Toh understand that The impact artificial intelligence can have on the economy, think of the tractor. Historians disagree as to who invented the humble machine. Some say it was Richard Trevithick, a British engineer from 1812. Others argue that John Froelich, working in South Dakota in the early 1890s, has a better claim. Still others point out that few people used the word “tractor” until the early 20th century. However, everyone agrees that it took a long time for the tractor to make any tracks. In 1920, only 4% of American farms had one. Even in the 1950s, less than half owned tractors.
speculation about the consequences hey– for jobs, productivity and quality of life – is in full swing. The technology is impressive. And yet heyThe economic impact will be muted unless millions of companies outside of Silicon Valley adopt it. That would mean far more than just using the odd chatbot. Instead, a comprehensive reorganization of the companies and their internal data would be required. “The diffusion of technological improvements,” argues Nancy Stokey of the University of Chicago, “is arguably as critical to long-term growth as innovation.”
Japan and France show how important dissemination is. Japan is unusually innovative, producing more patents per person per year than any other country except South Korea. Japanese researchers can take ownership of the invention of the invention qr code, the lithium-ion battery and 3D Press. But the country does poorly at diffusing new technologies into its economy. Tokyo is far more productive than the rest of the country. Cash still dominates. In the late 2010s, only 47% of large companies were using computers to manage their supply chains, compared to 95% in New Zealand. According to our analysis, Japan is around 40% poorer than would be expected based on its innovative strength.
France is the opposite. Although the country’s track record for innovation is average, it is excellent at spreading knowledge throughout the economy. In the 18th century, French spies stole technical secrets from the British Navy. At the beginning of the 20th century, Louis Renault visited Henry Ford in America and learned the secrets of the automobile industry. More recently, former hey Experts from Meta and Google founded Mistral hey in Paris. France also tends to do well when it comes to spreading new technologies from the capital to the periphery. Today, the productivity gap between a top and a mid-tier company in France is less than half that in the UK.
In the 19th and 20th centuries, businesses around the world became more “French” and new technologies spread faster and faster. Diego Comin and Martí Mestieri, two economists, find evidence that “cross-country differences in the delay in adoption have narrowed over the past 200 years.” The stream spread through the economy faster than tractors. It only took a few decades for personal computing in the office to break the 50% acceptance threshold. The internet spread even faster. Overall, the spread of technology helped fuel productivity growth in the 20th century.
However, since the mid-2000s, the world has been becoming Japanese. It’s true that consumers are adopting technology faster than ever. According to one estimate, the number of users of TikTok, a social media app, went from zero to 100 million in one year. Chatgpt itself was the fastest-growing web app in history until Threads, a rival to Twitter, launched this month. But companies are becoming increasingly cautious. All sorts of stunning innovations have hit the market over the past two decades. Still, according to the latest official estimates, only 1.6% of American companies used machine learning in 2020. In America’s manufacturing sector, only 6.7% of companies use 3D Press. Only 25% of business operations take place in the cloud, a number that hasn’t changed in half a decade.
Horror stories abound. In 2017, a third of Japanese regional banks still used it cobol, a programming language invented a decade before man landed on the moon. Last year Britain imported more than £20 million ($24 million) worth of floppy discs, MiniDiscs and cassettes. A fifth of the rich world’s businesses don’t even have a website. Governments are often the worst offenders – they insist on paper forms, for example. We estimate that bureaucracies around the world spend $6 billion a year on paper and printing, which is about as much in real terms as it was in the mid-1990s.
The best and the rest
The result is a two-tier economy. Companies that embrace technology withdraw from the competition. In 2010, the average worker in Britain’s most productive businesses produced goods and services worth £98,000 (in today’s monetary terms), which had risen to £108,500 by 2019. The worst-performing companies did not increase. In Canada, the productivity growth of frontier firms was about 40% faster than that of non-frontier firms in the 1990s. From 2000 to 2015 it was three times as high. A book by Tim Koller of McKinsey, a consulting firm, and colleagues finds that when companies were ranked by their return on invested capital, the 75th percentile had a return that was 20 percentage points higher than the median in 2017 – twice as much as in 2000. Some companies see huge gains from purchasing new technology; many see nothing at all.
While the economics may sound abstract, the real world implications are frighteningly familiar. People who can no longer use old technologies suffer, as do their salaries. In the UK, average wages in the least productive 10% of firms have fallen slightly since the 1990s – although average wages in the best firms have risen sharply. According to Jan De Loecker of ku Leuven and colleagues: “Most of the growth in inequality between workers is due to increasing average wage differences between companies.” Then what went wrong?
Three possibilities explain the lower penetration: the nature of the new technology, sluggish competition and increasing regulation. Robert Gordon of Northwestern University has argued that the “great inventions” of the 19th and 20th centuries had a far greater impact on productivity than more recent ones. The problem is that as technology advances, so does adoption, as companies have fewer incentives and face less competitive pressure to modernize. Electricity powered machines with light and energy. Cloud computing, on the other hand, is only needed for the most intensive operations. Recent innovations like machine learning may be more difficult to leverage and require more skilled workers and better management.
In the early decades of the 21st century, business momentum in the rich world declined. population aged. Fewer new companies were founded. Employees changed companies less frequently. All of this reduced dissemination as workers spread technology and business practices as they make their way through the economy.
In industries that are government-run or heavily managed, technological change occurs slowly. As Jeffrey Ding of George Washington University notes, innovation in the centrally planned Soviet Union was world-breaking—think Sputnik—but adoption was nonexistent. The lack of competitive pressure reduced the incentives for improvement. Politicians often have public goals, such as maximizing employment, that are incompatible with efficiency. Highly regulated industries now make up a large part of Western economies: Such sectors, including construction, education, healthcare and utilities, make up a quarter of the American economy GDP.
Could hey out of the ordinary and spreading faster in the economy than other newer technologies? Possibly. It’s easy for almost any organization to come up with a use case. No more administration! A tool to file my taxes! Covid-19 may have given some dynamism to Western economies as well. New firms are being formed at the fastest rate in a decade, and workers are switching jobs more frequently. George Mason University’s Tyler Cowen adds that weaker companies may have a special incentive to acquire heybecause they have more to gain.
hey can also be built into existing tools. Many programmers – maybe most – are already using it hey daily due to integration with everyday coding tools through Github’s CoPilot. Word processors, including Microsoft Word and Google Docs, will soon be launching dozens of them hey Characteristics.
No dinner party
On the other hand, the biggest advantages are new forms of hey will come as companies completely reorganize around the new technology; through customization hey Models for internal data, for example. This will require time, money and, above all, competitive pressure. Collecting data is tedious and running the best models is horribly expensive – a single complex query in the latest version of Chatgpt may cost $1-2. If you run 20 in an hour, you’ve exceeded the average American hourly wage.
Those costs will come down, but it could be years before the technology is inexpensive enough for mass deployment. Bosses concerned about privacy and security tell it regularly The economist that they are unwilling to send their data to modify models living elsewhere. Small business surveys are not encouraging. One from GoDaddy, a web hosting company, suggests that around 40% of people in America aren’t interested hey Tool. The technology is undoubtedly revolutionary. But are companies ready for a revolution? ■
