Economics of Software and the Importance of Human Capital
By Richard R. Nelson and Paul M. Romer
have long appreciated the centrality of technical advance in the
process of economic growth, a complete understanding of the key
processes, investments, and actors that combine to produce it has
not come easily. Indeed, these processes are very complex and variegated.
Economists broadly understand that the advance of technology is
closely associated with advances in knowledge. It also is clear
that new knowledge must be embodied in practices, techniques, and
designs before it can affect an economic activity. Beyond this,
different economic analyses focus on or stress different things.
stress the "public good" aspects of technology, seeing
new technology as ultimately available to all users. Others treat
technology as largely a "private good," possessed by the
company or person that creates it. Many economists have studied
research and development as the key source of new technology. Those
that have focused on R&D done by private, for-profit business
firms naturally assumed that the technology created through corporate
R&D is, to some extent at least, a private good. By contrast,
economists who have stressed the "public good" aspects
of technology have focused on government investments in R&D,
"spillovers" from private R&D, or both. (These spillovers
are another manifestation of the divergence between the public and
private returns noted above.) Still others argue that a single-minded
emphasis on organized R&D as the source of technical advance
sees the sources too narrowly. They point to evidence that learning-by-doing
and learning-by-using are important parts of the processes whereby
new technologies are developed and refined.
on which economists have been of different minds is whether technical
advance and economic growth fueled by technical advance can adequately
be captured in the mathematical models of economic equilibrium that
economists developed to describe a static world. Joseph Schumpeter
and economists proposing "evolutionary" theories of growth
have stressed that disequilibrium is an essential aspect of the
process. By contrast, recent theories that descend from neoclassical
models presume that the essential aspects of technical advance and
economic growth can be captured by extending the static equilibrium
While we do
not want to underplay the important open questions about how economists
ought to understand technical advance, a workable consensus for
policy analysis seems to be emerging from these divergent perspectives.
Technology needs to be understood as a collection of many different
kinds of goods. These goods can have the attributes of public goods
and private goods in varying proportions. Some are financed primarily
by public support for R&D, others by private R&D. Both business
firms and universities are involved in various aspects of the process.
Other parts of technology are produced primarily through learning-by-doing
and learning-by-using, both of which can interact powerfully with
research and development. There are aspects of the process that
are quite well treated by equilibrium theories, with their emphasis
on foresight, stationariness, and restoring forces. Still other
aspects are better suited to the evolutionary models, with their
emphasis on unpredictability and the limits of rational calculation.
One way to summarize
this emerging view is to focus on three types of durable inputs
in production. We will take our imagery and language from the ongoing
digital revolution and refer to these three different types of inputs
as hardware, software, and wetware. Hardware includes all the nonhuman
objects used in production, both capital goods such as equipment
and structures and natural resources such as land and raw materials.
Wetware, the things that are stored in the "wet" computer
of the human brain, includes both the human capital that mainstream
economists have studied and the tacit knowledge that evolutionary
theorists, cognitive scientists, and philosophers have emphasized.
By contrast, software represents knowledge or information that can
be stored in a form that exists outside of the brain. Whether it
is text on paper, data on a computer disk, images on film, drawings
on a blueprint, music on tape, yen thoughts expressed in human speech,
software has the unique feature that it can be copied, communicated,
The role of
software, hardware, and wetware can be discerned in a wide variety
of economic activities. Together they can produce new software,
as when a writer uses her skills, word processing software, and
a personal computer to write a book. They can produce new hardware,
for example, when an engineer uses special software and hardware
to produce the photographic mask that is used to lay down the lines
in a semiconductor chip. When an aircraft simulator and training
software are used to teach pilots new skills, they produce new wetware.
types of inputs can be discerned in activities that are far removed
from digital computing. In the construction of the new city of Suzhou
in Mainland China, the government of Singapore says that its primary
responsibility is to supply the software needed to run the city.
The hardware is the physical infrastructure, roads, sewers, and
buildings, etc., that will be designed according to the software.
