Few would disagree that technological change is a major ingredient of long-term economic growth, or that it is beset by a high degree of uncertainty. Understanding the nature of this uncertainty and the obstacles to surmounting it is not a trivial matter. It goes to the heart of how new technologies are devised, how rapidly and far they spread, and how they affect economic performance.

The deep uncertainty associated with innovation makes it hardly surprising that innovating firms have historically experienced high failure rates. Indeed, the vast majority of attempts at innovation fail. But this is only part of the story. A more intriguing field of enquiry might be the apparently widespread inability to anticipate the future impact of successful innovations, even after their technical feasibility has been established.

Uncertainty has a number of peculiar properties that shape the innovation process and hence the way in which technological change exercises its effects on the economy. In considering what has determined the trajectory of new technologies, I propose to focus on those that have made a powerful impact. A study that included unsuccessful as well as successful innovations might yield insights of a very different nature.

After a new technological capability has been established, new uncertainties begin to assert themselves

It is easy to assume that uncertainties disappear after the first commercial introduction of a new technology. Indeed, by this point some uncertainties will have faded. However, after a new technological capability has been established, the questions change, and new uncertainties, especially those of an economic nature, begin to assert themselves.

Historical perspectives

Consider the laser, one of the most powerful and versatile technological advances this century. In the 30 years since its invention, its range of uses has been breathtaking. Lasers allow the reproduction of music in compact disc players, and of text via laser printers. They are widely used for precision cutting in the textile, metallurgy, and composite materials industries. The laser has become the instrument of choice in many surgical procedures, including eye, gynecological, and gall bladder surgery.

Perhaps the most profound impact of the laser has been in telecommunications where, in combination with fiber optics, it is revolutionizing transmission. In 1966, the best transatlantic telephone cable could carry only 138 conversations simultaneously. The first fiber-optic cable, installed in 1988, could carry 40,000. Those installed in the early 1990s can carry nearly 1.5 million.¹ Yet despite what turned out to be a striking record of success, the patent lawyers at Bell Labs were initially unwilling to apply for a patent for the laser, believing it could have no relevance to the telephone industry.

Many other case histories illustrate what now seems a remarkable inability to foresee the uses to which new technologies would soon be put. The inventor of the radio, Marconi, thought it would mainly be used between two points where communication by wire was impossible, as in ship-to-ship or ship-to-shore communication. He envisaged the users of his invention as steamship companies, newspapers, and navies needing to transmit private messages over long distances. The idea of communicating to a large audience of listeners rather than to a single point seems never to have occurred to the pioneers of radio.

This failure of social imagination was widespread. One man, later to become a leader of the broadcasting industry, announced that it was hard to see what uses public broadcasting could serve. His sole suggestion was the transmission of Sunday sermons—the only occasion where one man regularly addressed a mass public.² When it became feasible in the second decade of this century, the wireless telephone was viewed in the same way as the radio: namely, as an extension of the existing wire system enabling remote and inaccessible places to be reached. In 1949, the computer was thought useful only for carrying out rapid calculations in certain scientific and data processing contexts. Even Thomas Watson, Sr, then the president of IBM, rejected the idea that the computer might have a much larger market. The prevailing view until the fifties was that world demand could be satisfied by just a handful of computers.

News of the invention of the transistor was nowhere to be seen on the front page of the New York Times

In December 1947, news of the invention of the transistor was nowhere to be seen on the front page of the New York Times. Instead, it figured in a small item buried deep in the inside pages. A regular weekly column, "News of Radio," suggested that this new device might be employed to develop better hearing aids for the deaf.

This catalog of failures to anticipate future uses for new technologies could be expanded almost without limit. We could, if we liked, amuse ourselves indefinitely at the inability of earlier generations to see the obvious. But that would be a mistake, given that our own ability to overcome the uncertainties associated with new technologies is unlikely to improve dramatically.

