation taking place. What gives?

The authors offer four possible explanations.

First, it is possible that the optimism about AI and machine learning technologies is misplaced. Perhaps they will be useful in certain narrow sectors of the economy, but ultimately their economywide impact will be modest and insignificant.

Second, it is possible that the impact of AI and machine learning technologies is not being measured accurately. Here it is pessimism about the significance of these technologies that prevents society from accurately measuring their contribution to economic productivity.

Third, perhaps these new technologies are producing positive returns to the economy, BUT these benefits are being captured by a very small number of firms and as such the rewards are enjoyed by only a minuscule fraction of the population.

Fourth, the benefits of AI and machine learning will not be reflected in the wider economy until investments have been made to build up complementary technologies, processes, infrastructure, human capital and other types of assets that make it possible for society to realize and measure the transformative benefits of AI and machine learning.

The authors argue that AI, machine learning and their complementary new technologies embody the characteristics of general purpose technologies (GPTs). A GPT has three primary features: It is pervasive or can become pervasive; it can be improved upon as time elapses; and it leads directly to complementary innovations.

Electricity. The internal combustion engine. Computers. The authors cite these as examples of GTPs, with which readers are familiar.

Crucially, the authors state that “a GPT can at one moment both be present and yet not affect current productivity growth if there is a need to build a sufficiently large stock of the new capital, or if complementary types of capital, both tangible and intangible, need to be identified, produced, and put in place to fully harness the GPT’s productivity benefits.”

It takes a long time for economic production at the macro- or micro-scale to be reorganized to accommodate and harness a new GPT. The authors point out that computers took 25 years before they became ubiquitous enough to have an impact on productivity. It took 30 years for electricity to become widespread. As the authors state, the changes required to harness a new GPT “take substantial time and resources, contributing to organizational inertia. Firms are complex systems that require an extensive web of complementary assets to allow the GPT to fully transform the system. Firms that are attempting transformation often must reevaluate and reconfigure not only their internal processes but often their supply and distribution chains as well.”

The authors end the article by stating: “Realizing the benefits of AI is far from automatic. It will require effort and entrepreneurship to develop the needed complements, and adaptability at the individual, organizational, and societal levels to undertake the associated restructuring. Theory predicts that the winners will be those with the lowest adjustment costs and that put as many of the right complements in place as possible. This is partly a matter of good fortune, but with the right roadmap, it is also something for which they, and all of us, can prepare.”

‘Does Machine Translation Affect International Trade?’ (August 2018)

In this paper, Brynjolfsson, Xiang Hui and Meng Liu explore the effect that the introduction of eBay Machine Translation (eMT) had on eBay’s international trade. The authors describe eMT as “an in-house machine learning system that statistically learns how to translate among different languages.” They also state: “As a platform, eBay mediated more than 14 billion dollars of global trade among more than 200 countries in 2014.” Basically, eBay represents a good approximation of a complex economy within which to examine the economywide benefits of this type of machine translation.

The authors state: “We show that a moderate quality upgrade increases exports on eBay by 17.5%. The increase in exports is larger for differentiated products, cheaper products, listings with more words in their title. Machine translation also causes a greater increase in exports to less experienced buyers. These heterogeneous treatment effects are consistent with a reduction in translation-related search costs, which comes from two sources: (1) an increased matching relevance due to improved accuracy of the search query translation and (2) better translation quality of the listing title in buyers’ language.”

They report an accompanying 13.1% increase in revenue, even though they only observed a 7% increase in the human acceptance rate.

They also state: “To put our result in context, Hui (2018) has estimated that a removal of export administrative and logistic costs increased export revenue on eBay by 12.3% in 2013, which is similar to the effect of eMT. Additionally, Lendle et al. (2016) have estimated that a 10% reduction in distance would increase trade revenue by 3.51% on eBay. This means that the introduction of eMT is equivalent of [sic] the export increase from reducing distances between countries by 37.3%. These comparisons suggest that the trade-hindering effect of language barriers is of first-order importance. Machine translation has made the world significantly smaller and more connected.”

‘The Productivity J-Curve’ (October 2018, revised January 2020)

In this paper, Brynjolfsson, Rock and Syverson develop a model that shows how GPTs like AI “enable and require significant complementary investments, including co-invention of new processes, products, business models and human capital. These complementary investments are often intangible and poorly measured in the national accounts, even when they create valuable assets for the firm” AND they “develop a model that shows how this leads to an underestimation of productivity growth in the early years of a new GPT, and how later, when the benefits of intangible investments are harvested, productivity growth will be overestimated.” Their “model generates a Productivity J-Curve that can explain the productivity slowdowns often accompanying the advent of GPTs, as well as the increase in productivity later.”

The authors find that, first, “As firms adopt a new GPT, total factor productivity growth will initially be underestimated because capital and labor are used to accumulate unmeasured intangible capital stocks.” Then, second, “Later, measured productivity growth overestimates true productivity growth because the capital service flows from those hidden intangible stocks generates measurable output.” Finally, “The error in measured total factor productivity growth therefore follows a J-curve shape, initially dipping while the investment rate in unmeasured capital is larger than the investment rate in other types of capital, then rising as growing intangible stocks begin to contribute to measured production.”

