AI Will Take Longer To Change The World Than Forecasters Say

Milton Ezrati

Chief Economist

Artificial intelligence (AI) has become the stuff of dreams. For some, these dreams look more like nightmares in which large sections of the population become unemployable. As my recent discussion in Forbes showed, both history and economic logic argue that such fears miss likely realities. Some, of course, will suffer job loss, but ultimately, these technological innovations, just like all those before them, will create more jobs than they destroy. Certainly, efforts to meet an unlikely challenge with a government-provided universal basic income (UBI) will do little except perhaps reduce the size of the workforce and allow a few technology oligarchs to feel “woke.”

AI dreams come with a very different character as well. These see a near future of robotic servants lifting the cares of everyday life. They see a limitless supply of affordable goods and services. Some describe flying driverless cars, others trains and planes that move at unheard of speeds. AI will not just take the pain out of living, it will take the travail out of travel. In these dreams, AI will make everyone smarter, too, by providing us with clever handmaidens to guide our decisions. And this is just a small sample of the prospects envisioned. Perhaps someday all these wonders will appear, but again, history and economic logic argue that such a day will take much longer to arrive than many anticipate.

Each past wave of innovation has also fostered such dream-like predictions (along with their dystopian cousins). None of the dreams have come true. The dystopian ones never and the wonderful ones never in the way they are initially cast or at the speed described by forecasters. The benefits take much longer to arrive than most typically envision. Before benefits become widespread, economies and societies must experiment with innovations. They must find applications to daily life and business, government and defense. When enough time has past for this process to gain momentum, and it always takes time, the ultimate change typically surpasses the initial, dreamy forecasts. There is every reason to expect a repeat of this pattern with AI, in terms of its general contours if not its specifics.

The computer revolution of the past 50-some years provides a near perfect illustration. Despite early fears of widespread unemployment and more pleasant dreams of high-speed business, neither happened, certainly not on the schedule originally forecast. It took 20 years from the first computers built during the Second World War to develop a commercially viable model, Univac, in the early 1960s. Even then, it took time before people and business recognized the potential. It is noteworthy in this regard that the 1975 Economic Report of the President (authored incidentally by Alan Greenspan) never used the word computer when discussing productivity. Indeed, the word does not appear once in the document.

Even as large firms and governments began using mainframes, it took another 20 years for the personal computer (PC) to develop so that computing power could become a part of small business and the everyday lives of individuals and office workers. It took another 20 years before the PC and the Internet together allowed e-commerce and e-communication that have so transformed business and daily life. That is more than half a century from the first innovation. Though Moore’s Law, which states that computing power doubles every two years, helped the process of transformation, it was less of a factor than the time it took for society and businesses to experiment with applications and then adjust to them.

If anything, the exciting revolutions in nuclear power and rocket science have taken even longer to have their transforming effects. It took 10-15 years from the military application of nuclear technology to the construction of viable reactors. During that time, along with the dystopian scenarios, forecasts of atomic-powered planes and cars littered popular discussion as well as projections of floating cities that would free humankind from the confines of geography. People are still waiting, even for a reactor with which the general public feels comfortable.

It is the same with space travel. When John F. Kennedy issued his challenge to the National Aeronautics and Space Administration (NASA) to land on the moon before the end of the 1960s, rocket science had already advanced steadily for some 25-30 years. When the space program began to show results, popular forecasts again took off. The world has seen tremendous advances since the 1969 moon landing, but space technology has yet to transform people’s lives as expected. Its biggest influence, satellites that have enabled rapid communication, is far from the dreams of the early years. Hilton has yet to open a hotel on the moon as per the late 1960s film “2001 Space Odyssey,” and despite the dreams of Jeff Bezos and Elon Musk, very few people travel into space.

It is not that practical, safe nuclear technology or space travel will not eventually have a transformational effect. They will. But it has already taken a lot longer than people thought when the innovations first broke on the public’s consciousness. The automobile has already transformed economics and society, but its changes, too, took longer than expected. It took more than 20 years or the first innovation to bring a practical design, the Model T, and then with 40-50 years before the economy and society adjusted and the car spawned interstate highways and the shape of the United States today. The highway system more or less accurately envisioned by General Motors at the 1939 World’s Fair did not arrive within a decade, as predicted. It did not arrive until the 1970s, in fact.

Energy provides a kind of negative illustration of typical forecasting errors. Thomas Malthus saw the imminent exhaustion of England’s coal resources late in the eighteenth century and forecast economic hardship as a result. He never considered longer-term technologies that would find ways to bring more coal to market and use oil as a substitute. As they did, society avoided the doom he forecast and then eventually transformed itself much more thoroughly than anyone expected early in the process. When in the 1970s the price of oil jumped 400%, forecasters saw “doomsday” for the U.S. economy unless, in the words of a U.S. News and World Report article, “a massive effort to solve the problem is launched immediately.” Nor were the 1970s the first time. Indeed, the standing joke among oil analysts is how forecasters have called for the world to run out of oil in ten years for at least the last 100 years. It has not. Now an excess of hydrocarbons has prompted today’s forecasters to claim immediate threats to the planet’s climate.

One of the problems with forecasting, especially in the period immediately after an innovation appears, lies in what one might describe as crossovers. Forecasters look at the strides in one area of technology and apply them to another where engineering issues are sometimes entirely different. Miniaturization offers one example. It had such power in electronics that forecasters saw similar patterns everywhere else, in automobile engines, for instance, or alternative energy sources. If miniaturization enabled computer engineers to take the power of a room-sized computer of the 1970s, multiply it many times and put it a small mobile phone, then cars would soon have engines no bigger than a man’s fist, allowing all sorts of other advantages. Wind turbines would soon fit neatly onto each homeowner’s roof. This would certainly extend what happened with electronics. But it seems that very small engines have trouble powering cars at speed or even moving them. They certainly will not arrive as fast as miniaturization did in electronics, and even there, things took longer than memories today suggest. Meanwhile the advances in wind turbines indicate that bigger is better. Instead of fitting on the roof or in the backyard, new designs these days stand taller than the Washington Monument.

Someday, AI technologies will almost surely change the economy and society. Driverless cars may indeed permit 100-mile-per-hour beltways and high-speed trains will run at remarkable speed through vacuum-sealed tubes, a recurring idea recently resurrected by Elon Musk. Many of the dreams connected to AI will come true, or something like them will. But none of this will occur until society and the economy have had time to make prototypes practical and then go through the long process to experiment with applications and adjust practice to them. Immediate forecasts of change entirely miss these time consuming intervals. Then, years after forecasts of radical change have failed to arrive on schedule, the economy and society will make their adjustments and the innovation will have much more far-reaching effects than even the dreamers predicted.

Originally published on