The nature of the problem

Much of the difficulty, I would suggest, derives from the fact that new technologies typically come into the world in a primitive condition. Their future uses will depend on an extended process of improvement that vastly expands their practical applications. Thus Thomas Watson, Sr was not so far off the mark if we bear in mind the state of the computer immediately after the Second World War. The first electronic digital computer, the ENIAC, contained no fewer than 18,000 vacuum tubes, was notoriously unreliable, measured more than 100 feet long, and filled a huge room.

This particular failure in prediction was an inability to anticipate the demand for computers after they had been made very much smaller, cheaper, and more reliable, and after their performance, and especially their calculating speed, had been improved by many orders of magnitude. In other words, it was an inability to foresee the trajectory of future improvements and the economic consequences of those improvements.

Around 80 percent of spending goes on improving products that already exist, rather than inventing new ones

The history of commercial aviation, like that of many other innovations, could be expressed in similar terms. The introduction of the jet engine was marked by a failure even among eminent scientists to anticipate the importance of future improve-ments. In 1940, a committee of the National Academy of Sciences was formed to assess the value of developing a gas turbine for aircraft. It concluded that such a thing would be impractical because it would have to weigh 15 pounds for each horsepower delivered, compared to just over one pound with existing internal combustion engines. Yet within a year, the British were operating a gas turbine that weighed a mere two-fifths of a pound per horsepower.³ Here we should note that most R&D expenditure is devoted to product improvement. Around 80 percent of spending goes on improving products that already exist, rather than inventing new ones. Instead of being committed to the search for breakthrough innovations, R&D aims to improve on the performance of technologies that have been inherited from the past.

On reflection, this is not surprising. The telephone has been around for more than a hundred years, but only recently has its performance been enhanced by facsimile transmission, electronic mail, voicemail, data transfer, on-line services, conference calls, and freefone numbers. The automobile and the airplane are both more than 90 years old; the camera dates back 150 years; and the Fourdrinier machine, the mainstay of today’s papermaking industry, was patented during the Napoleonic Wars. The improvement process clearly deserves much more attention.

The role of uncertainty in technological change goes far beyond the issue of technological feasibility

As history suggests, the role of uncertainty in technological change goes far beyond the issue of technological feasibility. Indeed, the uncertainty surrounding the eventual uses of the laser or the computer might be better described as ignorance.

Dimensions of uncertainty

Why is it so difficult to foresee the impact of even technically feasible inventions? Many have emphasized the question: "Will it work?" Though this is clearly a major source of uncertainty, fixating on it has served to divert attention from other important factors:

Hidden usefulness

New technologies come into the world not only in a primitive state, but also with characteristics whose usefulness cannot be immediately appreciated. Identifying uses for new technologies is inherently difficult. It took many decades to explore applications for electricity after Faraday discovered the principles of electromagnetic induction in 1831. Uses for the laser, as we have seen, are still expanding three decades after its invention.

Neither electricity nor the laser represented an obvious substitute for anything that already existed. Neither had a clearly defined antecedent. Each was a new discovery emerging from pure scientific research.

Learning how to translate a newly developed visualization technology into a clinically useful capability can take a long time

In medical diagnostics, learning how to translate a newly developed visualization technology into a clinically useful capability can take a long time. This has been true of computerized axial tomography (CAT) scanners, magnetic resonance imaging (MRI), and, most recently, echo cardiography. Extensive additional research may be needed before it is possible to make a reliable, clinically helpful diagnostic interpretation of what has been visualized. Positron emission tomography (PET) is currently at precisely this stage.

Unlike CAT and MRI, which are valuable for anatomical observation, PET scanners provide quantitative analysis of certain physiological functions. They can supply information on, for example, the effectiveness of drug therapy in the treatment of brain tumors. But the application of PET in such fields as neurology, cardiology, and oncology has been limited by the continuing difficulty of translating measurements of physiological functions into meaningful clinical interpretations.

