3.0 Innovation

• 3.1 Definitions
• 3.2 Examples of Innovation
• 3.3 Incentives
• 3.4 Interactions

Innovation should be defined using a bounded-rational model of goal-directed behavior. If we assume that political economic agents are rational, consumers maximize their utility in selecting a preferred bundle of goods subject to their budget constraint and producers maximize their profits. Under rationality, any change an agent makes is an innovation by definition because if political economic agents optimize, they do not make mistakes in a deterministic model or in the expected value sense in a stochastic model. In the past several decades research by behavioral psychologists and some economists have shown that human behavior is suboptimal, a condition for which Simon has coined the phrase bounded rational. We first define innovation using a bounded-rational model of behavior. We do this because in such a model change is not necessarily an innovation because bounded rational economic agents can make mistakes. Next, we discuss examples of innovation in goal-directed behavior. Then, we discuss the incentives for innovation. Finally, we discuss the interactions between discovery, invention, and innovation.

3.1 Definition
A simple definition of innovation is a better way of doing things. Individuals and institutions innovate in all their goal-directed behavior that is defined as an effort of an individual or an institution at achieving performance as measured by a criterion, whether objective or subjective. An example of objective goal-directed behavior is the effort of a firm to maximize profits. Another example would be the striving of a politician to obtain reelection. A third example, based on a subjective criterion, is the effort of a consumer to select a purchase most desirable as defined by his or her tastes. Thus, goal-directed behavior and consequently, innovation encompass a broad range of economic, political and social behavior. With respect to such goal-directed behavior, a formal definition of an innovation is the creation or implementation of a new alternative which achieves higher performance as measured by the respective criterion. In discussions of alternatives we will focus much of our attention on the choice between alternative procedures. Thus defined, innovation is a generalization of Schumpeter's definition of innovation for a firm.

Let us now consider an example of innovation in profit maximization. A manager of an existing product line might consider producing the product line at the firm's current plant, contracting the product line to be produced in the Far East, or constructing a new plant which would implement new automation technology. If the manager implements the new automated plant and achieves higher profits, the implementation of this alternative would constitute an innovation.

Since the decision to develop a successful invention by a firm is the selection of an alternative which improves profit-striving performance, inventive activity by firms can be considered a special category of innovation. A point made throughout the book is that invention is quite different from other categories of innovation. For example, property rights for invention differ markedly from property rights for other types of innovation. To keep the terminology as simple as possible, innovation, in all subsequent discussion, will refer to all categories of innovation other than invention. The media, such as BusinessWeek, should use the term innovative firm to indicate a firm that rapidly advances the technology of its products. Under our terminology such a firm would be called inventive.

A discussion of the factors that promote innovation by individuals or groups requires a more detailed description of innovation. Because innovation has been defined as the selection or creation of a new alternative that leads to improved performance in a goal directed behavior, such a description requires a model of goal-directed behavior. In this book the model of goal-directed behavior will combine aspects from bounded rationality and cognitive decision theories^9. Also, the model will be based on the empirical fact that decision-makers expend resources in making decisions.

A basic tenet of these theories of behavior is that humans have limited cognitive skills. Let us for now consider human decision unaided by information technology. Most aspects of purposeful behavior are intuitive. Moreover, intuition, is largely a product of the individual's personal history. Individuals perceive selectively, based on prior experience, and given their limited ability to integrate facts, they are sequential problem solvers. To solve a complex problem, then, they must simplify the problem by using heuristics, that is generalizations which enable them to construct a simple model of behavior. In constructing such a model which accounts for prior experience, individuals must mentally reconstruct prior events.

Numerous factors limit the goal-directed performance. Psychological research has cataloged numerous defects in individuals' intuitive judgment and choice processes¹¹. With respect to innovation, judgment is required in evaluating the alternatives. For most tasks individuals must forecast the future behavioral consequences of selecting an alternative. These forecasts are generally intuitive judgments or estimates. Human judgment, however, is dependent upon the order in which data are presented. Confronted with apparently complete data presentations, humans will tend to overlook critical data omissions. Similarly, too much data can overload human mental capacities, thereby reducing consistency of judgment. Also, humans are poor at making nonlinear forecasts such as predicting the social implications of the advances in computation.

