Wikipedia Deep Dive

Erik Brynjolfsson

14 min read

The Economist Who Wants You to Race With the Machine, Not Against It

Here's a question that keeps executives, policymakers, and workers up at night: Is artificial intelligence coming for your job? Erik Brynjolfsson has spent more than three decades trying to answer this question, and his conclusion might surprise you. He doesn't think the question itself is quite right.

Brynjolfsson argues that we're asking about replacement when we should be asking about augmentation. The difference isn't just semantic. It shapes everything from how companies deploy AI to whether the technology makes society richer or more unequal.

Martin Wolf, the chief economics commentator at the Financial Times, once wrote that no economist has done more to explain the revolutionary implications of information technology than Brynjolfsson. That's quite a statement, given that information technology has reshaped virtually every industry on the planet over the past half century.

From Denmark to the Digital Economy

Brynjolfsson was born in Roskilde, Denmark, a city about twenty miles west of Copenhagen that was once the capital of the Danish kingdom. His father, Ari Brynjolfsson, was a nuclear physicist, and the family has Icelandic roots—which might explain the distinctive surname that trips up so many English speakers.

The family eventually settled in Wayland, Massachusetts, a suburb about twenty miles west of Boston. There, at Wayland High School, Brynjolfsson distinguished himself enough to graduate as valedictorian. He then headed to Harvard, earning both a bachelor's degree and a master's degree in applied mathematics and decision sciences by 1984.

This was a fascinating moment in computing history. The IBM personal computer had only been released three years earlier. Apple's iconic "1984" advertisement for the Macintosh was about to air during the Super Bowl. The field of artificial intelligence was in the middle of what researchers now call the "AI winter"—a period of reduced funding and interest following inflated expectations in the 1970s.

But Brynjolfsson saw something others might have missed. After college, he returned to Harvard to teach courses on building expert systems and applications of artificial intelligence. Expert systems were an early approach to AI that tried to encode human expertise into computer programs using if-then rules. If a patient has a fever and a cough and recently traveled to a tropical region, then consider malaria. That kind of logic, but with thousands of rules.

In 1987, he co-founded the Expert Systems subgroup of the Boston Computer Society, which at the time was the largest personal computer user group in the world. The society eventually dissolved in 1996, but it served as an important community for early technology enthusiasts—people who sensed that these machines were going to change everything, even if they couldn't quite articulate how.

The Productivity Paradox

Brynjolfsson went on to earn his Ph.D. in Managerial Economics from the MIT Sloan School of Management in 1991. He would stay at MIT for the next three decades, becoming one of the most influential voices in digital economics.

His early research tackled what economists called the productivity paradox. The term comes from a famous quip by the economist Robert Solow, who won the Nobel Prize for his work on economic growth. In 1987, Solow observed that "you can see the computer age everywhere but in the productivity statistics."

This was genuinely puzzling. Companies were spending enormous sums on computers and software. Surely all that investment should show up as increased output per worker? But the official statistics showed productivity growth actually slowing down during the 1970s and 1980s, precisely when computers were spreading through the economy.

At Solow's urging, Brynjolfsson wrote a comprehensive review of this paradox. His conclusion? The puzzle was at least partly a measurement problem. Computers were generating value that our traditional economic statistics failed to capture.

But measurement wasn't the whole story. Brynjolfsson's research showed that simply buying computers didn't automatically make a company more productive. What mattered was how you used them. Companies that combined information technology with complementary investments—things like training workers, redesigning business processes, and rethinking organizational structures—saw real productivity gains. Those that just plugged in computers without changing anything else often saw little benefit.

The J-Curve

This insight led to another important idea: the productivity J-curve. When companies adopt new technologies, they often experience a period of stagnant or even declining productivity before things improve. Picture the letter J. You start at the top left, dip down, and then curve back up.

Why the dip? Because the intangible investments—the training, the process redesign, the organizational restructuring—take time and resources. Workers have to learn new systems. Managers have to figure out new ways of coordinating. Companies might have to write off old equipment and methods. All of this shows up as cost without immediately showing up as output.

Then, if the implementation succeeds, productivity takes off. The curve shoots up the right side of the J.

This framework has proven remarkably useful for thinking about technology adoption. It explains why revolutionary technologies often disappoint at first. The telegraph, the electric motor, the personal computer—each took decades to fully transform the economy, because each required complementary changes in how humans organized themselves.

Measuring What Matters

Brynjolfsson has long argued that our conventional economic statistics miss much of the value created by digital technologies. Consider Wikipedia, the free online encyclopedia. In purely economic terms, Wikipedia contributes essentially nothing to the Gross Domestic Product, or GDP—that standard measure of economic output that gets reported in the news. Nobody pays to use Wikipedia, so it generates no revenue to count.