The inherent complexity of the human body has made teasing out causal relationships extremely hard

Medical innovation involves some special difficulties. The inherent complexity of the human body—and the heterogeneity of human bodies—have made teasing out causal relationships extremely hard. Take aspirin, probably the world’s most widely used drug. Though it has been taken for almost a century, its efficacy in reducing the incidence of heart attacks by virtue of its blood-thinning properties was established only recently.

Although the discovery of harmful side-effects has received much more public attention, unexpected and beneficial new uses for old medications frequently emerge. Adrenergic beta-blocking drugs were originally prescribed to treat arrythmia and angina. Today, they are used in the treatment of more than twenty different conditions—including gastrointestinal bleeding, hypertension, and alcoholism—thanks to new applications uncovered after these drugs had been introduced into cardiology. Similar stories can be told about AZT (currently employed in the treatment of AIDS), oral contraceptives, RU-486, streptokinase, alpha interferon, and Prozac.

Complementary inventions

The impact of an innovation depends on improvements not only in the invention itself, but also in complementary inventions. The laser was of no use in telecommunications without fiber optics. Today, the combined potential of these two technologies is transforming the entire industry. Optical fiber did in fact exist in a primitive state when the first lasers were developed in the early 1960s, though not in any form that could accommodate the requirements of telephone transmission.

As is often the case, it took several years for the benefits of fiber-optics technology to become apparent

As is often the case, it took several years for the benefits of fiber-optics technology to become apparent: the lack of electromagnetic interference, the conservation of heat and electricity, and the enormous expansion in bandwidth owing to the fact that the light spectrum is approximately a thousand times wider than the radio spectrum.

My general point is that the impact of invention A will often depend upon invention B—which may not yet exist. Put a different way, inventions will frequently give rise to a search for complementary inventions. An important impact of invention A is to increase the demand for invention B.

After the introduction of the dynamo in the early 1880s, the falling price of electricity sparked off a search for technologies that could exploit this form of energy. However, the time frame over which complementary innovations could be developed turned out to vary considerably. An electrochemical industry employing electrolytic techniques emerged almost immediately, but a much longer period elapsed before the launch of the electric motor.

Similarly, the fact that transistors had not yet been incorporated into the computers of the day was partly responsible for the early predictions of a modest future for this new technology. The introduction of the transistor, and later the integrated circuit, transformed the industry. Indeed, in one of the most remarkable technological achievements of this century, the integrated circuit itself eventually became a computer with the advent of the microprocessor in 1970.

Long gestations

Major new technologies take many years to replace an established technology. This delay is partly to do with the need to develop numerous components of a larger technological system. Restructuring a factory to use electricity instead of steam or water power often meant a complete redesign. Among other things, electric power represented a revolution in the principles of factory organization, allowing machinery layouts to be much more flexible.

Learning how best to exploit a versatile new power source with wholly different methods of transmission involved decades of experimentation and learning. Indeed, major technological innovations usually entail profound organizational change.

At the same time, firms with huge investments tied up in manufacturing plants that still had long productive lives ahead of them were naturally reluctant to discard such facilities. Hence, those that adopted electricity between 1900 and 1920 tended to be new industries setting up produc-

tion facilities for the first time. In older industries, the introduction of electric power had to await the depreciation of existing plants.

In general, then, a radical new technology like electricity must undergo a long period of gestation before the opportunities it embodies are properly understood and can be thoroughly exploited. In 1910, only 25 percent of US factories used electric power. Twenty years later, the figure had risen to 75 percent.

If we date the origin of the modern computer to the invention of the microprocessor in 1970, we are still only a quarter of a century into the computer age. It took some 40 years or so for electric power to assume a dominant role in manufacturing. Here, then, is cause for optimism: the greatest economic benefits of the computer may still lie before us.

Unknown systems

Thinking about new technologies is handicapped by the tendency to conceive them in terms of old technologies

Major innovations often constitute entirely new technological systems. To conceptualize an unknown system is extremely difficult. As a result, our thinking about new technologies is likely to be handicapped by the tendency to conceive them in terms of the old technologies which they will eventually replace.