In any human data acquisition and processing of a large number of alternatives we assume humans are sublinear processors in the sense that human data acquisition and processing is so slow that it is prohibitively expensive for a human to apply even the simplest linear operator such as a single observation per alternative for a large set such as 250,000. The same is true for even a simple operation such as addition on a large amount of data. Prior to the rise of information technology humans developed numerous approaches to process large numbers of alternatives using sublinear processes. We shall call this conjecture the sublinear hypothesis.

Consider, for example, the economic model of a consumer maximizing his utility in purchasing a bundle of goods. First the consumer purchases goods sequentially to reduce the number of alternatives to be considered. For example, suppose a consumer actually considered purchasing goods in bundles and not sequentially. If the bundle had only two goods and there were 10 alternatives for the first good and 10 alternatives for the second good, the sequential approach has 20 alternatives to consider whereas the bundle approach has 100. That is the sequential approach is additive while the bundle approach is multiplicative. If we consider a bundle of ten goods with ten alternatives each, considering all possible bundles involves 10¹⁰ = 10,000,000,000 alternatives, whereas the sequential approach involves only 10 x 10 = 100 alternatives. Second, suppose Barnes and Noble Book Store with over 250,000 books simply piled all the books in a big heap so that each customer had to look through the pile to find the book he or she wanted. If the process only took a second per book, on average each customer would expend almost 35 hours to find a desired book. In order to facilitate the customer using a sublinear process, Barnes and Noble organizes the books into categories with a map showing their location. Thus, the customer can immediately select the desired subset and use more complex selection processes on a small number of alternatives. Stores in general facilitate customers using sublinear decision rules to narrow their search to a small number of alternatives.

Humans have considerable ability to recognize patterns that enable them to eliminate sets of alternatives without expending energy. For example, human grand masters in chess have superior skills in recognizing patterns of pieces on the board that enables them to focus on a small number of alternatives for the next move. In contrast computer chess processes all of the extremely large number of moves that can occur when all move possibilities are considered.

Another approach of decision makers to reduce their effort is to have subordinates prune the list of alternatives such as potential employees or candidates for admission to law school such that they can focus on the subset that requires their attention. In addition, aggregation reduces the amount of effort to evaluate alternatives. In large corporations or government bureaucracies, data is aggregated as it proceeds up the hierarchy. This reduces the effort of senior managers in analyzing performance.

Another heuristic to reduce a complex task involving a large number of alternatives to a subset is to seek the satisficing rather than the optimal alternative¹⁰. Simply put, an individual faced with a large number of alternatives will seek a good, but not necessarily the best, alternative. Once he or she finds the good alternative the individual ignores the alternatives not already processed.

Moreover, most judgment situations involve uncertain phenomena for which man lacks the neurological circuits for optimal intuitive processing. One example is that humans do not always have transitive preferences. Another defect is handling uncertain phenomena, such as man tends to assume erroneously that the characteristics of a small sample are characteristic of the general population. Furthermore, in combining uncertain data, humans tend to favor the concrete over the abstract. Also, in judgment situations where they should combine new data with old data, humans will frequently ignore the old data. When they do combine the two, they are generally more conservative than optimal.

To illustrate the problems individuals have in combining uncertain data, consider the problem of forming a judgment concerning the reliability of a new automobile. If a friend has recently bought a model which turned out to be a lemon, this concrete data will tend to be given much more weight than a previously conducted study indicating that the model, on average, is reliable. The individual may, in fact, ignore the statistical study altogether. To illustrate the conservative nature of data on judgment consider the forecast that in the 90's US automobiles should become as reliable as the Japanese automobiles. But because Americans are likely to rely more on experience than statistical studies, there will be a time lag before the public acquires the general perception that US automobiles are as reliable as the Japanese.