But does anyone seriously believe Wikipedia has zero value? It's one of the most visited websites on Earth. Students, researchers, curious minds of all kinds consult it millions of times every day. If Wikipedia suddenly disappeared, people would pay substantial sums for alternatives.

Working with colleagues including Avinash Collis and Felix Eggers, Brynjolfsson developed new methods to measure this kind of value. They used what they call "massive online choice experiments"—essentially asking large numbers of people how much they would need to be paid to give up various digital services.

The results were striking. Users valued Facebook at roughly $550 per year. Search engines came in around $17,500 per year. Email? About $8,400 annually. These figures don't show up in GDP because users don't actually pay these amounts, but they represent real value—what economists call consumer surplus, the difference between what something is worth to you and what you actually pay for it.

This research led Brynjolfsson and his colleagues to propose a new measure they call GDP-B, where the B stands for "benefits." Unlike traditional GDP, which counts only market transactions, GDP-B tries to capture the consumer surplus from free digital goods. The goal isn't to replace GDP—which remains useful for many purposes—but to complement it with metrics that better reflect how digital technology affects human wellbeing.

The Second Machine Age

In 2014, Brynjolfsson and his frequent collaborator Andrew McAfee published "The Second Machine Age." The book became a touchstone for discussions about technology and the economy, translated into more than twenty languages.

The title refers to the steam engine, which powered the first machine age beginning in the late 1700s. Steam and its successors—internal combustion, electricity—extended human physical capabilities. Suddenly we could move faster, lift heavier loads, manufacture goods at scale. The economic historian Joel Mokyr has called this "the lever of riches."

Brynjolfsson and McAfee argued that digital technology represents a second lever, one that extends human mental capabilities rather than physical ones. Computers can now recognize speech, translate languages, identify objects in images, play games that require strategy, and increasingly, generate text and images that pass for human-created.

One reviewer described the book as "pioneering a fundamentally new economics, one based not on the old reality of scarcity but on a new reality of abundance that we are only just beginning to comprehend."

That's a grand claim, but there's something to it. Traditional economics is built around scarcity. There's only so much oil in the ground, only so many hours in the day, only so many skilled workers. Prices emerge from the tension between limited supply and unlimited wants.

But digital goods have peculiar economics. Once you've created a piece of software, you can copy it infinitely at essentially zero cost. The same goes for digital books, music, movies, and the algorithms that power AI systems. This abundance doesn't eliminate scarcity—you still need hardware, energy, human attention—but it fundamentally changes the economic calculus for many goods.

Race With, Not Against

The question that haunts every discussion of AI and automation is whether machines will replace human workers. Brynjolfsson's answer has been consistent for over a decade: the question is wrong.

In 2011, he and McAfee published "Race Against the Machine," which examined how digital technology was transforming employment. The key insight? Humans shouldn't try to compete with machines at tasks machines do well. They should partner with machines, combining human creativity, judgment, and adaptability with machine speed, precision, and tirelessness.

Two years later, Brynjolfsson gave a TED talk making this argument to a broader audience. He was subsequently labeled a "techno-optimist," though he prefers the term "mindful optimist." The distinction matters to him. Techno-optimism can imply that technology automatically makes things better. Brynjolfsson's view is more nuanced: technology creates possibilities, but human choices determine outcomes.

He ended that TED talk with words he often repeats: "Technology is not destiny. We shape our destiny."

The Turing Trap

In 2022, Brynjolfsson published an article called "The Turing Trap," which crystallized years of thinking about AI and labor. The Turing test, proposed by the mathematician Alan Turing in 1950, asks whether a computer can exhibit intelligent behavior indistinguishable from a human. If you're chatting with something and can't tell whether it's human or machine, the machine passes the test.

Brynjolfsson argued that using the Turing test as a goal for AI development has unintended consequences. If you're trying to build systems that perfectly imitate humans, you're implicitly building systems to replace humans. Every success brings you closer to making human workers obsolete.

But what if we set a different goal? Instead of asking "Can AI do what humans do?", we could ask "Can AI do what humans can't?" This reframes the technology from replacement to augmentation. The goal becomes extending human capabilities rather than replicating them.

The difference shows up in economic outcomes. If AI replaces workers, the benefits flow primarily to whoever owns the AI systems—typically shareholders and executives. If AI augments workers, making them more productive, the benefits can be shared more broadly through higher wages and new job opportunities.

