The Second Machine Age is a 2014 book by MIT economists Erik Brynjolfsson and Andrew McAfee, published in January 2014 by W. W. Norton. It argues that digital technologies — computing, AI, robotics, and networks — constitute a "second machine age" comparable in scope to the Industrial Revolution, and pairs that argument with an account of the resulting wealth and inequality dynamics and a set of policy prescriptions. It follows the authors' shorter 2011 book Race Against the Machine. Many of the AI-economy sources in the labor-economics literature descend from its framing.
Core argument
The book contrasts two eras. The first machine age, the Industrial Revolution, substituted machine power for human muscle; drawing on anthropologist Ian Morris's index of social development, the authors argue that the steam engine bent the curve of human history more than any earlier development. The second machine age does for mental power what steam did for physical power, and the authors argue it operates at exponential, digital, and combinatorial rates that the first did not. They state three conclusions: that digital technologies have reached an inflection point at which their full force is being felt; that the transformation will be profoundly beneficial; and that it will also bring difficult challenges, especially for workers. They identify three characteristic dynamics:
- Exponential — Moore's Law-style doubling continues across compute, storage, and network bandwidth.
- Digital — zero-marginal-cost reproduction enables scale that industrial technology could not.
- Combinatorial — the building blocks multiply, and every new block enables new combinations, so innovation compounds.
Chapter-by-chapter summary
Part one: the technology (chapters 1–6)
1. The big stories. The authors introduce the two machine ages and the book's three conclusions, citing Martin Weitzman's view that long-term growth in advanced economies is dominated by technical progress.
2. The skills of the new machines: technology races ahead. The authors describe tasks that computers had recently begun performing after being considered beyond them. In 2004 Frank Levy and Richard Murnane's The New Division of Labor had cited driving in traffic as a task computers would struggle with; within a few years Google's autonomous cars were driving on public roads. Other examples are Apple's Siri (introduced with the iPhone 4S in fall 2011), machine translation, IBM Watson's Jeopardy! victory, and robotics, where the authors discuss Moravec's paradox (that high-level reasoning is easier for computers than perception and mobility), Honda's ASIMO falling on stairs in a 2006 demonstration, and Rethink Robotics' low-cost Baxter robot. They conclude that these developments indicate an inflection point.
3. Moore's Law and the second half of the chessboard. The authors argue that Moore's Law is not a physical law but a sustained record of engineering success made possible by the looser physical constraints on digital components. Using Ray Kurzweil's retelling of the fable of rice on a chessboard, they note that exponential growth becomes hard to comprehend in the second half of the board. Taking 1958 (when the Bureau of Economic Analysis first tracked corporate IT investment) as a start and an eighteen-month doubling period, they calculate that US business entered the second half of the chessboard after thirty-two doublings, in 2006, a calculation they describe as a simple exercise. They note that improvements extend beyond processors to sensors, storage and bandwidth.
4. The digitization of just about everything. Digital information is non-rival and has near-zero marginal cost of reproduction, and the volume of digital data is growing rapidly. The authors use the navigation app Waze to show innovation built on layers of digitization (digital maps, GPS, social data and sensor data from users' phones).
5. Innovation: declining or recombining? The authors engage with Robert Gordon's and Tyler Cowen's arguments that the major innovations have been used up. They argue that digital technology is a general purpose technology (defined, following Gavin Wright, as a deep new idea with the potential for important impacts across many sectors) and that innovation is fundamentally recombinant, following Weitzman: new ideas combine existing ones, and digital building blocks, which are not used up, multiply the possible combinations. Instagram, built by a small team on existing infrastructure and reaching more than 30 million users, is one example.
6. Artificial and human intelligence in the second machine age. The authors argue that machines able to complete cognitive tasks are now available, showing broad abilities in pattern recognition and complex communication along with progress in natural language processing, machine learning and computer vision, and they expect AI to lower costs and improve outcomes across many uses. Their example of a life-changing application is OrCam, an Israeli device introduced in 2013 that reads text aloud to visually impaired users.
Part two: bounty and spread (chapters 7–11)
7. Computing bounty. Productivity growth, not more hours, drives living standards; the authors state that the average American would need only eleven hours of work per week to produce what a forty-hour week produced in 1950. Productivity growth slowed after 1973, prompting Robert Solow's 1987 remark that computers were visible everywhere but in the productivity statistics. Brynjolfsson's 1993 work on the "productivity paradox" argued that complementary innovations were needed, and the mid-1990s productivity surge is attributed mostly to IT. The authors cite Chad Syverson's comparison with electrification, where productivity gains lagged adoption by more than twenty years.