In addition to affecting human judgment, cognitive limitations also affect the ability of individuals to make choices. Two aspects of making choices in goal-directed behavior are generating the objective and selecting among alternatives Individuals must constantly adjust the objectives of their tasks to new conditions in a political economy with a rapid rate of discovery, invention, and innovation. The basic assumption underlying these adjustments is that the objectives generated by individuals are, in fact, only estimates of the true objectives. Moreover, experimental evidence demonstrates that individuals do not form consistent objectives¹² and tend to confound their beliefs about behavior with their preferences or tastes. In making choices man uses rules of thumb-- simple common sense rules to select alternatives--to reduce the mental effort.

Given the human limitations, the question becomes whether or not human performance ever approximates the optimal performance suggested by rational man theories of behavior. Tasks which are repeated offer individuals and organizations an opportunity to improve their performance. Competition provides strong incentives to constantly improve one's performance, but competition in the sense of the survival of the fittest only guarantees that a successful firm keep more efficient than its rivals not that it achieves a true optimum. We assume that the higher the rate of change in terms of technological advance and the more variations in prices the further individuals and organizations will be from their optimal solutions.

Second, we assume that there is a tremendous variation in the complexity of tasks to be solved in goal-directed behavior. For example, consider the problem of scheduling. If a firm makes a single product for a customer with a steady demand, the scheduling problem is simple to solve. But, scheduling in its most general case of multiple products, multiple machines, and multiple variable demands is an NP problem that is not tractable even with computers. The human heuristic solution prior to computers to the problem of scheduling was to produce for inventory. The latest scheduling algorithms that produce good approximations to difficult scheduling problems are genetic algorithms. In the case of difficult scheduling problems, humans without computers will not converge to a good solution in that inventory brings in no return until it is sold, and even with computers humans will only approximate the optimal solution.

At any particular time human performance can be far from optimal. Consider the consumer. Because it is efficient to purchase goods sequentially, consumers must solve a budget allocation problem over time. Given the number of consumers who have problems with credit card debt, learning to budget over time is a difficult problem to solve. Also, consumers can make major mistakes in individual purchases because determining the attributes of products in the marketplace can be very expensive. Recently, it became public knowledge that some brands of latex paint contain mercury and arsenic and that children in houses and apartments painted with such paint suffered harm. Yet because of the cost involved, individual consumers almost never consider performing chemical analysis of the products they buy, assuming, sometimes erroneously, that they will be protected by government agencies or the product liability laws. Also, humans can make errors with dire consequences in judgment in predicting future behavior involving such factors as reliability and safety. For example, consumers who bought three wheeled recreational vehicles have tragically lost children in accidents when these vehicles turned over.

Firms and governments are also prone to serious errors in goal-directed behavior even though institutions generally expend much greater resources in decision-making than individuals do. For example, in the 80s General Motors tried to automate their plants ahead of the state of the art and were forced to back off with their Saturn plant. Ford and Chrysler were more successful with more modest automation. Firms have, on occasion, introduced duds into the marketplace such as Ford's introduction of the Edsel or IBM's introduction of the PCjr to the home computer market. Government legislation is usually based on some theory of behavior. When theories change, government policies are frequently wrong in relationship to the new theories. For example, much 30s legislation was to prevent competition from bankrupting existing firms. An example is regulation Q that curtailed competition among banks. After WWII the world economy became increasingly competitive making this 30s anticompetitive legislation a drag on the economy because regulation Q was a a factor in each recession until repealed in Reagan's administration in 1980.

Because both individuals and institutions are prone to making serious errors in decision-making, what constitutes an innovation is far from obvious. Innovation is not synonymous with the selection of a new alternative. To determine what constitutes an innovation requires a procedure for judging whether the selection of a new alternative improves performance. The problem is to determine the objectives for which performance is to be measured. Because individuals, firms and the government face constant technological change, their respective generated objectives are assumed to be only estimates of their true objectives. This creates a problem because innovation should be defined as the selection or creation of a new alternative which improves the performance of a goal-directed task as measured against the true, but unknown objective.

Let us start to solve this problem by considering the properties of the true, but unknown objectives. Because true objectives can not be precisely defined, criteria are needed by which to judge whether the creation and implementation of a new alternative by an individual, firm, or government is an innovation. It is assumed that the true, but unknown objective of a firm should be to maximize profits. While this may appear objective and clearly defined, an obvious issue is to what extent should a firm emphasize short term profits in relationship to long term gains. Also, how the profit motive for the entire firm should be transcribed to the objectives of subordinate units within the firm is not a simple problem. For the firm, the selection of a new alternative can be considered an innovation if the change increases profits. In a competitive market, then changes which persist and are imitated can be considered innovations.