This isn't just theory. Consider a chess player using a computer for assistance. The combination of human intuition and machine calculation can outperform either alone. The same principle applies to doctors using diagnostic AI, lawyers using document review software, or designers using generative AI tools. The human brings judgment, creativity, and the ability to handle novel situations. The machine brings speed, pattern recognition, and the ability to process vast amounts of information.

The Task-Based Approach

How do you actually figure out which jobs AI will affect and how? Brynjolfsson, working with computer scientist Tom Mitchell and economist Daniel Rock, developed what they call the "task-based approach."

The key insight is that jobs consist of bundles of tasks. A radiologist reads images, but also consults with patients, coordinates with other physicians, keeps current on medical literature, and handles administrative duties. AI might excel at image analysis while being useless at patient communication.

This means that AI rarely eliminates entire occupations at once. Instead, it transforms jobs by automating some tasks while leaving others untouched—or even making them more important. The tasks that require distinctly human skills—empathy, creativity, physical dexterity, dealing with novel situations—become more valuable as routine tasks get automated.

Brynjolfsson co-founded a company called Workhelix that applies this framework to help organizations assess their opportunities for using AI. The methodology has also influenced policy discussions. Brynjolfsson co-chaired two committees for the National Academies of Sciences, Engineering, and Medicine—one on "Automation and the US Workforce" in 2017 and one on "Artificial Intelligence and the Future of Work" released in 2024. He has testified before Congress and participated in AI summits at the White House.

The AI Index and Long-Term Thinking

In 2016, Brynjolfsson co-founded the AI Index, an initiative to track the progress of artificial intelligence across multiple dimensions. The project produces annual reports documenting advances in AI capabilities, adoption in industry, research trends, and policy developments.

The same year, he co-authored the original report for the One Hundred Year Study of Artificial Intelligence, a project convened by Stanford to assess AI's impact on society. The study's very name signals a recognition that understanding AI requires thinking on timescales longer than the next product cycle or quarterly earnings report.

In 2020, after thirty years at MIT, Brynjolfsson moved to Stanford. He now holds the Jerry Yang and Akiko Yamazaki Professorship—named after the co-founder of Yahoo and his wife—and directs the Digital Economy Lab at the Stanford Institute for Human-Centered Artificial Intelligence.

His graduate course, "The AI Awakening: Implications for the Economy and Society," has featured guest lectures from some of the most prominent figures in AI development, including Mira Murati, who served as chief technology officer of OpenAI; Jeff Dean, who leads Google's AI research; and Mustafa Suleyman, a co-founder of DeepMind.

Beyond Academia

Brynjolfsson hasn't confined himself to academic pursuits. He has co-founded three companies and holds several patents, including ones for forecasting skills and tasks and for optical storage media. He has served on the boards of two publicly traded companies: Computer Science Corporation and CSK Holdings.

In an earlier life, he even designed video game software called Dragonfire II—a reminder that the people shaping our technological future sometimes got their start tinkering with games and toys.

Perhaps most notably, Brynjolfsson co-founded the MIT Inclusive Innovation Challenge, a global competition designed to encourage entrepreneurs to use technology for creating a more equitable future. The winners have collectively generated over $170 million in revenue, raised over a billion dollars in capital, created more than 7,000 jobs, and served 350 million people.

These numbers point to something important about Brynjolfsson's worldview. He's not content to simply study how technology affects the economy. He wants to shape that relationship—to demonstrate that the benefits of technological progress can be shared widely rather than concentrated among a few.

The Road Ahead

The questions Brynjolfsson has spent his career investigating have never been more urgent. Large language models like GPT-4 and Claude can now write essays, code software, and engage in nuanced conversation. Image generators can create photorealistic pictures from text descriptions. These capabilities seemed like science fiction just a few years ago.

Yet the fundamental tension Brynjolfsson has identified remains unresolved. Will we use these tools to replace human workers, concentrating the gains among a small elite? Or will we use them to augment human capabilities, expanding what people can accomplish and sharing the benefits broadly?

The answer isn't predetermined. Technology creates possibilities, but as Brynjolfsson has reminded audiences countless times, we shape our destiny. The economist who has spent decades measuring productivity, proposing new metrics for wellbeing, and arguing for augmentation over automation will continue making the case that the choice is ours to make.

In a world increasingly anxious about AI, Brynjolfsson offers something valuable: not naive optimism that technology will automatically make things better, nor fatalistic pessimism that humans are doomed to obsolescence. Instead, he offers a framework for thinking clearly about the choices we face and evidence for the possibility that we can choose wisely.

Whether we actually will remains the open question of our age.