Machine, Platform, Crowd: Harnessing Our Digital Future

by Andrew McAfee

In all companies and industries, machine, platform, and crowd have counterparts. For machine intelligence, the counterpart is the human mind. Accountants with spreadsheets, engineers with computer-aided design software, and assembly line workers next to robots are all examples of mind-and-machine combinations. The counterparts of platforms are products—in other words, goods and services. A ride across town is a product, while Uber is the platform people use to access it. The same is true with accommodations and Airbnb, or news stories and Facebook. For the crowd, the counterpart is the core: the knowledge, processes, expertise, and capabilities that companies have built up internally and across their supply chains. The core of GE Appliances designs, manufactures, and markets refrigerators and ovens; NASA’s core builds spaceships and tries to better understand our universe; Microsoft’s core capabilities include developing personal computer operating systems and applications.

Research in many different fields points to the same conclusion: it’s exactly because incumbents are so proficient, knowledgeable, and caught up in the status quo that they are unable to see what’s coming, and the unrealized potential and likely evolution of the new technology. This phenomenon has been described as the “curse of knowledge” and “status quo bias,” and it can affect even successful and well-managed companies. Existing processes, customers and suppliers, pools of expertise, and more general mind-sets can all blind incumbents to things that should be obvious, such as the possibilities of new technologies that depart greatly from the status quo.

survey conducted by the economist Shaw Livermore and published in 1935 found that over 40% of the industrial trusts formed between 1888 and 1905 had failed by the early 1930s. Another 11% were “ ‘limping’ units, whose records were . . . a mixture of good and bad. . . . In general, the bad results have been witnessed in the more recent years of the period under review.” Of the trusts that survived, most became much smaller. A study by economist Richard Caves and his colleagues of forty-two manufacturing firms that were dominant in 1905 and still in existence in 1929 found that their average market share declined by over one-third, from 69% to 45%.

The big gains came not from simple substitution of electric motors for steam engines, but from the redesign of the production process itself.

The successful companies of the second machine age will be those that bring together minds and machines, products and platforms, and the core and crowd very differently than most do today.

System 1 operates automatically and quickly, with little or no effort and no sense of voluntary control. System 2 allocates attention to the effortful mental activities that demand it, including complex computations. The operations of System 2 are often associated with the subjective experience of agency, choice, and concentration.

That practical conclusion, we believe, is that we need to rely less on expert judgments and predictions.

But the overall pattern is clear: in case after case, when a model can be created and tested, it tends to perform as well as, or better than, human experts making similar decisions. Too often, we continue to rely on human judgment when machines can do better.

Because System 1 operates automatically and cannot be turned off at will, errors of intuitive thought are often difficult to prevent. Biases cannot always be avoided, because System 2 may have no clue to the error.

How can we make use of all this knowledge about biases and glitches in System 1 and System 2? How can it lead us to be smarter about making decisions, and to make better ones? The most obvious approach is to consider letting the machines make the decisions when and where they can—to let pure digital instances of System 2, turbocharged by Moore’s law and fed from a fire hose of data, come up with their answers without input from System 1. Over time, this is exactly what more and more companies are doing.

Careful selection of the right data inputs and the right performance metrics, especially the overall evaluation criterion, is a key characteristic of successful data-driven decision makers.

Among excellent companies a fundamental shift is taking place: away from long-range forecast, long-term plans, and big bets, and toward constant short-term iteration, experimentation, and testing. These organizations follow the computer scientist Alan Kay’s great advice that the best way to predict the future is to invent it. They do this in many small steps, getting feedback and making adjustments as necessary, instead of working in private toward a distant event with a confidently predicted outcome.

In many important cases, we simply don’t and can’t know what rules we ourselves are using to get something right.

Computers are getting good at tasks like determining people’s emotional states by observing their facial expressions and vocal patterns, but this is a long, long way from doing the things we just listed. We’re confident that the ability to work effectively with people’s emotional states and social drives will remain a deeply human skill for some time to come. This implies a novel way to combine minds and machines as we move deeper into the second machine age: let the computers take the lead on making decisions (or judgments, predictions, diagnoses, and so on), then let people take the lead if others need to be convinced or persuaded to go along with these decisions.

What are the characteristics of the winners in the platform battles we’ve observed, and those that will be waged in the future? While they are not all identical, we’ve already seen that winning platforms—those that grow quickly and deliver value to both their participants and their owners—tend to share a few characteristics. 1. They’re early to the space. They don’t have to be the first (Android certainly wasn’t), but they had better not be so late that many potential participants have already chosen a platform and network effects have taken hold. 2. They take advantage of the economics of complementary goods whenever possible, realizing that low prices for one complement lead to increased demand for others. 3. They open up their platforms to a broad range of contributors and contributions. This variety increases total consumer surplus, especially if some of the contributions are offered to users for free, and it pushes the demand curve outward in a series of nudges. 4. While they maintain a broad rule of openness, they also curate their platforms to deliver a consistent and positive experience to participants, and to minimize unpleasant surprises.

