This article was originally published in the “Carnegie Endowment for International Peace” by CHARLOTTE STANTON, VIVIEN LUNG, NANCY (HANZHUO) ZHANG, MINORI ITO, STEVE WEBER, KATHERINE CHARLET. Read the Full Article on Carnegie Endowment for International Peace.
INTRODUCTION
Thanks to major improvements in computing power, increasingly sophisticated algorithms, and an unprecedented amount of data, artificial intelligence (AI) has started generating significant economic value. With algorithms that make predictions from large amounts of data, AI contributes, by some estimates, about $2 trillion to today’s global economy. It could add as much as $16 trillion by 2030, making it more than 10 percent of gross world product1.
AI’s outsize contribution to global economic growth has important implications for geopolitics. Around the world, governments are ramping up their investments in AI research and development (R&D), infrastructure, talent, and product development. To date, twenty-four governments have published national AI strategies and their corresponding investments.
So far, China and the United States are outspending everyone else while simultaneously taking steps to protect their investments from foreign competition. In 2017, China passed legislation requiring foreign companies to store data from Chinese customers within China’s borders, effectively hamstringing outsiders from using Chinese data to offer services to non-Chinese parties. For its part, the U.S. Committee on Foreign Investment blocked a Chinese investor from acquiring a leading U.S. producer of semiconductors, which are essential components for computing. While this was officially a national security action, it could also benefit U.S competitiveness by protecting its stake in semiconductor production. 2
Both data and certain classes of semiconductors are core elements of the AI value chain. Given AI’s economic and geopolitical significance, they’re also increasingly being considered strategic assets. The extent to which countries can participate in this value chain will determine how they fare in the emerging global economic order and the stability of the broader international system. Indeed, if the gains from AI are distributed in highly variable ways, extreme divergence in national outcomes could drive widespread instability.
So what does the AI value chain look like? And where in the physical world are the key nodes of value creation and control emerging? This article addresses these questions, introducing the idea of a machine learning value chain and offering insights on the geopolitical implications for countries searching for competitive advantage in the age of AI.
THE MACHINE LEARNING VALUE CHAIN
Machine learning, the science of getting computers to make decisions without being explicitly programmed, is the subfield of AI responsible for the majority of technical advances and economic investment. In recent years, machine learning has led all categories of AI patents (and, in fact, constituted the third-fastest-growing category of all patents granted behind 3D printing and e-cigarettes) and attracted nearly 60 percent of all investment in AI.
A value chain describes the sequence of steps through which companies take raw materials and add value to them, resulting in a finished, commercially viable product. For machine learning, that value chain consists of five stages: data collection, data storage, data preparation, algorithm training, and application development.
- Data collection involves the gathering of raw data from any number of sources.
- Data storage involves amassing raw data in data centers.
- Data preparation involves efforts to clean, convert, format, and label raw data.
- Algorithm training involves configuring an algorithm to make predictions from data.
- Application development converts algorithmic predictions into commercially viable products.
WHAT ARE THE GEOPOLITICAL IMPLICATIONS?
While no two countries look alike in their machine learning investments, most fall into three categories: fast movers, moderate movers, and slow starters. Fast movers, namely China and the United States, are heavily investing across most if not all nodes of the machine learning value chain—effectively ensuring that both economies medal in the so-called race to win AI. Moderate movers, by contrast, are concentrating their investments in particular nodes of the value chain. Germany, Japan, and Taiwan, for instance, are heavily investing in the physical capital required for data storage and algorithm training (like HDCs and supercomputers). Australia and South Korea are investing in the requisite intellectual capital (for example, R&D and STEM graduates).
Slow starters have yet to invest significantly in any stage of the machine learning value chain. Most developing countries are slow starters. Notable exceptions include Brazil, which entered the HDC market by capitalizing on its cheap cost of energy, and Kenya, whose relatively high internet penetration rate (83 percent) enables significant data collection. But slow starters need not be left out: the most immediate opportunities for these countries are in data collection and data labeling. The market for data labeling, and specifically image labeling, has an especially low barrier to entry for developing countries where English is not the first language since classifying images doesn’t require English literacy.
This offers hope that there are ways to drive a more globally inclusive machine learning economy. But dedicated attention is necessary. Governments should study where in the machine learning value chain they may have a comparative advantage. Development agencies can develop tools to assist such analysis, and they can invest in strategies to help slow starters better position themselves to participate in the machine learning value chain. Researchers can study which stage of the machine learning value chain creates the most value in order to determine whether any given country’s best opportunities for investment are in data collection, data storage, or another stage. Such research must account for country-specific factors and the magnitude and relative shelf life of the value created at each stage. For instance, the profitability of operating an HDC directly depends on local energy prices and data localization laws, among other things. Likewise, a hardware component may become obsolete after three years, while a well-trained data scientist may yield value for decades, possibly becoming more valuable over time with the benefit of experience. Unpacking the profitability of each stage for each country will not be easy, but it can help countries enhance their competitive advantage in the age of AI.
Finally, it’s important to note that the geographical concentration of machine learning value among the fast movers and even moderate movers will have first- and second-order impacts on the distribution of wealth and power within and between countries. Concentrating talent and wealth in certain countries will likely exacerbate economic inequality between countries. A similar geographical concentration of talent and wealth in certain cities could also impact land and housing prices, causing demographic shifts. Taken together, these first- and second-order impacts suggest increasing inequality between and within countries—a different trend than what took place over the last quarter century, when globalized industrial manufacturing increased inequality within countries but decreased inequality across countries. Policymakers must prepare now for the geopolitical consequences of countries’ varied capabilities and investments in the machine learning value chain.
Read the Full Article on Carnegie Endowment for International Peace.
Authors
- Jacques Bughin, Jeongmin Seong, James Manyika, Michael Chui, and Raoul Joshi, Notes From the AI Fronter: Modeling the Impact of AI on the World Economy (New York: McKinsey Global Institute, 2018), 13; and “GDP Long-Term Forecast,” Organization for Economic Cooperation and Development, 2018, https://data.oecd.org/gdp/gdp-long-term-forecast.htm.
- Tim Hwang, “Computational Power and the Social Impact of Artificial Intelligence,” SSRN Electronic Journal (March 2018): 27; and Michael Brown and Pavneet Singh, “China’s Technology Transfer Strategy: How Chinese Investments in Emerging Technology Enable A Strategic Competitor to Access the Crown Jewels of U.S. Innovation,” Defense Innovation Unit Experimental (DIUx), January 2018, 8–15.