Myths and Realities of Technology Change

22
Myths and Realities of Technology Change by Jeffrey Funk National University of Singapore

description

This book analyzes six myths of technology change and it replaces them with more accurate descriptions of reality. These myths are largely based on metaphors and anecdotal evidence that were presented decades ago and that have not been systematically re-examined. This book summarizes the empirical research that proves these myths wrong and it describes how the more accurate descriptions of reality suggest more appropriate policies and strategies that are very different from the ones suggested by the myths. #1 Performance vs. time curves resemble an S-curve #2 Slowing rates of improvement in old technologies drive the development of new technologies #3 All technologies have the potential for rapid rates of improvement. #4 Product design changes drives performance increases and process design changes drives cost reductions, with product preceding process design changes in a technology’s life cycle. #5 Costs fall as cumulative production rises in a learning curve #6 The future of new technologies cannot be analyzed.

Transcript of Myths and Realities of Technology Change

Page 1: Myths and Realities of Technology Change

Myths and Realities of Technology Change

by

Jeffrey Funk

National University of Singapore

Page 2: Myths and Realities of Technology Change

Chapter 1, Introduction

Technology change is one of the most powerful forces in the world. It has given us dramatic

improvements in economic productivity and standards of living while at the same time creating

winners and losers at the individual, firm, and country level. Technologies such as steam engines,

electricity, internal combustion engines, automobiles, computers, and integrated circuits have had

a particularly large impact on economic productivity and our lives. They have given us marvelous

new products and services, enabled dramatic changes in our life styles, and enabled the substitution

of machines for humans in an increasing number of jobs.

The growing importance of technology means that more is expected from technology than ever

before in human history. People expect technology to give us better food, homes, and vacations,

more time for our families, a cleaner environment, a safer workplace, a happier household, longer

lives, more time to be creative, a better sex life, and a fairer society. We expect scientists and

engineers to effectively develop the right technologies, managers to effectively implement

technologies, policy makers to promote good technological solutions, and universities to help

students deal with an increasingly complex world of technology. These high expectations requires

manager, policy makers, and academics to better understand technology change than is currently

done and to understand it at a much deeper level.

Most technology is embodied in complex systems. Complex systems provide us with food,

shelter, water, manufactured products, entertainment, transportation, and work spaces. These

systems can be subdivided into sub-systems each of which can be further subdivided in what might

be called a hierarchy of sub-systems. For example, manufactured products can be thought of as

systems in which components are processed and assembled in global supply chains. Even the

processing of basic components such as steel pipes is a complex system of basic material processes

that also rely on the mining and transport of raw materials.

These systems are constantly being reconfigured in response to changes and improvements in

technology. Some reconfigurations occur at lower levels and some occur at higher levels. Often

the reconfigurations at higher levels come from improvements in lower-level systems. These

changes percolate up through hierarchies of sub-systems where changes at one level impact on the

next higher level through decisions made by profit-seeking managers.

For example, improvements in integrated circuits made possible better computers, mobile

Page 3: Myths and Realities of Technology Change

phones, and set-top boxes while improvements in areal recording density made smaller forms of

hard disks and tape players possible. Some firms responded to these improvements with more

innovative electronic products while other firms responded very slowly. The slow responses by

other firms suggest that they did not understand the implications of these improvements and the

changes in higher-level systems that these improvements made possible.

Sometimes these changes occur over decades and involve not just managers but also policy

makers. For example, the emergence of complex global supply chains occurred as managers and

policy makers adjusted to improvements in information technology, cargo vessels, ports, rail lines,

trucks, containers, and in many other subsystems. These improvements provided managers and

engineers with new choices about how to design products and services and where to purchase

inputs for them. These improvements also provided policy makers with new choices about

regulations, trade rules and their country’s infrastructure for the global supply chains. The falling

cost of information and transportation enabled products to be global designed and manufactured

and the global supply chains emerged from millions of decisions that were made over many years

by managers and policy makers, each responding to the falling cost of information and

transportation.

This book is concerned with how these choices emerge from technology change. How does

technology change create new choices for managers, engineers, and policy makers in the design

of systems? How does it lead to changes in the design of systems or to the emergence of new

systems? What technologies are experiencing significant changes and why and thus what

technologies are impacting on higher-level systems? Can we use this information to more

effectively design higher-level systems and to more effectively allocate resources?

Economists often call this change creative destruction while management scholars often call it

innovation. Both terms provide insights, but neither capture the essence of this change. For creative

destruction, although a new system is often created that destroys an old system, this destruction

occurs over many years and through millions of decisions and we are interested in these micro-

changes and decisions that lead to the replacement of old systems with new ones. Global supply

changes emerged from millions of decisions that were made in response to the falling cost of

information and transportation technology over many years.

Innovation is also too general and abstract for what we are trying to describe. The term does

not help us understand the millions of micro-changes and decisions that led to the emergence of

Page 4: Myths and Realities of Technology Change

the global supply chains. While some of these decisions were more innovative than others (e.g.,

container shipping), a focus on such innovations does not help us understand the impact of

improvements in lower level systems on higher level systems. For example, the emergence of

global supply chains primarily came from incremental improvements in information systems,

improvements in information systems primarily came from incremental improvements in

computers, and incremental improvements in computers primarily came from incremental

improvements in integrated circuits. Which improvements were innovations and which were not?

