How is artificial intelligence reshaping the global economy?
Artificial intelligence (AI) is a general purpose technology that is quickly reshaping economic activity and human behavior. Modern AI is still in its early days (though the term itself and earlier versions date back to the 1950s). Several estimates suggest that AI could create trillions or tens of trillions of dollars in wealth. Given the wide potential applications of the technology, those numbers seem plausible.
But as with most bursts of technological innovation, the benefits and risks of AI are not distributed equally worldwide. In AI as with other technologies, what we have isn’t really a global economy, but a set of partially connected national economies playing out on the global stage.
How does machine learning work?
Computer programs typically are created by programmers who write down a set of instructions that the computer executes. This works well when software can clearly state a set of rules or processes for doing something straightforward—basic mathematical operations, for instance.
But if we want a computer to carry out a task that doesn’t have a set of easy-to-follow instructions, because we don’t know exactly how to do it or how it can be efficiently done—such as recognizing a face or keeping our balance while catching a football—then we need a different technique.
Machine learning uses statistics to train computers to independently extract patterns from massive data sets, develop their own algorithms based on those patterns, and fine-tune the algorithms’ predictive ability by running them against new data sets.
How does machine learning boost economic growth?
Machine learning can create enormous value in a variety of ways, but probably the simplest way is by ramping up productivity. Any real production process—whether a factory that makes cars or a process that determines the likelihood that a biopsied skin cell is malignant—has thousands of steps. Each one of those steps could be done more efficiently. But to figure out how to do so requires considering possible changes to these thousands of steps, and perhaps rearranging them and adding new ones. This ushers in a cascade of nearly infinite possibilities.
Given this complexity, machine learning is a more efficient way to sift through these possibilities because the algorithms can predict whether a change will improve things. In practice, this means machine learning can yield productivity gains that human beings might not have otherwise imagined or discovered.
But those algorithms themselves need to be built, maintained, and improved. That involves a set of processes that we call a “machine learning value chain”. The term “value chain” still calls up images of factories and industrial products, but new technologies, such as AI, have value chains too.
An AI value chain basically includes five steps: data collection, data storage, data preparation and engineering, algorithm training, and algorithm development. That’s not the only way to categorize it, and it’s not the end of the story by any means. But thinking about AI in these terms shows that some countries are in better shape to join the AI competition than others.
How have policymakers approached AI so far?
It wasn’t that long ago that many people in the technology sector were able to convince policymakers that the digital world existed on a plane separate from and independent of governments. The internet of the 1990s, and even through much of the 2000s, was imagined to be a policy-free zone in which the power of technology would simply overwhelm or bypass the power of the state. It’s good that this naive notion is now mostly gone. But that really ups the ante for policymakers, who now have no excuse for simply letting the technology sector carve its own path.
Policymakers thinking about long-term economic growth as well as military and security issues need to determine which parts of the machine learning value chain are most important, and realistically achievable for their country. Policy will be an important determinant of where countries land in the global machine learning economy over the next decade. That has become increasingly clear in the last year or so, as governments develop AI strategies to position themselves on what may be an even more uneven landscape than the one that late twentieth-century globalization produced.
How will the consequences of AI differ from the effects of globalization?
In simple terms, globalization refers to the increasing mobility of goods, money, ideas, and, to a lesser extent, people across national boundaries. This heightened mobility, in turn, has led many global companies to create cross-national production networks. For instance, the technology and business services firm IBM promoted a vision of a globally integrated enterprise, where work moves (as former chief executive officer Sam Palmisano put it) to where it can be done best, essentially without regard for political borders.
Right now, the AI and machine learning economy is not evolving that way. That’s partly a function of the global rise in economic nationalism. But in many ways, this difference is also specific to AI. Given AI’s propensity to boost productivity and growth, policymakers see that the technology is poised to bestow significant and possibly sustained advantage on the companies and countries that lead. This makes countries less keen to share AI-related scientific and business knowledge.
AI also has direct military and intelligence applications—autonomous weapons are just one example. Countries have also become sensitive, though in very different ways, to the uses and abuses of very large data sets collected by firms and governments. They are keenly aware of the value of such data and are leery of letting it fall into others’ hands.
Like with the steam engine or other technologies of the past, the basic science and technologies behind machine learning cannot really be controlled or contained. But pieces of the value chain can be at least partially contained within national borders. It is possible to restrict the export of data, and access to bespoke hardware that runs machine learning systems much more efficiently than less specialized hardware. Governments can encourage or discourage the development of markets for certain kinds of machine learning products that can jumpstart AI-powered economic growth. Restricting the movement of the most talented machine learning researchers and engineers (through visa restrictions, for instance) is feasible. And it’s possible, though not easy or cheap, that some technologies could be brought out of the private sector and back into the public and defense sectors, if governments decide it is necessary for national security reasons.
The bottom line is that no law of nature says that machine learning will diffuse across national borders and be equally available to all countries and companies. There will be leaders and laggards. In the fast-moving technology space, it’s likely that the market leaders will, at least for a time, accelerate ahead of others at an even faster pace, as their initial advantages build on themselves. That has important consequences for global economic inequality, which may start to take a different shape than it has over the last thirty years.
How can policymakers prepare for those new developments in economic inequality?
Globalization has contributed to increased inequality inside many countries. But economic inequality between countries has diminished in meaningful terms. Economist Richard Baldwin coined the term “the Great Convergence” to describe this phenomenon. This convergence was partly driven by Chinese growth and some of its positive effects in other developing nations—all of which were enabled by cross-national production networks and liberalized trade in manufactured goods. The relative increase in wealth among many less developed countries was a geopolitical stabilizer, even as worsening domestic inequality created internal political fractures in many countries.
The machine learning economy looks set to shape a different inequality landscape. First-mover advantages, positive feedback growth loops, and the critical importance of small pockets of highly talented people will combine to create superstar economies. Just as San Francisco and the Bay Area have rocketed ahead of rural California, the countries that lead in machine learning are likely to speed ahead of others, who will find it increasingly hard to catch up.
This looks like very bad news for poor developing countries, who can no longer count on low-cost manufacturing as the first rung of a ladder to faster growth. There might be an equivalent for machine learning, like labeling data sets, but that is extremely low-value-add work that can be done anywhere in the world over the internet, and from which little, if any, learning occurs that would jump-start the development of more sophisticated and higher-value machine learning activities, like algorithm design and product development. Middle-income countries with more capital may find themselves in a similarly challenging situation. How can they use that capital to avoid being left behind?
The age of industrial globalization had a formula for economic development, but machine learning does not yet. Policymakers need to solve that problem as soon as possible—and that includes policymakers in countries that are winning the race. Leaving everyone else behind only looks good until the losers decide to disrupt or attack the leaders rather than try to compete. Given the vast challenges of digital insecurity that could undermine our collective future, that is not a scenario that anyone should want to see come to pass.
Steven Weber is faculty director of the Center for Long-Term Cybersecurity at the University of California, Berkeley’s School of Information. His new book Bloc by Bloc: How to Organize a Global Enterprise for the New Regional Order contains an extended discussion of many of these issues.