The wetware initially will be the minds of experts from Singapore,
but eventually will be supplied by Chinese officials who will be
trained in Singapore to staff the legal, administrative, and regulatory
bureaucracies. The software comprises all the routines and operating
procedures that have been developed in Singapore, examples of which
range from the procedures for designing a road, to those for ensuring
that police officers do not accept bribes, to instructions on how
to run an efficient taxi service.
models of growth describe output as a function of physical capital,
human capital, and the catch-all category, "technology."
The alternative proposed here has the advantage of explicitly distinguishing
wetware (i.e., human capital) from software. This is an essential
first step in a careful analysis of the intangibles used in economic
activity. The next step is to identify the reasons why software
differs from both hardware and wetware.
two key attributes that distinguish different types of economic
goods: rivalry and excludability. A good is rival if it can be used
by only one user at a time. This awkward terminology stems from
the observation that two people will be rivals for such a good.
They cannot both use it at the same time. A piece of computer hardware
is a rival good. So, arguably, are the skills of an experienced
computer user. However, the bit string that encodes the operating-system
software for the computer is a nonrival good. Everyone can use it
at the same time because it can be copied indefinitely at essentially
zero cost. Nonrivalry is what makes software unique.
is physically possible for a nonrival good to be used by many people,
this does not mean that others are permitted to use it without the
consent of the owner. This is where excludability, the second property,
comes in. A good is said to be excludable if the owner has the power
to exclude others from using it. Hardware is excludable. To keep
others from using a piece of hardware, the owner need only maintain
physical possession of it. Our legal system supports each of us
in our efforts to do this.
It is more difficult
to make software excludable because possession of a piece of software
is not sufficient to keep others from using it. Someone may have
surreptitiously copied it. The feasible alternatives for establishing
some degree of control are to rely on intellectual property rights
established by the legal system or to keep the software, or at least
some crucial part of it, secret.
Our legal system
assigns intellectual property rights to some kinds of software but
not others. For example, basic mathematical formulas cannot be patented
or copyrighted. At least at the present time, there is no way for
the scientists who develop algorithms for solving linear programming
problems to get intellectual property rights on the mathematical
insight behind their creation. On the other hand, the code for a
computer program, the text of a novel, or the tune and lyrics of
a song are examples of software that is excludable, at least to
classification of goods according to excludability and rivalry creates
four idealized types of goods. Private goods and public goods are
the names given to two of these four types. Private goods are both
excludable and rival. Public goods are both nonexcludable and nonrival.
The mathematical principles used to solve linear programming problems
are public goods. Because they are software, they are nonrival;
it is physically possible to copy the algorithms out of a book.
Because the law lets anyone copy and use them, they are nonexcludable.
to private goods and public goods, there are two other types of
goods that have no generally accepted labels but are important for
policy analysis. The first are goods that are rival but not excludable.
The proverbial example is a common pasture. Only one person's livestock
can eat the grass in any square foot of pasture, so pastureland
is a rival good for purposes of grazing. If the legal and institutional
arrangements in force give everyone unlimited access to the pasture,
it is also a nonexcludable good. Frequent allusions to "the
tragedy of the commons" illustrate one of the basic results
of economic theory: Free choice in the presence of rival, nonexcludable
goods leads to waste and inefficiency.
The fourth category,
and one of central importance to the study of technical advance,
is of nonrival goods that are excludable, at least potentially.
We stress the term "potentially" here because society
often has a choice about the matter. It can establish and enforce
strong property rights, in which case market incentives induce the
production of such goods. Alternatively, it can deny such property
rights. Then if the goods are to be provided, support through government
funding, private collaborative effort, or philanthropy is needed.
Many of the most important issues of public policy regarding technical
advances are associated with this latter choice. For rivalrous goods,
establishing and enforcing strong property rights is generally a
good policy (although there are exceptional cases.) But for nonrivalrous
goods, the matter is much less clear.