Identifying the impact of improvements in lower level systems on higher level ones helps us better

understand how higher level systems change than does identifying which of the changes qualifies

as an important innovation, a not so important innovation, or not an innovation at all.

Management scholars will usually focus on who captures the most value from innovations and

thus the types of strategies that lead to high profits. This book is less interested in these strategies

and more interested in the front end of this process: how do choices about new systems become

available and thus how can managers, engineers and policy makers find new choices and

effectively consider then? How does technology change cause new systems to become

economically possible and thus provide managers with new choices about designs and strategies?

One premise of this book is that any discussion of new systems and how they emerge from

technology change must focus on performance and cost. They are important because managers and

policy makers are concerned with their absolute levels and because they respond to changes in

them. Managers are concerned with absolute levels of performance and cost and changes in them

because they impact on their value propositions to customers and their profits. One of the ways

they respond to changes in the performance and cost of inputs is by changing the ways systems

are designed and configured.

Policy makers also respond to both changes in performance and cost and to their absolute levels.

Some policy makers are more concerned with the former while others are more concerned with

absolute levels. Absolute levels of performance and cost are important because they directly impact

on productivity, standards of living, and quality of life, issues that often concern voters in the long

term. For example, without increases in crop yield (e.g., bushels of wheat per acre of land) and

labor productivity (output per hour of farm workers) it would be difficult to feed the world’s seven

billion people and to keep doing so as the world population increases. Furthermore, the challenges

of improving these performance and cost indices are exacerbated by water shortages and demands

Page 5: Myths and Realities of Technology Change

for lower use of pesticides, insecticides, and fungicides. In other words, meeting these new

challenges require us to increase crop yields and productivity with respect to an increasing list of

inputs.

Almost every problem can be represented by these types of output-to input ratios, of which

examples are shown in Table 1. Better mobile phones have higher faster speeds (bit per second),

higher spectral efficiencies (bits per second and bandwidth), more storage capacity (bits per chip),

better cameras (number of pixels) and faster microprocessors (cycles per second), which may

occur through parallel processing (number of cores). Better computers are measured in terms of

instructions per unit time, per unit-cost, or per kw-hour. More cost effective DNA sequencing and

synthesizing require lower sequencing or synthesizing per unit cost.

Addressing climate change requires improvement in energy output per weight of carbon

emissions along with more efficient and cost effective solar cells. The cost of solar cells is

measured in terms of cost per peak Watt. Energy efficient lighting and displays require higher

luminosity per Watt light bulbs such as LEDs (light-emitting diodes). Economically effective

electric vehicles require higher energy and power storage densities from batteries or alternatives

such as capacitors and flywheels.

A second reason for our interest in performance-to cost ratios is that new technologies and

systems composed from new technologies must reach specific levels of cost and performance for

them to become economically feasible. For example, the cost of solar cells on a per-kw-hour or a

per-peak Watt level must fall to a certain levels before they will begin to diffuse. More generally

speaking, technologies with low carbon emissions must also have much lower costs than they

currently do in order for them to become economically feasible without large subsidies and

contribute towards reductions in carbon emissions. Thus, a major challenge for clean energy

advocates is to find those technologies that are likely to experience cost reductions than are other

technologies and become economically feasible in the near future.

But this raises a number of questions. What drives improvements in a technology’s cost and

performance? Is it improvements on the factory floor, better components, or new product and

process designs? And why do some technologies experience rapid rates of improvement than do

other technologies? Is it because of greater production, demand, R&D, or something else? Unless

we understand the answers to these questions, it is difficult to understand creative destruction and

innovation, to address climate change, and to help engineers and other young entrepreneurs create

Page 6: Myths and Realities of Technology Change

Table 1. Examples of Output-to Input Ratios that are Important to Humanity

Technology

Domain

Measure of Performance

or Sub-Technology

Examples of Output-to Input Ratios

Agriculture Productivity of land,

people, water, and

fertilizers

Crop production per acre of land, per person-hour, per

water volume, and per volume or mass of chemicals

and fertilizers

Housing Floor space Area per person

Electricity

Generation

Cost, efficiency Cost per kw hour of electricity, energy per weight of

carbon emissions, energy generated per area of land,

cost per peak Watt of Solar or Wind, efficiency of

solar cells, cost of carbon sequestration of fossil fuels

Electricity

Transmission

Cost, efficiency Cost per distance of transmitting electricity, energy

loss per length of transmission line, current x length or

current x length per dollar for transmission materials

such as superconductors

Electricity

Storage

Cost, efficiency Energy stored (e.g., joules) per mass (e.g., kg), volume

(e.g., liters), or cost; power (watts) generated per mass,

volume, or cost

Transportation Cost, efficiency Cost per weight-mile for transporting cargo, cost per

person-mile of transporting humans, amount of energy

needed to transport cargo (weight) or humans per mile

Lighting and

Displays

Lighting and Displays Luminosity per Watt, lumens per dollar

Displays Square meters per dollar

Lasers Power density, cost per Watt

Health Care Quality of life Longevity and quality of vision, hearing, smell, other

organs, and of mobility

Cost Percentage of GNP devoted to health care, cost of

specific operations and procedures

Information

Processing

Microprocessor ICs Number of transistors per chip, cost per transistor and

cycle time, number of cycles per second

Camera chips Pixels per dollar, light sensitivity

new systems. Evidence of a lack of understanding can be found in the emphasis on wind turbines

and batteries, both of which are experiencing very slow rates of improvement. To do this, however,