8. Beyond GDP. Free digital goods raise welfare while lowering measured GDP, as when free texting, Craigslist or Skype replace paid services. The authors propose consumer surplus as a better measure, citing Brynjolfsson's 1993 estimate that price declines in computers added about $50 billion a year in welfare and later work with Joo Hee Oh valuing internet use by the time people spend on it. They also discuss unmeasured new goods and intangible assets and call for new metrics.
9. The spread. Although the economy has grown, the authors report that between 1983 and 2009 the bottom 80 percent of the income distribution saw a net decrease in wealth (per Ed Wolff and Sylvia Allegretto); that the top 1 percent's earnings rose 278 percent between 1979 and 2007 against 35 percent for the middle; that median hourly wages grew by only 0.1 percent per year between 1973 and 2011; and that labor's share of US GDP, which averaged 64.3 percent from 1947 to 2000, fell to 57.8 percent in the third quarter of 2010. They argue that technology, rather than tax policy, trade or finance, is the main driver, noting that inequality rose faster in Sweden, Finland and Germany than in the United States over twenty to thirty years. They describe three pairs of winners and losers: skilled versus less-skilled workers (skill-biased technical change, including job polarization as routine tasks are automated), capital versus labor, and superstars versus everyone else, and they stress organizational coinvention as a complement to technology.
10. The biggest winners: stars and superstars. Drawing on Sherwin Rosen's 1981 economics of superstars and Robert Frank and Philip Cook's work on winner-take-all markets, the authors argue that digitization lets the best provider in a market serve all of it, so rewards follow relative rather than absolute performance. Examples include TurboTax's automation of tax preparation and J. K. Rowling as the first billionaire author. They describe income distributions shifting toward power laws.
11. Implications of the bounty and the spread. The authors describe the bounty in broad terms (more choice, variety and quality, cheaper necessities, medical advances) and weigh it against the spread. They review the debate over technological unemployment from the Luddites of 1811–1817 and John Bates Clark to John Maynard Keynes, consider an "android experiment" in which machines can do all human work, and assess globalization as an alternative explanation for wage trends, concluding that technology is the larger factor.
Part three: recommendations (chapters 12–15)
12. Learning to race with machines: recommendations for individuals. Using freestyle chess, where teams of amateurs with computers and a good process beat grandmasters and supercomputers, the authors argue that people should race with machines rather than against them. They identify ideation (generating good new ideas), large-frame pattern recognition and complex communication as areas where humans retain an advantage, and recommend education that develops these skills, including through self-directed learning and online courses.
13. Policy recommendations. The authors argue that the standard "Econ 101" playbook still applies in the near term because digital labor is far from a complete substitute for human labor. Their recommendations are: improve education (including better teacher pay with accountability and the use of technology such as MOOCs); restart startups and entrepreneurship; make more matches between workers and jobs; support basic research and use prizes; upgrade physical infrastructure (citing the American Society of Civil Engineers' 2013 grade of D+ and a backlog of over $3.6 trillion); welcome skilled immigrants; and tax wisely, through Pigovian taxes on negative externalities and taxes on economic rents, such as land.
14. Long-term recommendations. The authors reject halting technology or replacing capitalism. They review the history of basic-income proposals (Thomas Paine, Bertrand Russell, Martin Luther King Jr., and economists including James Tobin, Paul Samuelson and John Kenneth Galbraith) but argue, citing research on the non-financial value of work (Daniel Pink, Andrew Oswald, William Julius Wilson), that policy should encourage work. They favor Milton Friedman's negative income tax, an expanded Earned Income Tax Credit, and reduced taxes and mandates on labor, funded partly by Pigovian taxes, consumption taxes and a value-added tax. They also discuss the peer economy (Airbnb, TaskRabbit) and crowd work such as Amazon's Mechanical Turk, and welcome unconventional ideas.
15. Technology and the future. The authors discuss non-economic risks of a complex, tightly coupled digital infrastructure, including Charles Perrow's "normal accidents," cyberattacks, and existential risks, and questions such as the singularity. They close as non-determinists: technology creates possibilities, and outcomes depend on choices, with the long-run promise of the second machine age being to free human ingenuity for creativity, exploration and community.
Bounty and spread
The book describes two macroeconomic consequences that, in the authors' account, operate simultaneously. Bounty is the rise in aggregate wealth and welfare — cheap software, free information, productivity gains, and global network effects. Spread is the increasingly unequal distribution of those gains, driven by winner-take-most dynamics, skill-biased technical change, and superstar effects. The authors frame the central policy challenge as preserving the bounty while addressing the spread, arguing that the two are not opposed — both are real — but require different institutional responses.
