Machines, Platforms, Crowds, and the Synergies Between Them

The book Machine Platform Crowd by Andrew McAfee and Erik Brynjolfsson provides a superb overview of three of the most important current trends in technology. The authors expand on ideas they introduced in The Second Machine Age (which I wrote about here), discussing which tasks are likely to be automated and how platforms bring about complementarities between people and algorithms. They outline how certain tools enabled by digital platforms—including crowdsourcing and blockchain—are being adopted by businesses or could impact them in future. In this blog post I point out one highlight of the book, then expand on a topic that intrigues me: how positive feedback loops between machines, platforms, and crowds might set up a virtuous cycle of expanding technological capabilities.

One especially valuable passage in the book is the accessible introduction to one of the most powerful economic theories of organization, Transaction Cost Economics (TCE). TCE emerged in the 1970s and its originator, Oliver Williamson, received the 2009 Nobel Prize in Economics, yet I have seen relatively little coverage of the theory in popular social science books. Put succinctly, TCE helps explain why organizations exist. It explains why some transactions are governed by market mechanisms (i.e. contracts) and others are governed by hierarchies (i.e. organizations), by analyzing how the relative costs of these competing forms of governance vary depending on the nature of the transaction. The large academic literature on TCE helps explain why firms shift toward or away from vertically integrated structures over time. Integration is favored when contracts are harder to write, for example when monitoring of effort or output is trickier or when rapid change generates uncertainty about the future.

Brynjolfsson and McAfee bring this theory to bear to examine what the emergence of digital platforms means for the size of organizations. The rise of Big Data reduces our uncertainty about transaction partners, as we know more about their track record. In addition, transaction platforms can incorporate tools to reduce the likelihood we get defrauded by counterparts we transact with over the platform. These factors lower the cost of using market mechanisms to govern transactions. More transactions can be governed by markets rather than hierarchies. As a result, digitization can drive down the size of organizations, and mean workers are increasingly treated as contractors rather than employees (a source of considerable controversy!).

The book is arranged in three sections each devoted to one of the phenomena mentioned in the title. (Machine here refers to machine learning technology and Crowd refers to decentralized modes of organizing). To expand on the book, one could undertake a structured analysis of how the three phenomena mutually interact and (potentially) reinforce one another. I’ve sketched out the beginnings of such an analysis in the Figure below. There are at least seven mechanisms generating synergies between machines, platforms, and crowds, several of which are mentioned in the book, and several of which I’ve added:

1. Machine learning improves digital platforms through two main mechanisms. First, it improves the quality of search tools which allow digital platforms to make better quality matches between users and complements. This helps users access the long tail of products that are hard to provide in offline settings. Example: Amazon’s recommendation algorithm helps users find items they want to buy.
   
2. Second, machine learning tools reduce costs related to processing transactions over platforms. Machine learning helps platforms verify identities and detect fraud.*
3. Crowds contribute to the effectiveness of platforms. User generated feedback is an important basis for establishing the trust that allows transactions to happen on digital platforms. For example, users rate their Uber drivers—and drivers rate passengers. User feedback weeds out bad actors, thereby improving users’ confidence in the platform.
4. Platforms, in turn, enable machine learning. McAfee and Brynjolfsson identify the five critical inputs of machine learning as (i) data, (ii) algorithms, (iii) networks, (iv) the cloud, and (v) hardware. Platforms directly improve machine learning tools by generating masses of data on user interactions. This data can be used to train algorithms to make predictions, which can feed into automated decisions made by the algorithm. For example, Google uses past data on who tends to clicks on what in order to optimize its ad-serving algorithms.
5. Platforms can be used to run innovation contests, which employ the “wisdom of the crowd” to solve difficult technical problems. Innovation contests may be used for many types of problems but have proven particularly effective for firms searching for better algorithms. In this way, platforms and crowds jointly contribute a key input to machine learning. For example, Netflix used an innovation contest to find a better user recommendation algorithm.
6. Machines and platforms empower crowds. Machines and platforms democratize entrepreneurship by providing tools for anyone to go about building their own business. Machine learning tools are not exclusively used by large firms: powerful tools such as TensorFlow have been made open source, meaning anyone may adopt them for free. Other tools, such as IBM’s Watson or Microsoft’s Azure, are available to entrepreneurs to incorporate as modules in their enterprise via an API.
7. And, finally, crowdfunding platforms allow entrepreneurs to raise financial capital to support their business. Some crowdfunding is “reward-based” (e.g. Kickstarter) allowing the entrepreneur to effectively make pre-sales which then fund production, while other platforms exist for founders to raise equity or debt finance from a broad pool of small investors.

These inherent complementarities represent a positive feedback loop which could generate change at an accelerating pace. This is, at once, both exciting and intimidating. With the world struggling to keep up with past technological changes, it is tempting to wish for things to slow down a little. But if the ever-expanding technological frontier can be used to solve problems and create value—rather than merely help a few people capture a larger share of value—then this positive feedback loop is cause for tremendous optimism.