Time and again, people view a new technology as a mere supplement that will resolve limitations inherent in an existing technology. In the 1830s and 1840s, railroads were thought of as feeders into the canal system, useful where the terrain was unsuitable for canals. Similarly, the telephone was originally conceived as a business instrument like the telegraph, to be used for exchanging specific messages such as the terms of a contract.

It is characteristic of a system that improvements in performance in one part have only limited impact without simultaneous improvements in other parts. In this sense, technological systems may be said to comprise clusters of complementary inventions. Improvements in power generation, for example, can make only a slight difference to the cost of electricity until improvements are made in the transmission network to bring down the cost of transporting electricity over long distances. This need for further innovation in complementary activities helps explain why even apparently spectacular breakthroughs usually have only a slowly rising productivity curve flowing from them.

Within technological systems, therefore, major improvements in productivity are seldom produced by single innovations, no matter how important they seem to be. But the cumulative effect of multiple improvements within a technological system may ultimately be immense.

Unexpected applications

Historically, one of the reasons why predicting the uses of a new technology is so difficult is that many inventions originate in attempts to solve specific problems. Once a solution has been found, it often turns out to have applications in totally unexpected contexts. Serendipity plays a large part in the life history of inventions.

Take the steam engine, invented in the eighteenth century to pump water out of flooded mines. A succession of later improvements made it a feasible source of power for textile factories, iron mills, and an expanding array of industrial facilities. In the early nineteenth century, steam power was adopted more widely in railroads and steamships. Later that century, it was used to produce a new kind of power, electricity, which in turn satisfied innumerable final uses to which steam power itself did not apply.

Finally, the steam turbine displaced the steam engine in electric power generation, and the qualities associated with electricity—ease of transmission over long distances, capacity for making power available in "fractionalized" units, and much greater flexibility of electric-powered equipment—sounded the death-knell of the steam engine.

Major innovations, once established, have the effect of inducing further innovations across a wide front

As this suggests, major innovations, once established, have the effect of inducing further innovations across a wide front. Indeed, being able to do this amounts to a defining quality of a major innovation, and helps distinguish technological advances that are merely novel from those that have the potential to make a genuine impact. The nature of the eventual impact, however, remains difficult to predict, since it will depend on the size and direction of subsequent complementary innovations.

Since innovations often arise as solutions to specific problems in particular industries, their flow to applications in different settings is bound to be highly uncertain. In some cases, an innovation may have multiple points of impact on another industry.

Consider the role of the computer in air transport. Changes in the performance of this industry have been influenced at least as much by new applications of the computer as by R&D spending:

• Supercomputers now carry out a good deal of aerodynamic research, including much that was formerly performed in wind tunnels.
• Computers have helped to reduce costs in the design of specic aircraft components. They played an important role in the wing design of the Boeing 747, 757, and 767, and the Airbus 310.
• Computers are responsible for much of the activity that takes place in the cockpit—including, of course, the automatic pilot.
• Together with weather satellites, which routinely monitor the movement of high-altitude jet streams, computers are widely used to determine optimal flight paths. The resulting fuel saving for the world’s commercial airlines probably exceeds $1 billion per year.
• Computers are at the core of the current global ticketing and seating reservation system.
• Computer simulation has become the preferred method of instruction in teaching novices how to fly.
• Along with radar, the computer is central to the operation of the air traffic control system.

As this example illustrates, R&D spending tends to be concentrated in a small number of industries. Each industry should be regarded as the locus of research activity that generates new technologies which may be diffused throughout the entire economy. Historically, a few industries have played this role in especially crucial ways, such as with the development of steam engines, electricity, computers, transistors, machine tools, and so on.

This brings us back to the notion that a major—or breakthrough—innovation may be defined as one that establishes a new framework for the working out of incremental innovations. Incremental innovations are the natural complements of breakthrough innovations. In turn, breakthrough innovations have often provided the basis for the emergence of entirely new industries.