Policy prescriptions
The book's prescriptions, several of which subsequently became common in the AI-economy conversation, are:
- Education reform — human-capital accumulation should match the velocity of technology, using digital tools such as MOOCs and pairing higher teacher salaries with accountability.
- Entrepreneurship, job matching and science — encouraging startups, better labor-market matching, basic research funding and prizes.
- Immigration — welcoming skilled foreign-born workers, a position the authors note is supported by both the Cato Institute and the Center for American Progress.
- Infrastructure — investment in physical infrastructure such as roads, bridges, ports and airports.
- Taxation — reducing taxes and mandates on labor, with Pigovian taxes, taxes on economic rents, consumption taxes and a value-added tax as alternatives.
- Safety nets — expansion of a negative income tax or the EITC rather than a universal basic income; Brynjolfsson and McAfee were more cautious on UBI than several later writers.
An earlier version of this page listed platform-based "Learning 2.0," combined digital-and-physical infrastructure, progressive consumption taxes and antitrust policy among the book's prescriptions. The book does not use the term "Learning 2.0" or propose antitrust measures; its infrastructure recommendation concerns physical infrastructure, and it mentions consumption taxes without the qualifier "progressive." The list above reflects the book's text.
Relation to later AI-economy literature
The book sits upstream of much of the subsequent labor-economics work on AI. Generative AI at Work (Brynjolfsson and Raymond) is a direct continuation that adds randomized-controlled-trial evidence, and Generative AI and the Nature of Work (same authors) updates the framework for the large-language-model era. The Simple Macroeconomics of AI (Acemoglu) responds to the book's arguments on productivity, and Applying AI to Rebuild Middle-Class Jobs (Autor) is related work by David Autor, whose research on skill-biased change and polarization the book cites. Brynjolfsson is a Vol. 1 co-editor of the The Digitalist Papers (Stanford, Volumes 1–2), whose Vol. 2 includes an Autor/Thompson essay on an expertise framework. Empirical follow-ups on displacement and augmentation include the Epoch AI analysis "How close is AI to taking my job?" (Source: epoch.ai) and the Anthropic Economic Index reports Anthropic Economic Index — January 2026: Economic Primitives and Anthropic Economic Index — March 2026: Learning Curves. The authors' 2017 sequel is Machine, Platform, Crowd — McAfee and Brynjolfsson (2017).
Reception and later assessment
Several of the book's framings remained influential into the 2024–2026 productivity-and-inequality debates. The bounty-and-spread distinction is reflected directly in those debates; the exponential, digital, and combinatorial dynamics map onto large language models; and the skill-biased technical-change framework remained foundational, though later qualified by the Autor/Thompson expertise framework.
Other elements have been contested or revised. On the magnitude of the "spread" in the AI era, Autor's 2025 expertise framework suggests a more nuanced picture in which AI can be accessibility-enhancing and democratizing as well as displacement-driving. On safety nets, the 2014 book was conservative on UBI, while later thinking — including work by Marinescu, Berggruen/Gardels, and Yelizarova in the digitalist papers — has moved toward more serious universal-benefit proposals. Platform dynamics, under-emphasized in 2014, were addressed in the sequel Machine, Platform, Crowd — McAfee and Brynjolfsson (2017).
Provenance
Authored by Erik Brynjolfsson and Andrew McAfee, both then at MIT, and published by W. W. Norton in January 2014. The raw file is a full-text PDF extraction including notes. Confidence is medium: the page summarizes a single source, and the schema reserves high for claims supported by three or more sources.
Relationships
- supports: Generative AI at Work / Generative AI and the Nature of Work / (Source: hbr.org) — direct continuations.
- supports: The Digitalist Papers (Stanford, Volumes 1–2) — Brynjolfsson's editorial framing descends from this book.
- related: The Simple Macroeconomics of AI — Acemoglu's response on productivity.
- related: Applying AI to Rebuild Middle-Class Jobs — Autor's related work.
- related: The Singularity Is Near — Kurzweil (2005) — source of the second-half-of-the-chessboard framing the authors borrow.
- successor: Machine, Platform, Crowd — McAfee and Brynjolfsson (2017) — same authors' 2017 sequel.
- contradicts: not directly, but AI Snake Oil — Narayanan and Kapoor (2024) and Why I Think AI Take-Off Is Relatively Slow (Cowen) would moderate the exponential and combinatorial framing.