Page 7: Myths and Realities of Technology Change

we must return to the basics and address the assumptions that form the basis for most management

and economic analyses of technology. In reassessing these assumptions, we have found that many

of them are very wrong and thus we call them “myths.” These myths distort the reality of

technology change and mislead the choices available for engineers, scientists, managers, and

policy makers. Our empirical research has identified six myths that impact strongly on how these

people think about technology change:

#1 Performance vs. time curves resemble an S-curve

#2 Slowing rates of improvement in old technologies drive the development of new

technologies

#3 All technologies have the potential for rapid rates of improvement.

#4 Product design changes drives performance increases and process design changes drives

cost reductions, with product preceding process design changes in a technology’s life cycle.

#5 Costs fall as cumulative production rises in a learning curve

#6 The future of new technologies cannot be analyzed.

These myths are largely based on metaphors and anecdotal evidence that were presented decades

ago and that have not been systematically re-examined. The one exception is the learning curve,

but even here the empirical evidence has been selective, ignoring the improvements that occur for

most technologies before commercial production begins. This book summarizes the empirical

research that proves these myths wrong and it replaces them with much more accurate descriptions

of reality. These more accurate descriptions of reality suggest more appropriate methods of

creating new systems that are very different from the ones based on the myths.

The first two myths prevent us from understanding the shapes of performance or cost vs. time

curves and thus rates of improvement for what Giovanni Dosi calls “technology trajectories.” The

myth of S-curves (See left side of Figure 1), which is explored in Chapter 2, makes us believe that

rates of improvement will strongly accelerate when demand, R&D spending or something else

increases and also that slowdowns and physical limits will quickly emerge. The first half of this

myth facilitates the exaggeration and hype of new technologies and makes it appear as if any

technology is right around the corner as long as governments, firms, and other potential investors

make the right investment and aren’t too focused on the short term. As one of my colleagues said

Page 8: Myths and Realities of Technology Change

as I discussed the slow rate of improvement for electric batteries, “but improvements will

accelerate once the demand increases.” In other words, real data isn’t as important as a belief in

the accelerations that form the basis for the S-curve. Purported accelerations have been used to

justify investments in solar cells, wind turbines, and electric vehiclesi.

Time

Performance

Figure 1. Myth vs. Reality of Performance vs. Time Curves on Logarithmic Scale

Slowdown and Limits

Acceleration

Time

Performance(logarithmicscale)

a. The Myth b. The Reality

Note: limits exist but they are often further away than ordinarily thought

The second half of this myth makes us believe that slowdowns and limits will quickly emerge

and thus we must quickly jump to a new technology. Michael Tushman and Philip Andersonii make

the strongest argument for steep accelerations that are quickly followed by slowdowns and limits

in that long periods of incremental change are punctuated by relatively rare innovations that

provide sharp price-performance improvements over existing technologies. Or in their words,

“technology evolves through relatively long periods of incremental change punctuated by

relatively rare innovations that radically improve the state of the art.” Building from this paper and

findings from evolutionary biology, other scholars use the term “punctuated equilibrium” to

describe a long-term process in which stable environments are punctuated by new technologies

that provide jumps in performance and that also instigate organizational transformation.

Punctuated equilibrium is one of the most powerful and often cited theories in management and it

increases the hype of S-curvesiii.

Unfortunately, there is little evidence for punctuated equilibrium or S-curves particularly if one

looks at logarithmic plots of data from engineering and science journals. One must use logarithmic

Page 9: Myths and Realities of Technology Change

plots because a straight line on a logarithmic plot means constant rates of improvement over time

while a straight line on a linear plot means a decreasing rate of improvement. Engineering and

science journals plot data for rapidly improving technologies on logarithmic plots and such plots

typically include some combination of best laboratory results and commercialized products.

Without data from best laboratory results, there may be gaps in the data that come from time

periods between product releases and that falsely imply jumps in performanceiv.

Our analysis of performance vs. time curves for 25 technologies and 32 measures show that the

shape of performance-versus time curves more closely resembles a straight line on a logarithmic

plot (See right side of Figure 2) than the classical S-curve and that punctuated equilibrium is a

highly misleading metaphor. Thus, we should not expect or plan for early accelerations or for early

limits. Instead, we should plan for and expect incremental improvements to occur at fairly constant

rates of improvement. These constant rates of improvement allow us to compare rates of

improvements for different technologies and to consider these comparisons when long term plans

for development and implementation are being considered. Firms can and do consider rates of

improvement when they choose research projects and clean energy advocates should consider rates

of improvement when they propose potential solutions.

The second myth, which is explored in Chapter 3, extends the myth of S-curves further into a

fantasy land in which slowdowns in a single old technology are perfectly linked with accelerations

in a single new technology (See left side of Figure 2). In other words, it is the slowdown in an old

technology that causes the development and improvement of a new technology and for an

acceleration in the rate of improvement of this “single” new technology to occur. This myth about

“linked S-curves” suggests that demand is the most important lever for policy and it is the most

important driver for firms when they create R&D strategies. According to this myth, a slowdown

in an old technology represents demand for a better solution and the acceleration in the new

technology represents the response to the slowdown.