 ___________
Footnote 
*Machine learning to improve fraud detection is discussed in some detail in the book Prediction Machines by Agrawal, Goldfarb, and Gans.

The Edison Conjecture: Why Successful New Technologies Offer Unfamiliar Solutions to Familiar Needs

The Google Glass and Apple Watch were launched in 2014 and 2015 respectively. Both devices were “wearable tech,” both embodied leading-edge technology, and both were launched to much fanfare with sophisticated marketing campaigns. But while the Apple Watch sold millions of units in its first quarter on the market, Google’s Glass flopped. What led to such different outcomes for these two innovative technologies?

To provide a thoughtful answer to this question, let me wind back the clock to 1878 when Thomas Edison and his employees were conducting pioneering experiments into electricity.¹ Edison faced a tremendous challenge, described in a classic research article by Andrew Hargadon and Yellowlees Douglas:² electricity had the potential to be put to numerous possible uses, but it was deeply unfamiliar to the general public at the time. It was regarded with a mix of curiosity and fear. The risk of death by electrocution was salient in the public’s mind after some high-profile incidents. Why should people let electricity’s mysterious, ghost-like presence into their home or workplace?

Part of what underlay Edison’s success in commercializing electricity was in how he framed it as a solution to familiar needs. Rather than launch with an immediate wave of electricity-powered domestic gadgets (which, indeed, came later), he focused on the familiar problem of providing light after the sun has set. There was an existing solution for this: natural gas was piped under streets to feed gas lamps which created light as a by-product of combustion. Rich people’s houses had gas lines in the walls feeding lamps on the walls and ceiling.

Edison’s success came partially from the pioneering inventions of his laboratory, which created the first carbon incandescent light bulb in 1879. But Edison’s success was also due to his savvy commercialization strategy. He organized the Edison Electric Light company under existing statutes that applied to gas companies. This allowed him to install underground lines and compete directly with gas companies, many of which would have happily seen him fail. Electric lightbulbs were launched as a direct substitute for gas lamps. Electric light fixtures were designed to resemble gas lamps, including the installation of lamp shades, which are not functionally needed for electric bulbs, but retaining them increased the sense of familiarity to the users.

A lamp shade's original function was to prevent
soot from gas lamps ruining the ceiling and walls

Edison’s success in commercializing his laboratory’s ideas is one reason we revere him as one of the most influential innovators in history.

What else can we learn from Edison’s success? I have previously written about how innovation can be viewed as a matching process in which needs are matched up with solutions (see here). The stories of Edison and Apple and Google point towards a powerful lesson for would-be innovators. When innovators match up needs and solutions, the general public may be more or less familiar with the needs they tackle and the solutions they proffer. The general public also have only a limited tolerance for things that are unfamiliar. For this reason, an innovator offering an unfamiliar solution to an unfamiliar need—even if it is a very real need—is unlikely to succeed. I refer to this as ‘The Edison Conjecture.’³

In this two-by-two matrix, the sweet spot is applying an unfamiliar solution to a familiar need. Applying a familiar solution to an unfamiliar need can also work. Applying an unfamiliar solution to an unfamiliar need is a ‘bridge too far’ for most potential adopters.

Returning to the Apple versus Google example: both companies launched technologically pioneering consumer electronic products that fall in the category we now call “wearable tech.” Both solutions were unfamiliar at the time they were launched. The biggest difference, I suggest, is that Apple’s Watch is couched in the familiar, much as Edison’s electric light bulbs were launched as a replacement for (familiar) gas lamps. The Apple Watch solves two familiar problems: telling the time (for which we’ve used watches for over a century) and receiving notifications (for which we had used pagers since the 1980s and mobile phones since the early 2000s). Google’s Glass, on the other hand, holds potential for numerous hypothetical applications, but did not solve any clear, familiar problem for the user.

As always, there are other important factors that helped the Apple Watch succeed and other factors that led the Google Glass to fail (e.g. Google underestimated the seriousness of privacy concerns relating to the Glass’s integrated camera⁴). But the analysis of the (un)familiarity of needs and solutions carries what I think is a valuable message for managers aspiring to introduce ground-breaking innovations. If you want to introduce an unfamiliar solution, it better be matched up with a familiar need, otherwise resistance to it will simply be too great.

_________

Footnotes ¹ Edison was not a lone genius; rather he ran a successful organization which churned out inventions. It has been said that “Edison is in reality a collective noun” (see the article referenced below).
² Reference: Hargadon, A. B.; Douglas, Y. (2001). When innovations meet institutions: Edison and the design of the electric light. Administrative Science Quarterly, 46(3), 476-501. (link)
³ I refer to it as a conjecture because it has not, to my knowledge, been tested in a large sample study.
⁴ An interesting post-script in the story of Google Glass: the concept has been pivoted to address a market in which privacy is less of a concern. The Glass has been relaunched as an B2B service for enterprise productivity applications.