By and large,
society has chosen to give property rights to the kind of software
commonly called "technology" and to deny property rights
but provide public support for the development of the software commonly
referred to as "science." Establishing property rights
on software enables the holder of those rights to restrict access
to a nonrival good. When such restriction is applied, for example,
by charging a license fee, some potential users for whom access
would be valuable but not worth the fee will choose to forego use,
even though the real cost of their using it is zero. So putting
a "price" on software imposes a social cost, positive-value
uses that are locked out, and in general the more valuable the software
is to large numbers of users, the higher will be the cost. To cite
just one example that influences the choices of working scientists,
there are experiments that could be carried out using PCR (polymerase
chain reaction) technology that would be done if the scientists
involved could use this technology at the cost of materials involved.
Some of these are not being done because the high price charged
by the current patent holder makes this research prohibitively expensive.
Note that this
is very different from what is entailed in establishing property
rights on rival goods. Only one user can make use of a rival good
at any one time. So property rights, or options to sell them, encourage
the rival good to be used by those to whom it is most valuable.
Our legal system
tries to take account of the ambiguous character of property rights
on software. We give patents for some discoveries, but they are
limited in scope and expire after a specific period of time. For
rival goods this would be a terrible policy. Imagine the consequences
if the titles to all pieces of land lapsed after seventeen years.
For some nonrival goods, such as works of literature or music, we
grant copyright protection that lasts much longer than patent protection.
This can be rationalized by the argument that costs from monopoly
control of these goods creates relatively little economic inefficiency.
For other goods, such as scientific discoveries and mathematical
formulas, the law gives no protection at all. This presumably reflects
a judgment that the cost of monopoly power over these goods is too
high and that we are better off relying on such nonmarket mechanisms
as philanthropic giving and government support to finance and motivate
the production of these types of software.
distinction between different types of software is the difference
in the amount and variety of additional work that needs to be done
before that software makes an actual contribution that consumers
would be willing to pay for. Property rights on software that is
directly employed by final consumers can lead to high prices, consider
the high prices on some pharmaceuticals, and cut out use by some
parties who would value use, but will not or cannot pay the price.
For software such as this, however, that is close to final use,
it is possible for users to make reasonably well founded benefit-price
It is quite
otherwise with software whose major use is to facilitate the development
of subsequent software. Any market for software, such as mathematical
algorithms and scientific discoveries far removed from the final
consumer, would risk being grossly inefficient. Over time, many
producers have to intervene, making improvements and refining the
basic idea, before such software can be finally embodied in a technique,
practice, or design that produces value and is sold to a final consumer.
Economic theory tells us that the presence of monopoly power at
many stages in this long and unpredictable chain of production can
be very bad for efficiency.
In the worst
case, property rights that are too strong could preempt the development
of entire areas of new software. In the computer software industry,
people capture this dilemma by asking the rhetorical question, "What
if someone had been able to patent the blinking cursor?" The
point applies equally well to many other important discoveries in
the history of the industry, the notion of a high-level language
and a compiler, the iterative loop, the conditional branch point,
or a spreadsheet-like display of columns and rows. Extremely strong
property rights on these kinds of software could have significantly
slowed innovation in computer software and kept many types of existing
applications from being developed.
In the production
of computer software, basic software concepts are not granted strong
property rights. Software applications, the kind of software sold
in shrink-wrapped boxes in computer stores, is protected. This suggests
a simple dichotomy between concepts and final applications that
mirrors the distinction noted in the beginning between the search
for basic concepts by a Niels Bohr and the search for practical
applications by a Thomas Edison. As the work of Pasteur would lead
us to expect, this dichotomy hides important ambiguities that arise
in practice. At the extremes, the distinction between concepts and
applications is clear, but in the middle ground there is no sharp
dividing line. Courts are forces to decide either that software
for overlapping windows or specific key sequences should be treated
as essential parts of an application that are entitled to patent
or copyright protections, or that they are basic concepts that are
not given legal protection. In the realm of software, there are
many shades of gray. The simple dichotomy nevertheless serves as
a useful framework for guiding the economic and policy analysis
of science and technology, for science is concerned with basic concepts,
and technology is ultimately all about applications.
is excerpted from a longer article entitled Science, Economic Growth,
and Public Policy, which appears in the March-April 1996 issue of
Challenge. It is also part of a forthcoming book by the authors.