Since accelerations were not found in our analysis of the first myth of S-curves, there is little

chance that such accelerations will occur at the same time as slowdowns in the linked S-curves

shown in Figure 2. The best one can hope for with this myth is that a slowdown occurs in an old

technology as a new technology experiences some form of improvements. Chapter 3 analyzes this

possibility using 15 pairs of old and new technologies and finds only two technologies experienced

statistically significant slower rates of improvement after a performance metric was recorded for

Page 10: Myths and Realities of Technology Change

a new technology, thus suggesting that slowdowns do not generally cause the development and

improvement of new technologies. Furthermore, Chapter 3 finds that multiple new technologies

are being developed simultaneously as replacements for an old technology (See right side of Figure

2), unlike the single performance-to cost curve that is presented in the linked S-curve theory (left

side of Figure 2).

Time

Performance

Figure 2. Myth vs. Reality of Slowdowns Driving Improvements in New Technologies

Slowdown

Acceleration

Time

Performance(logarithmic Scale)

a. The Myth b. The Reality

Many technologies are simultaneouslybeing developed in a very decentralized manner and their timingdepends on supply-side factors

Chapter 3’s analysis is also more consistent with the “technology push” than of “market pull”

theories of technology change, a debate that has continued for more than 40 years with Clayton

Christensen’s theory of disruptive innovation providing recent evidence for market pull.

Proponents of technology push argue that technology change is driven by universities and other

laboratories that “push” technologies into the marketplace while others argue that technology

change is driven by a market that “pulls” ideas for new technologies from the minds of scientists

and engineers in universities and laboratories through targeted research. One of the persuasive

arguments for technology push in the 1970s was that few of the needs addressed by the important

“innovations” of the 20th century had been recognized before the new technology was developed.

Looking at subsequent innovations such as cellular phones and the Internet, one can also perceive

this lack of early recognition in the needs for these innovations.

Chapter 3’s analysis suggests that performance vs. time curves are being driven much more by

Page 11: Myths and Realities of Technology Change

supply than demand-side factors. Although demand becomes more important as a new technology

experiences improvements and thus nears economic feasibility, the initial improvements have

more to do with supply side factors such as advances in science or accidental discoveries than with

a slowdown in an old technology. It is only as new technologies are developed and improved that

the state of the old technology becomes a more relevant driver of the improvements in the new

technologies.

The third myth, which is explored in Chapter 4, is that all technologies have the same potential

for improvements and thus the same potential for rapid rates of improvement. Although this myth

has not been explicitly stated anywhere to our knowledge, it is implicitly stated almost everywhere.

One type of evidence for the existence of this myth is that rates of improvement are rarely

discussed and few people are aware of them, one exception is Moore’s Law. An ignorance of rates

of improvement is particularly common among social scientists even those who specialize in

technology. For example, Clayton Christensen’s books and papers on disruptive innovations do

not mention rates of improvement even though rates of improvement clearly impact on the

replacement of old technologies with new ones and even though his theory is largely based on an

analysis of hard disk drives, which have experienced unusually rapid rates of improvement.

Christensen conveniently ignores such details perhaps in an attempt to promote the universality of

his theory.

Ignoring rates of improvement is also common among advocates of sustainability. Most papers,

books, presentations, and courses on sustainability rarely mention rates of improvement and the

particularly slow rates of improvement for wind turbines and batteries. A notable exception is

Vaclav Smil’s books on energy and sustainability but his books sell far fewer copies than do those

of Thomas Friedman, who does not mention rates of improvement in his books or in his New York

Times columns on clean energy. Even discussions with professors of sustainability or

sustainability-related organizations such as the Intergovernmental Panel on Climate Change or the

Society of Risk Analysis will merely produce blank stares, laughs, and sometimes anger. A

common response is “it’s not just about technology,” as if considering rates of improvement

prevents ones from considering other factors.

Equally troubling, the few analyses of rates of improvement confuse rates of improvement for

each doubling of cumulative production with annual rates of improvement. This causes them to

overestimate the rates of improvement for some technologies since annual rates and those for

Page 12: Myths and Realities of Technology Change

doubling will only be the same when cumulative production is doubling each year, which rarely

occurs. Most technologies experience much slower rates of production growth than 100% per year

and thus their annual rates of improvement are 1/10 to ¼ the rates for each doubling of cumulative

production.