Also, a long time and a great deal of discovery and invention may be required to make a change an innovation. First let us consider the innovation of replaceable parts. This idea was possibly first employed by Eli Terry in the manufacture of wooden clocks around 1800. Eli Whitney, who is generally credited with the idea, put in a great deal of effort improving the accuracy of machine tools to make this innovation successful. Another innovation is the use of the electric motor which was first introduced in the 19th century. But it was not until the 1920s that the use of electric motors could be observed to improve productivity. A more recent example is the use of desktop computers in the workplace in the 80s. It has taken numerous advances in software and communications to make the use of desktop computers an innovation currently.

The true, but unknown objective of government will be given the archaic term, ``to promote the common weal'', in order to distinguish the concept from contemporary social welfare theory based on utility theory²³. Social welfare theory has provided results of limited usefulness for social system design. Arrow has demonstrated that it is impossible to create a social welfare function from individual preferences that satisfies a small number of reasonable assumptions. Also, the Pareto criterion for evaluating alternative social alternatives is of no practical usefulness in that there are always losers and winners in the move from one social state to another.

We shall instead focus our attention on a different political problem. The task of government to promote the common weal involves the responsibility to deal with external threats, maintain internal order, promote prosperity and ensure the quality of life. No agreement, however, exists in identifying the basic elements of the common weal or in estimating the tradeoffs between them. What we will focus on is having policies that can be evaluated by measurable criterion. In general, the selection of a new alternative by government can be considered an innovation if the new alternative has the properties of consistency, efficiency, and general benefits. If the interested reader wishes, he or she could consider the impact of alternative criteria on the design. The lower the level of government, the more competition there is among governments competing for resources such as firms investing in new plant and equipment in their domain. In these activities change will be considered an innovation if it is copied and persists.

In the case of the individual even less can be stipulated. It can only be said, at this point, that the true, but unknown objective of individuals is not considered to be equivalent to the desires of the individual. For example, the desire for drugs by a drug addict is not in the drug addicts long term interests. Change in individual tasks can be considered an innovation if it persists, is imitated and does not diminish the common weal. The final requirement of not diminishing the common weal is needed to eliminate pathological behavior such as drug addiction.

Let us summarize our theory of behavior to emphasize the most important points for subsequent reference:

• Man is an approximate optimizer who can make serious errors
  • Performance improves in repeated tasks, but does not necessarily converge to optimal performance
  • Performance is problem specific to the difficulty of the task
    • In dealing with large numbers of alternatives humans use sublinear procedures
• Change is not necessarily an innovation
  • Criteria are needed to distinguish between change and innovation
  • A great deal of discovery and invention and a considerable effort over a long time may be required to make a change into an innovation.

3.2 Examples of Innovation

Now having defined criteria for determining what is an innovation, let us consider some specific cases. Let us start by considering an example of change that was not an innovation. The creation of an organizational structure with fourteen levels of management at General Motors as compared with five at Toyota was not an innovation¹³. Currently, General Motors is struggling to achieve a more horizontal organization.

Let us now consider a few examples of innovation by firms in the task of production. Two important innovations in the 19th century were the concept of the assembly line and interchangeable parts. To recent innovations in production have been the reduction in work in progress through just in time and MRP inventory control and the application of statistics to reduce defects in products. Not every change in production is an innovation. General Motors in the 80s tried to advance automation of their assembly plants beyond the state of the art and was forced to back down in the creation of the plant to produce Saturns.

Because tasks have a hierarchical structure, innovation can also take place in the decision-making process and in the execution of a task. The model of goal-directed behavior implies that each of the steps in a task can, in itself, be considered a subordinate task. Thus, an innovation in the task of production can be a new, superior way to determine alternatives. For example, computer manufactures can now use PCOrder¹⁴ to obtain a complete online organization of the components to build PCs, complete with current prices. Using this software is an innovation because it greatly reduces the time and cost of evaluating alternatives and purchasing the desired components. In general, innovations in the first four steps of a task can be classified as innovations in the decision-making process, and innovations in the fifth step can be described as an innovation in the execution of a task.