Unmet needs

The ultimate impact of a new technological capability is not merely a matter of technical feasibility or improved performance. It also has to do with identifying specific categories of human need and catering to them in novel, cost-effective ways. All innovations need to pass an economic test, as well as a technological one. Concorde might be a spectacular success in terms of its flight performance, but it has proven a financial disaster, costing British and French taxpayers billions of dollars.

What is at stake here is not just technical expertise, but a leap of the imagination. Understanding the scientific basis for wireless communication, as Marconi did, was no help in anticipating how the radio might be used to enrich human experience. In fact, it was an uneducated Russian immigrant, David Sarnoff, who envisaged how the new technology could be employed to transmit news, music, and other forms of entertainment and information. In short, he appreciated the commercial possibilities of radio, and his vision eventually prevailed when he led RCA after the First World War.

Social change and economic impact are not things that can be extrapolated out of a piece of hardware

Social change and economic impact are not things that can be extrapolated out of a piece of hardware. New technologies are unrealized potentials—building blocks whose eventual impact will depend on what is designed and constructed with them. The shape they ultimately take will be determined by our ability to visualize how they might be applied in new contexts.

Sony’s development of the Walkman is a brilliant example of how existing technological capabilities can be recombined to create an entirely new product. Batteries, magnetic tape, and earphones had all been around for some time. What was new was the idea of providing entertainment in unexpected settings, such as while people were out jogging. Admittedly, the components did need to be reengineered, but the real breakthrough was Akio Morita’s identification of a market opportunity that had previously been overlooked.

In the history of the video cassette recorder, the American pioneers, RCA and Ampex, gave up long before a usable product had been developed. Matsushita and Sony, by contrast, went on to make thousands of small improvements in design and manufacture. The initial concept of the VCR had been of a capital good for use by television stations. Progress came with the realization that there might be a mass domestic market for the product if its performance, and especially its storage capacity, could be enhanced.

The crucial difference between the Americans and the Japanese seems to have been the latters’ confidence that they could achieve the necessary cost reductions and performance improvements. The rapid transformation of the VCR into one of Japan’s largest export products was thus an achievement of both imagination and engineering capability.

Dismissed as a hacker’s toy, the personal computer was thought to present no threat to the primacy of mainframes

The blinkered view once held by American firms of the VCR’s potential bears comparison with the disdain of mainframe computer makers toward the personal computer when it began emerging about fifteen years ago. Dismissed as a hacker’s toy, it was thought neither to have a future in the business world nor to present any threat to the primacy of mainframes.

Reviving old technologies—or killing them off?

So far, we have considered barriers to the exploitation of new technologies. However, in highly competitive societies where there are strong incentives to innovate, these incentives apply to improving old technologies as well as to inventing new ones. In fact, innovations often seem to provoke vigorous and imaginative responses from firms confronted with substitutes for their own products. The competitive pressure exerted by a new technology tends to lead to an accelerated improvement in the old technology.

Wooden sailing ships enjoyed some of their biggest advances between 1850 and 1880—just after the introduction of the iron hulls and compound steam engines that were to displace them by the beginning of the next century. There were radical changes in hull design to accommodate more cargo and increase speed, and labor-saving equipment was introduced that cut crew requirements by two-thirds. In the same way, gas lamps for interior lighting were most dramatically improved shortly after the advent of the incandescent electric light bulb.

In telecommunications, postwar research has not only led to the development of productive new technologies, but also increased the capacity of existing transmission systems. Pairs of wires, coaxial cables, microwaves, satellites, and fiber optics have all benefited from later improvements in capacity, often achieved with only minor modifications to these technologies. Some improvements have produced order-of-magnitude gains that have effectively postponed the introduction of new generations of transmission technology. Time-division multiplexing, for instance, now allows a pair of wires to carry 24 voice channels instead of just one.