The reality is that some technologies have much faster annual rates of improvement than do

other technologies. As shown in Figure 3, 2/3 of the 120 technologies have rates of less than 9%

and 89% of them have rates of less than 15% per year. Furthermore, the empty space between 14%

and 24% and our analysis of chemical vs. non-chemical technologies suggest that there are

multiple distributions in Figure 3, which enable us to draw conclusions about how these

improvements occur. Improvements in chemicals occur through different mechanisms than do

non-chemicals since product innovations are more difficult with fixed-formula chemicals than with

non-chemicals and since chemicals benefit more from increases in production scale than do other

technologies. By analyzing the multiple distributions in Figure 3, we are able to better understand

the design changes that enable improvements. After analyzing other possible reasons for the

differences such as greater production, demand, and R&D, Chapter 4 concludes that some

technologies have greater opportunities for improvements from product and process designs than

do other technologies, a conclusion that was reached by Nathan Rosenberg more than 40 years ago

but contradicts the recent emphasis on demand-based theories (e.g., Clayton Christensen’s theory

0

5

10

15

20

25

-10 -8 -6 -4 -2 0 2 4 6 8

10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 >42

FIgure 3. Number of Technologies by Annual Rates of Improvement

Annual Rates of Improvement

Page 13: Myths and Realities of Technology Change

of disruptive innovation) and on demand-based subsidies as a tool for promoting clean energy

technologies.

The fact is that technologies with rapid rates of improvement generally have had and will

continue to have a larger impact on our world than do other technologies. Technologies such as

integrated circuits (ICs), hard disk drives, lasers, and computers have had a large impact on our

world partly because they have experienced one or two orders of magnitude improvements each

decade resulting in 5 to 10 orders of magnitude improvements over the last 50 years. They have

enabled new forms of systems to emerge, one of which is the Internet and another of which is the

global supply chains that were mentioned earlier and that depend on the Internet. They have also

enabled other systems to emerge, many of which are defined as disruptive innovations by Clayton

Christensen. Thus, if we are trying to understand technology change or we are looking for

disruptive innovations, we should be focusing on technologies with rapid rates of improvement as

this will provide us with better ideas than will looking for Christensen’s low-end innovations.

Technologies with rapid rates of improvement in performance and/or cost will also have faster

rates of diffusion than will ones with slower rates of improvement. While rates of diffusion

depends on a variety of factors, the largest factor identified in diffusion studies for the U.S. is the

profitability of a technology for usersv. The users that profit most from a new technology, whether

these users are industrial ones or consumers, are the first adopters and faster rates of improvement

in performance and/or cost mean faster rates of improvement in the profitability for users. One

reason for the large importance of profitability for users in determining rates of diffusion is that

the U.S. and other developed countries have less barriers to entry for “creative destruction” than

do other countries, which is of course why the U.S. and other developed countries are richer than

are poorer ones.

In other words, slow rates of improvement are a type of implementation problem that must be

overcome if a technology is to succeed. Scholars of technology management often focus on

implementation problems and thus argue that rates of improvement are less important than are

organizational or regulatory challenges. This book argues that slow rates of improvement are

probably a larger problem than are organizational and regulatory challenges in developed countries

because these countries have become good at solving organization and regulatory problems

through existing and new firms, venture capital, and changes in regulations by policy makers.

The importance of slow rates of improvement as an implementation problem means that

Page 14: Myths and Realities of Technology Change

understanding why some technologies have more rapid rates of improvement than do others is a

critical question. Understanding the answer to this question can help us achieve more rapid rates

of improvement in existing technologies and more importantly identify technologies with a greater

chance of achieving rapid rates than do other technologies. Answering this question about the

reasons behind rapid rates requires us to address two other myths and both these two myths and

the prior two myths also contribute to the myth that all technologies have the same potential for

improvements. When one believes in linked S-curves, accelerations occurring early in a

technology’s lifecycle, the slowdowns that drive these accelerations and two other myths, it is easy

to believe that all technologies have the same potential for improvements.

Fre

qu

en

cy o

f In

no

vati

on

an

d R

ate

of

Imp

rovem

en

t

Figure 4. Myth vs. Reality of Product and Process Innovations

Product Innovation

Process

Innovation

Time

Increases in performance from

product innovations

Reductions in price from

process innovations

Fre

qu

en

cy o

f In

no

vati

on

an

d R

ate

of

Imp

rovem

en

tIncreasing number of

inter-related product and

and process innovations

Time

a. The Myth b. The Reality

Relatively constant rate

of improvements in

cost and performance

The fourth myth, which is explored in Chapter 5, is that product design changes lead to increases

in performance early in the life cycle and process design changes lead to reductions in costs later

in the life cycle (See Figure 4). This myth causes firm strategies and government policies to focus

on product design changes to achieve performance increases and to focus on process design

changes to achieve cost reductions. “Process” in general may include any activities that contribute

to cost and the strategic importance of process innovation for competition increases through the

life cycle of an industry, as the opportunities decline for product innovations that lead to

Page 15: Myths and Realities of Technology Change

improvements in functional performance. Many scholars cite some version of this assessment and

some take this assessment further and argue that increasing returns to process R&D lead to a

shakeout, i.e., dramatic decline, in the number of firmsvi.

This myth is addressed by analyzing the cost and performance of 17 different technologies, 22

unique time-series pairs of performance and cost, 358 unique data points and 705 total years of

data for these time-series pairs. Technologies such as chemicals, agricultural products and

materials are excluded since they often have a fixed chemical composition and thus improvements

in performance do not occur and most of their improvements in cost are driven by process

innovations or by increasing the scale of their production equipmentvii. But for other technologies,

our analysis shows that improvements in performance and cost for most products are highly

correlated over many decades suggesting that the same design changes are impacting on

improvements in both cost and performance.