Innovation in the decision-making process can also occur in estimating the objective. A problem in large firms is creating objectives for subordinate units in the firm so that the performance of the subordinate units best contributes to the profit maximization the firm. One recent example in this regard is total quality management, TQM, which focuses subordinate units is providing good service to customers. Because TQM has been widely imitated in competitive industries, this change qualifies as an innovation.

Innovations which improve all aspects of task performance are frequently new forms of organization and incentive systems. The original form of the manufacturing corporation was called a functional organization. Coordination of the firm took place between the president and the vice presidents in charge of manufacturing, finance and other functions of the firm. Managers below the level of the vice presidents were specialists such as accounting managers. As manufacturing firms increasingly began to manufacture multiple products this form of organization exhibited a growing defect--that managers were more sharply focused on their functional specialties than on the profits of the various products and product lines. The resulting shift from functional organization to profit-oriented divisional organization in the 1920's was a change which more sharply focused managerial attention on the profits of each product by coupling managerial rewards to the profits of their respective products.¹⁵. Since profit-oriented divisional organization has persisted and has been widely imitated, this change is an innovation.

In the realm of government, an example of innovation in the task of distributing social security checks is direct deposit into the bank accounts of the recipients for the sake of convenience and security. A new tax loophole, however, would not be considered an innovation since such a loophole would lack general benefits. Innovation in the decision-making process can also occur generating the governments estimate of the true, but unknown objective. For example, the framers of the constitution incorporated checks and balances into the governmental design in order to achieve a good estimate of the common weal. Subsequently, in the 20th century the political reforms, such as direct election of senators, campaign finance laws, and the principle of one-man one-vote, have sought to make the political process more representative. These changes have had a profound effect of the estimate of the common weal. Many social observers consider these changes innovations²⁶.

An example of individual innovation in the task of working is telecommuting by professionals, whereby professionals perform their work from their homes or local offices through a computer communications network instead of having to travel to the central workplace²⁵. Another individual innovation in the task of maintaining a financial assets portfolio is the use of personal computers together with information services, such as the Dow Jones Information Service, to identify, evaluate, and buy and sell alternative financial assets. The emergence of a lifestyle based on smoking crack would not be considered an innovation, however, as the common weal is diminished by the loss of productive capability and the increased likelihood of crime and child neglect.

3.3 Incentives

The incentives for innovation are neither fame nor intellectual property. The incentives for innovation are simply achieving better performance in a goal-directed activity. A firm in a competitive market has strong incentives to constantly innovate to maintain a competitive edge over its rivals.

Currently, invention is promoted by intellectual property, whereas most aspects of innovation are not. Moreover, it is doubtful that an intellectual property right in the form of a patent could be created for innovation, since innovation is not a precisely defined object but rather the implementation of a concept by an individual, firm, or government. Frequently, each implementation of the concept is specific to the implementer. Moreover, advances in innovation are generally incremental. This means that few innovations would have such clear definition and originality to merit the equivalent of a patent.

The fact that innovation is not protected by property rights means that the motivation for innovation is to improve the performance of a goal-directed task itself rather than to sell innovation as a product on the market. Because innovation lacks intellectual property rights, any success by an innovator will be rapidly imitated¹⁶ in a competitive market. Nevertheless, for some types of innovation, the innovator has positive incentives to promote imitation. For example, innovations by government are imitated by other governments at the same level, and officials responsible for the innovation enhance their reputations by providing the details to interested imitators. A business innovator can obtain financial rewards for promoting imitation of an innovation by establishing a consulting service or by writing a how-to book.

Also, because innovation lacks intellectual property rights, businessmen have mixed motives concerning the dissemination of innovation. Firms generally prefer to maintain secrecy about innovations which give them an advantage over their rivals, unless, of course, there are considerable profits to be made from creating a consulting service to sell their expertise in creating the innovation. On the other hand, firms have a great incentive to disseminate knowledge about innovations based on inventions they are selling, and the extension of copyrights to software provides similar incentives for software firms to disseminate information about innovations based on their software.