In fiber optics, capacities of almost 1,000 megabytes per second are predicted for the near future

The same pattern can be discerned in fiber optics. When AT&T began field trials in the mid-1970s, information was transmitted at 45 megabytes per second. By the early 1990s, the standard for new fiber cables had reached 565 megabytes per second. Capacities of almost 1,000 megabytes per second are predicted for the near future.

As we have seen, the introduction of an innovation often has to await the availability of complementary innovations; meanwhile, established technologies may achieve renewed competitive vigor through continual improvement. But this is not always the case. Innovations sometimes turn out to be substitutes for—rather than complements to—existing technologies. Such innovations will cut short the life expectancies of technologies that once seemed to possess rosy futures.

The prospects for communication satellites declined unexpectedly in the 1980s on the introduction of fiber optics, with their huge and reliable expansion of channel capacity. In the same way, fiber optics, whose first major application was in medical diagnostics in the early 1960s, may be approaching the beginning of the end of its useful life. Fiber-optic endoscopes made it possible to use much less invasive techniques in visualizing the gastrointestinal tract. Recently, however, new sensors from the realm of electronics, charged couple devices, have begun to produce images of a resolution and detail that fiber-optic devices cannot match.

The CAT scanner is giving way to an even more powerful diagnostic tool: magnetic resonance imaging

The CAT scanner, similarly, is giving way to an even more powerful diagnostic tool: magnetic resonance imaging. Upheavals of this sort impart considerable risk to long-term investments in expensive new technologies. The process that eventually resolves such uncertainties is not the textbook competition between producers all seeking to deliver the same product to market at the lowest cost. Rather, it is a competition between different tech-nologies, illustrating that one of the greatest uncertainties facing new technologies is the invention of yet newer ones.

Our lack of knowledge about the relationships between the different dimensions of uncertainty precludes any understanding of its overall impact on technological change. Consider the refinement of complementary technologies and the potential for any technology to form the core of a new system. The complementary technologies might exercise a coercive and conservative pressure, compelling the new technology to be placed inside the current system. Conversely, these complementary technologies might be exactly what is needed to create an entirely new system.

The implications of uncertainty

Researchers are constantly exhorted to ensure the relevance of their work to social and economic needs. Often, however, there is no way of knowing which new discoveries may turn out to be relevant, or to what realm of human activity they may eventually apply. Uncertainty pervades not only basic research, where it is generally recognized, but also product design and new product development. This means that any early commitment to a specific, large-scale project—as opposed to a more limited, exploratory approach—is likely to be risky.

The pervasiveness of uncertainty suggests that governments should resist the temptation to champion any one technology

The pervasiveness of uncertainty suggests that governments should resist the temptation to champion any one technology, such as nuclear power. It is more prudent to manage a deliberately diversified research portfolio that may throw light on a range of alternatives. The approach should be to open many windows and provide the private sector with financial incentives to explore the technological landscapes that can only faintly be discerned from these windows.

Private firms will naturally allocate their R&D funds to projects they hope will turn out to be relevant. Aware that they confront huge uncertainties in the marketplace, they are capable of making their own assessments and placing their bets accordingly. Bad bets are, however, common—so much so that it is tempting to conclude that the manner in which competing firms pursue innovation is wasteful. But such a judgment ignores the role of uncertainty.

It is a singular virtue of the marketplace that, in the face of great uncertainty about the uses of a new technological capability, it promotes exploration along a multitude of paths. In the early stages of a technology, when uncertainties are at their highest and individuals with different views need to be encouraged to pursue their hunches, this property is especially valuable. Technological progress relies on differences of opinion.

The marketplace also provides strong incentives to terminate, quickly and unsentimentally, directions of research whose once glowing prospects have been unexpectedly dimmed by new knowledge, changes in the economic environment, or the restructuring of social or political priorities.

The simultaneous advance in new technology and upgrading of old technology underlines the uncertainty confronting decision makers in a world of rapid change. To imagine that a paradigm could be developed to handle all the relevant factors systematically would be naive. But a more rigorous analysis of the issues raised here might conceivably improve the way we think about the innovation process.