What types of design changes might cause simultaneous improvements in both cost and

performance and also cause very different rates of improvement in different technologies? Some

changes in product design lead to both improvements in performance and cost. For example,

improvements in the efficiency of solar cells, lights and displays and in the speed of electronic

devices, computers and other electronic products also lead to reductions in the cost of these

technologies. Other design changes involve inter-related product and process design changes that

can be implemented by multiple firms in what many call modular design or by a single firm in

what may be called “integral” design. For example, although firms such as Intel design both the

product and process for new microprocessors, improvements in many ICs are achieved through

cooperation between foundries and design houses. However, whether they are implemented via

modular or integral design, is not the issue here, it is the types of inter-related product and process

design changes that enable the improvements and in particular the rapid improvements. Our

analysis of the next myth, the fifth one, helps us better understand the specific types of inter-related

product and process design changes that lead to improvements in cost and performance and

explains why some technologies experience more rapid rates of improvements than do other

technologies.

The fifth myth, which is analyzed in Chapter 6, is that costs fall as cumulative production rises

and as improvements are made to processes on the factory floor (See Figure 5). Since the

publication of Theodore Paul Wright’s analysis of fighter jet costs in 1936, empirical analyses

Page 16: Myths and Realities of Technology Change

correlating cost reductions to cumulative production have grown extensively in what some call

learning curves. The early work on learning curves was mostly done on single designs in specific

factories and thus analyzed the impact of factory level changes on factory productivity.

Subsequently, learning curves have been applied to technologies that are manufactured with new

designs and in new factories where the output variable might be cost or performance, albeit these

models are now often called experience curves. For example, the costs of ships, solar cells,

semiconductor memory, chemicals, primary metals, and food have been analyzed using this

approach, across significant design changes and often throughout all global factoriesviii.

Cumulative Production(log scale)

Costs(log scale)

Figure 5. Myth vs. Reality of Cost Reductions (i.e., learning curve)

Time

Costs(logscale)

a. The Myth b. The Reality

Start of Commercial Production

Linking cumulative production to reductions in cost or improvements in performance can lead

to confusion about how the improvements in cost and performance are being achieved. Some

believe that such a linkage suggests most of the improvements are occurring on the factory floor

while others note that cumulative production indirectly leads to improvements in performance.

Increases in production are linked with expected future production and lead to increased incentive

to perform process-related and general R&Dix where the results of the increased R&D spending

lead to improvements in performance or cost. This argument is also implicit in Christensen’sx

analysis of hard disk drives, computers and other “disruptive” technologies in that the emergence

of a low-end product lead to increases in R&D spending and thus rapid improvements in the new

product, which in turn leads to replacement of the dominant technology by the low-end innovation.

Page 17: Myths and Realities of Technology Change

Linking cumulative production to reductions in cost or improvements in performance also

makes it easier for policy makers to justify demand-based subsidies for desirable new technologies

such as clean energy ones. It is particularly easy to justify demand-based subsidies if the cost

reductions are coming from activities on the factory floor and thus the subsidies will probably

encourage the cost reductions. However, what if they are not coming from activities on the factory

floor but instead are coming from R&D activities in the laboratories? If it is the latter, then a

different set of policies are needed to achieve cost reductions than the current emphasis on demand-

based subsidies.

Chapter 5 and 6’s analysis suggests a different reality than the myth of increases in cumulative

production lead to cost reductions. The correlation that is found between improvements in cost and

performance in Chapter 5 suggests that costs are falling because of changes in product and process

design that also impact on improvements in performance. Thus, increases in cumulative production

and changes in process design by themselves cannot be the reasons for falling costs. This argument

is taken one step further in Chapter 6 by conducting two forms of analysis that enable us to better

understand how rapid improvements occur.

Thirteen new technologies are analyzed that experienced rapid improvements of greater than

10% per year without commercial production. These include organic materials for transistors, solar

cells, and light-emitting diodes (LEDs), quantum dots for displays and solar cells, new forms of

non-volatile memory integrated circuits such as resistive or magnetic RAM (random access

memory), carbon nanotubes, superconducting materials for both energy transmission and

integrated circuits, and quantum computers. Since there was no commercial production during

most of the time periods in which the rapid improvements occurred, the improvements must be

from factors other than factory floor activities and this chapter examines the specific product and

process design changes that enabled the rapid improvements to occur. Our analysis suggests that

these design changes occur in laboratories and there are primarily two types of inter-related product

and process design changes: 1) creating new materials (and their associated processes); and 2)

reducing the scale of features in the product through process improvements.

Cost breakdowns of higher-level systems in which components have a larger impact on costs

than does the assembly of the system are also analyzed. For example, cost breakdowns of

electronic products such as computers, mobile phones, game consoles, set-top boxes, and eReaders

have found that 95% of their system costs are represented by the cost of components and thus

Page 18: Myths and Realities of Technology Change

increases in cumulative production and learning in assembly can only impact on about 5% of the

system costs. Furthermore, since many of the assembly operations are standard ones that are used

across many different electronic products, learning in factories probably has a small impact even

on the 5% of the costs that are represented by the assembly operations. Instead, it is improvements

in components that enable improvements in system cost and performance. Furthermore, since these

improvements in components are also occurring before a new system is commercially produced,

it is improvements in components and not the manufacture of the new system that causes it to

become economically feasible.