Organizations have a lot of virtues, but they often get in their own way; they do things that are counterproductive and that worsen their performance in innovation, R&D, and virtually every other area. Organizational dysfunctions are real things—not only the subjects of countless Dilbert cartoons—and they do keep the core from working as well as it could.

starts with the basic rule of thumb that markets often have lower production costs (all the costs that come with making goods and services), while hierarchies typically have lower coordination costs (all the costs associated with setting up the production and keeping it running smoothly). The technologies discussed in this book are great cost reducers, and especially good at reducing coordination costs. It’s easy to see how search engines, cheap global communication networks, and the free, perfect, and instant economies of information goods in general would drive down coordination costs.

Of the many elaborations of TCE, those that are most relevant here are the concepts of incomplete contracts and residual rights of control. In pathbreaking work, Sandy Grossman and Oliver Hart asked, “What rights does the owner of a firm have that a non-owner doesn’t?” They reasoned that ownership has value only to the extent that contracts are incomplete; if every possible contingency for use of a building, machine, or patent were spelled out in contracts, then labeling one party the “owner” of the asset would confer no additional rights.

However, when contracts are incomplete, owners have the residual rights of control, meaning they can do whatever they want with the asset except for what’s in the contract.‡‡ If there’s nothing in any contract about what colors you can paint your car, or when to change the oil, or whether to replace the music system, or even whether to sell it to a little old lady down the street for $1, then you, as the owner, pretty much have the right to make those decisions. Hart dove deeper into these questions, including through a particularly influential set of papers with John Moore§§ and with Bengt Holmström.

In practice, this means that when two people work together on a project and one of them owns an essential asset, like a machine or factory necessary to produce the output, then that owner has the residual rights of control. If either of them comes up with some great new idea to increase the output of the machine, the owner can implement it without further consultation. The nonowner, in contrast, needs the owner’s permission. That requirement gives the owner bargaining power to, for example, insist on a cut of the additional output. TCE calls this the “hold-up problem.” As a result, ownership affects the incentives for innovation, whether large (like a new product idea) or small (like a better way to sort inventory). The bottom line is that changing ownership changes incentives, and therefore results. Employees working with someone else’s assets have different incentives from those of independent contractors who own their own assets. That’s an important reason why firm boundaries matter. A crucial question in the efficient design of a company, a supply chain, or a whole economy is how the assets, and thus incentives, are arranged.

The two of us are pessimistic that totally decentralized, purely crowd-based entities like The DAO will ever be economically dominant, no matter how technologically solid they become. They simply can’t solve the problems of incomplete contracting and residual rights of control that a company solves by letting management make all the decisions not explicitly assigned to other parties. Smart contracts are interesting and powerful new tools and there will be a place for them, but they don’t address the fundamental problem that keeps companies, as it were, in business. Companies exist in large part because well-functioning complete contracts are impossible to write, not because they’re too difficult or costly to enforce.

But things have not worked out that way. According to the Bureau of Labor Statistics, managers represented approximately 12.3% of the US workforce in 1998, but by 2015 this figure had increased to 15.4%. And there’s strong evidence that a lot of other jobs have become substantially more management-like over time. In 2015, economist David Deming published an intriguing study that looked at the demand for different skills throughout the US economy between 1980 and 2012. As expected, demand for routine skills, both cognitive and physical, declined sharply over this period as the standard partnership between minds and machines, which we described in Chapter 2, spread throughout the economy.

Deming was also able to assess demand shifts for what he calls the “social skills” of coordination, negotiation, persuasion, and social perceptiveness. He found that “social skill task inputs”—in other words, the overall use of these tasks—increased 24% between 1980 and 2012, while the use of “non routine and analytical skills” grew only 11%. What’s more, jobs that required high social skills increased as a share of total employment during this period, whether or not those jobs also required high math skills. Not all of these jobs are managerial, but it’s clear that the economy as a whole has, over the years, been demanding more of the things that good managers excel at: sensing people’s emotions and priorities, and getting them to work well together.

either case, the key practice for managers within these companies is that they try not to let their own biases and judgments play too large a role in determining which of the ideas they hear are the good ones, and thus worthy of implementation. Instead, they fall back whenever possible on the processes of iteration and experimentation to find unbiased evidence on the quality of a new idea. Managers, in other words, step away from their traditional roles as evaluators and gatekeepers of ideas. This shift is uncomfortable for some, who fear (with justification) that some bad ideas will see the light of day, but many of the most impressive companies and managers we’ve encountered believe the benefits are far greater than the risks. At the online education company Udacity, for example, egalitarianism about ideas led to a major positive change to the company’s business model and cost structure.

In addition to egalitarianism, and often in support of it, second-machine-age companies have high levels of transparency: they share more information more widely than has been typical. Wall Street Journal technology columnist Christopher Mims points out that information transparency and a flat, fast, evidence-based management style are highly complementary. As he puts it, “What makes this relatively flat hierarchy possible is that front-line workers have essentially unlimited access to data that used to be difficult to obtain, or required more senior managers to interpret.” Mims summarizes that the combination of egalitarianism and transparency “isn’t the end of middle management, but it is an evolution. Every company I talked to had middle and even senior managers who operated as player-coaches, tasked with both doing things and directing others.”