This explains the observed phenomenon that when a new technology like AI and machine learning, or something like blockchain and distributed ledger technology, is introduced into an area such as supply chain, it generates furious debate about whether it creates any value for incumbent suppliers or customers.

If we consider the reported time it took before other GPTs like electricity and computers began to contribute measurably to firm-level and economywide productivity, we must admit that it is perhaps too early to write off blockchains and other distributed ledger technologies, or AI and machine learning, and their applications in sectors of the economy that are not usually associated with internet and other digital technologies.

Give it some time. However, I think we are near the inflection point of the AI and Machine Learning Productivity J-curve. As I have worked on this #AIinSupplyChain series, I have become more convinced that the companies that are experimenting with AI and machine learning in their supply chain operations now will have the advantage over their competitors over the next decade.

I think we are a bit farther away from the inflection point of a Blockchain and Distributed Ledger Technologies Productivity J-Curve. I cannot yet make a cogent argument about why this is true, although in March 2014, I published #ChainReaction: Who Will Own The Age of Cryptocurrencies? — part of an ongoing attempt to understand when blockchains and other distributed technologies might become more ubiquitous than they are now.

Conclusion

Examining this topic has added to my understanding of why disruption happens. The authors of the Productivity J-Curve paper state that “the more transformative the new technology, the more likely its productivity effects will initially be underestimated.”

The long duration during which incumbent firms underestimate the productivity effects of a relatively new GPT is what contributes to the phenomenon studied by Rebecca Henderson and Kim Clark in “Architectural Innovation: The Reconfiguration of Existing Product Technologies and the Failure of Established Firms.” It is also described as Supply Side Disruption by Josgua Gans in his book, “The Disruption Dilemma,” and summarized in this March 2016 HBR article, “The Other Disruption.”

If we focus on AI and machine learning specifically, in an exchange on Twitter on Sept. 27, Brynjolfsson said, “The machine translation example is in many ways the exception. More often it takes a lot of organizational reinvention and time before AI breakthroughs translate into productivity gains.”

By the time entrenched and industry-leading incumbents awaken to the threats posed by newly developed GPTs, a crop of challengers who had no option but to adopt the new GPT at the outset has become powerful enough to threaten the financial stability of an industry.

One example? E-commerce and its impact on retail in general.

If you are an executive, what experiments are you performing to figure out if and how your company’s supply chain operations can be made more productive by implementing technologies that have so far been underestimated by you and other incumbents in your industry?

If you are not doing anything yet, are you fulfilling your obligations to your company’s shareholders, employees, customers and other stakeholders?

If you are a team working on innovations that you believe have the potential to significantly refashion global supply chains, we’d love to tell your story in FreightWaves. I am easy to reach on LinkedIn and Twitter. Alternatively, you can reach out to any member of the editorial team at FreightWaves at [email protected].

Dig deeper into the #AIinSupplyChain Series with FreightWaves.

Commentary: Optimal Dynamics – the decision layer of logistics?

Commentary: Combine optimization, machine learning and simulation to move freight

Commentary: SmartHop brings AI to owner-operators and brokers

Commentary: Optimizing a truck fleet using artificial intelligence

Commentary: FleetOps tries to solve data fragmentation issues in trucking

Commentary: Bulgaria’s Transmetrics uses augmented intelligence to help customers

Commentary: Applying AI to decision-making in shipping and commodities markets

Commentary: The enabling technologies for the factories of the future

Commentary: The enabling technologies for the networks of the future

Commentary: Understanding the data issues that slow adoption of industrial AI

Commentary: How AI and machine learning improve supply chain visibility, shipping insurance

Commentary: How AI, machine learning are streamlining workflows in freight forwarding, customs brokerage

Author’s disclosure: I am not an investor in any early-stage startups mentioned in this article, either personally or through REFASHIOND Ventures. I have no other financial relationship with any entities mentioned in this article.

One Comment

Comments are closed.

Brian Aoaeh

Brian Laung Aoaeh writes about the reinvention of global supply chains, from the perspective of an early-stage technology venture capitalist. He is the co-founder of REFASHIOND Ventures, an early stage venture capital fund that is being built to invest in startups creating innovations to refashion global supply chain networks. He is also the co-founder of The Worldwide Supply Chain Federation (The New York Supply Chain Meetup). His background covers the gamut from scientific research, data and statistical analysis, corporate development and investing for a single-family office, and then building an early stage venture fund from scratch - immediately prior to REFASHIOND. Brian holds an MBA in General Management, with a specialization in Financial Instruments and Markets, from NYU’s Stern School of Business. He also holds a Bachelor’s Degree in Mathematics & Physics from Connecticut College. Brian is a charter holding member of the CFA Institute. He is also an adjunct professor of operations management in the Department of Technology Management and Innovation at the New York University School of Engineering.