Chapter 7 presents our reality of technology change. It uses the analyses from the previous

chapters to summarize our view of technology change. Performance vs. time curves more closely

resemble straight lines on a logarithmic plot than an S-curve. Accelerations do not occur and

physical limits take decades if not longer to emerge. The improvements in the new technology are

not driven by a slowdown in the old technology and in fact multiple technologies are competing

to replace a new technology even without a slowdown in the old technology. This competition

primarily occurs in laboratories where advances in science and accidental discoveries determine

the timing of the first recorded performance metrics and the rates of improvement largely

determine the eventual winning technologies. A lack of S-curves means that pre-production early

rates of improvement can provide an important signal for future rates and thus help us identify the

technologies with the large potential for rapid improvements.

Different technologies have different rates of improvement primarily because of differences in

opportunities for improvements. Simply put, some technologies experience more rapid rates of

improvement because they benefit more from the two inter-related product and process design

changes mentioned above than do other technologies. For the first one, technologies that benefit

from reductions in scale experience very rapid rates of improvement and only some technologies

benefit from reductions in scale. Thus, by itself this type of design change can explain rapid rates

of improvement and help us identify technologies with the potential for rapid rates of

improvement.

It is less clear why some technologies experience more rapid rates of improvement from the

creation of materials (and their associated processes) than do other technologies. Interviews with

several physicists suggest that technologies involving the deposition of thin crystalline films of

less than one micro-meter on other crystalline materials are easier to improve than are other

Page 19: Myths and Realities of Technology Change

materials. All of the technologies that benefited from the creation of materials and their associated

processes involve the thin-film deposition of crystalline materials while other technologies with

slower rates of improvement do not involve thin film deposition of crystalline materials.

In any case, rapid improvements increase the chances that a technology will become

economically feasible or enable new forms of higher-level systems to emerge. Thus, if one is

looking for creative destruction, disruptive innovation, or some other new system that might

replace an existing one, one should be looking at technologies that experience rapid improvements.

Finding these systems is a goal of managers, policy makers, university professors and even

students.

Chapter 8 uses this better reality of technology change to addresses the sixth myth: we can’t

analyze the future. This is probably the most controversial and emotional myth because most social

scientists dismiss analyses of the future as useless while failing to recognize that managers, policy

makers, professors, and students are constantly making decisions that imply predictions about the

future. When managers introduce R&D budgets or new product or students choose university

majors, they are making a prediction about the future. The firms are predicting that the R&D

budgets and new products will probably provide greater benefits than their costs. The students (or

their parents) are predicting that the university degrees will provide them with benefits that are

greater than the costs of the education. Even social science professors, who enjoy criticizing

predictions, are implicitly making a prediction when they draw conclusions about specific policies,

strategies, or organizational designs. By concluding that something has worked better than

something else, they are implying that this conclusion will be true in the future.

More importantly, criticizing predictions prevents social scientists from understanding the

challenges of searching for entrepreneurial opportunities, solving global problems such as

sustainability, helping students search for opportunities and address sustainability, and setting

effective R&D policies. Where should young entrepreneurs search for opportunities? How should

they try to address sustainability? Which technologies and markets should they consider? What

kinds of systems should they introduce? These types of questions must be addressed in order to

search for entrepreneurial opportunities, solve global problems such as sustainability, help students

search for opportunities and address sustainability, and set effective R&D policies. Unless social

scientists help their students make such decisions, they cannot help them address these issues and

helping them make such decisions requires them to make predictions about which solutions have

Page 20: Myths and Realities of Technology Change

the largest chance of success.

An aversion to helping young entrepreneurs choose the technologies or markets to look for

opportunities reminds me of a joke that economists like to tell. A man walking down the street sees

a man crawling around on his hands and knees under a lamp. The pedestrian asks the crawler, what

are you looking for. The crawler says “my keys. I lost them over there, as he points to a place not

under the lamp.” The pedestrian then asks while pointing at the place where he is crawling, why

are you looking there?” The crawler says, “this is where the light is.”

Analyzing the future requires us to look in places where the light is not very bright. Although

the previous chapters of this book have increased the luminosity of the light by disproving myths

and replacing them with more useful insights into of technology change, the future is still dimly

lit. It is all about probabilities. Which types of systems have the largest probability of becoming

economically feasible? Searching for entrepreneurial opportunities, solving global problems such

as sustainability, helping students search for opportunities and address sustainability, and setting

effective R&D policies requires us to think about the technologies with the greatest chance of

achieving improvement and thus the decisions that will probably lead to the best futures.

Chapter 8 distinguishes between predictions of the future and analyzing the future. After

analyzing predictions made by MIT’s Technology Review, it then shows that this book’s insight

about rapid improvements provide us with a better way to analyze the future than does Technology

Review’s method of expert elicitation. Not only do these rapid rates of improvement provide us

with better predictions than does Technology Review’s method, it also enables us to reduce forecast

error by gathering and analyzing more data. The other chapters reduce forecast error by helping us

better understand technology change. This chapter focuses on reducing forecast error by better

understanding rates of improvement, comparisons between old and new technologies, the

composition of systems, and the extent to which improvements are needed.

This chapter draws on research done by the author for a course on technology change. This

course helps students understand the composition of existing and potential systems, comparisons

between old and new systems, rates of improvement that impact on these systems, and the extent

to which improvements are needed before these new systems become economically feasible.

Effectively analyzing these factors and the uncertainty in them can help students better understand

when new technologies become economically feasible. More specifically, better understanding the

composition of systems can lead to better cost and performance estimates, better comparisons

Page 21: Myths and Realities of Technology Change

between new and old systems, and better estimates for when the new systems might become

economically feasible. This is one goal of this book and has been the goal of the author for more

than 10 years.

i McGrath R 1998. Technological discontinuities and media patterns: assessing electric vehicle batteries,

Technovation 18(11): 677-687. Anderson B and Jacobsson S 2000. Monitoring and assessing technology

choice: the case of solar cells, Energy Policy 28(14): 1037-1049. Schilling M and Esmundo M 2009.

Technology S-curves in renewable energy alternatives: Analysis and implications for industry and

government, Energy Policy 37(5): 1767-1781

ii Tushman, M. and Anderson, P. 1986. Technological Discontinuities and Organizational Environment,

Administrative Science Quarterly 31: 439-456.

iii For punctuated equilibrium in evolutionary biology see Eldredge, Niles and S. J. Gould

(1972). "Punctuated equilibria: an alternative to phyletic gradualism" For punctuated equilibrium in

organizations, see Gersick C 1991. Revolutionary Change Theories: A Multilevel Exploration of the

Punctuated

Equilibrium Paradigm, The Academy of Management Review 16(1): 10-36. Miller D and Friesen P 1984.

Organizations: A quantum view. Englewood Cliffs, NJ: Prentice-Hall. Tushman M. and Romanelli E 1985.

Organizational evolution: A metamorphosis model of convergence and reorientation. Research in

Organizational Behavior 7: 171-222. Loch C and Huberman B 1999. A Punctuated-Equilibrium Model of

Technology Diffusion, Management Science 45(2): 160-177.

iv For example, most of the commercial product types for which Sood and Tellis (2005) report step-

wise functions do not show large numbers of data points for the relevant periods of time. In

extensions to this analysis, they develop a step-wise model (Sood and Tellis, 2009) and contrast

this model to S-curves and other models (Sood et al, 2012). In doing so, they focus on products,

many of which initially failed, thus causing large gaps to emerge in the time series before another

product is released. Although they correctly cite these gaps as evidence of a step-and weight

strategy for products, improvements probably continued to be made on the technologies in

Page 22: Myths and Realities of Technology Change

laboratories even as the firms did not introduce new products. This is an important methodological

limitation of analyzing the performance of products before a technology has consistent market

success. If one is to look for an acceleration during the early years of a technology, one must

include best laboratory results. After all, a major objective of such an analysis of performance vs.

time curves is to understand whether an acceleration will occur as demand, investment, or

something else increases.

v Griliches Z 1957. Hybrid Corn: An Exploration in the Economics of Technological Change,

Econometrica 25 (4), 501-522. Mansfield E 1968. The Economics of Technological Change, NY: W.W.

Norton & Company Inc.

vi Early and heavily cited papers concerning this theory include: Utterback J and Abernathy W, 1975. A

dynamic model of process and product innovation, Omega, 3(December):639-656. Abernathy W and

Utterback J, 1975. Patterns of technological innovation, Technology Review 80, June/July 1978, pp. 40-

47. Utterback, J., 1994, Mastering the dynamics of innovation, Harvard Business School Press. Adner R

and D Levinthal, 2001. Demand Heterogeneity and Technology Evolution: Implications for Product and

Process Innovation, Management Science 47(5): 611-628. For the impact on a shakeout in the number of

firms, see Klepper, S. 1996. Entry, exit, growth and innovation over the product life cycle, American

Economic Review 86(3) 562–583. Klepper S 1997. Industry Life Cycles, Industrial and Corporate

Change 6(1): 145-81

vii Stobaugh, R 1988. Innovation and Competition: The Global Management of Petrochemical

Products. Boston: Harvard Business School Press

Utterback, J., 1994, Mastering the dynamics of innovation, Harvard Business School Press.

viii Wright T P, 1936. Factors Affecting the Cost of Airplane, Journal of Aeronautical Sciences,

3(4): 122 – 128. Argote L 1999. Organizational Learning: Creating, Retaining and Transferring

Knowledge, NY: Springer. Argote L and Epple D 1990. Learning Curves in Manufacturing, Science

247(4945): 920- 924. Ayres, R. 1992. Experience and the life cycle, Technovation 12(7): 465-486 Thornton

R and Thompson P 2001. Learning from Experience and Learning from Others, American Economic

Review 91(5): 1350-1368. Nagy B, Farmer D, Bui Q, Trancik J 2013. Statistical Basis for Predicting

Technological Progress. PLoS ONE 8(2): e52669. doi:10.1371/journal.pone.0052669NREL, 2013.

ix Schmookler, J. 1966. Invention and Economic Growth, Cambridge, Harvard University Press.

Sinclair G, Klepper S, and Cohen W 2000. What’s experience got to do with it? Sources of cost

reduction in a large specialty chemicals producer. Management Science, 46: 28–45.

x Christensen, C 1997, The innovator’s dilemma, Harvard Business School Press, Boston, MA