We shall assume the government officials enjoy the exercise of power and that they generally have strong incentives to get reelected. For lower levels of government that are in fierce competition with other governments to promote growth and prosperity in their domains, reelection incentives generally coincide with innovation. The higher the level of government the less competition among governments is a factor. At the federal level of government we assert it is the design of the system of checks and balances is what couples the desire to get reelected to innovation.

Innovation among individuals is motivated by better performance and the rewards of being an innovator.

3.4 Interactions

Innovation interacts with theoretical discovery and a profitable invention always requires an innovation. However, most important is the applied discovery needed to improve the performance of a new alternative to the point where it becomes an innovation. The efficiency of this learning process depends on the learning strategy.

Innovation, like invention, can involve a two-way relationship with theoretical discovery. One example of a theoretical discovery promoting innovation is the development of the theory of statistical quality control. The applications of this theory to production have led to a significant decrease in the number of rejects. Innovation can also promote theoretical discoveries. The series of innovations in the automation of manufacturing is stimulating all types of discoveries in machine intelligence.

Generally, an innovation is required in order to make an invention profitable. The invention of spreadsheet software, for instance, became profitable as businessmen innovated a large number of applications for spreadsheet analysis. But, an innovation does not necessarily need to be based on an invention. For example, the development of the assemblyline in manufacturing is an innovation which is based on the organization of production rather than a specific invention.

For an invention to become profitable, at the very least its use must be an innovation in the sense that substituting the new invention for the older product leads to greater performance in some goal-directed activity. More often, the use of an inventions frequently spawns a series of innovations whose success involves the discovery of new skills. Consider, for instance, the innovations in processing paper in modern offices. The innovation of wordprocessing as a displacement of typing required the development of wordprocessing skills and their dissemination through instruction manuals and training courses. The diffusion of stand alone wordprocessors set the stage for numerous inventions such as storage devices, laser printers, and local area networks to improve the flow of paperwork in the office. In addition, new inventions linking the office equipment together electronically promoted numerous innovations such as electronic filing of documents and electronic mail. New software such as desktop publishing programs generate new skills in creating documents, and these skills in turn become the basis for a round of innovations in the use of desktop publishing in business.

Innovation, like invention, frequently requires a large increase in knowledge to achieve better performance. For example, a plant based on new technology can require several years learning by doing before it become profitable. This learning can involve formal discovery in engineering in making revisions to the plant and surface skills as the workforce learn numerous new skills related to the new technology.

The efficiency of learning to achieve better performance in an innovation is a function of methodology and organization. Improving the organization of innovation is frequently much more difficult than improving the organization of invention. In almost all cases, firms have separated inventive activity from production; however, in many cases it is difficult to separate innovation from the other activities of the firm. For example, even General Motors does not have sufficient resources to establish automobile assembly plants as research and development facilities for pure experimentation in production processes. New assembly plants with new innovations in production are run to make a profit. The same is true in major new organization forms for corporations. The firm shifts from the old to the new and then struggles to maximize profits with the new organization.

Generally, discovery for innovation takes place in a very different learning environment than discovery for invention. In most invention situations, inventing is a separate task in which the inventor can usually conduct experiments to perfect a design. In contrast, the cost of experimentation in innovation is so great that the innovator can generally try only a single alternative. Moreover, applied discovery to perfect an innovation is rarely a separate task, but rather this applied discovery is subordinate to the goal-seeking behavior. For a formal perspective learning when an activity can be isolated is an example of the design of experiments. When learning must take place at the same time as profit maximization the learning model is known as estimation and control. In estimation and control learning is subordinate to profit maximization and in most cases is likely to take place at a slower rate than in the model of design of experiments.

To provide a simple example illustrating the difference consider the following two cases.

1. Pure learning (Invention)¹⁷ : Suppose an investigator wants to learn the value of a in the following model: y_t = ax_t + e_t for t = 1,2,...,T where e_t is a uniform distribution and a is an unknown coefficient. If the investigator can arbitrarily set the value of x_t in repeated experiments, the larger the value that he or she sets x_t, the quicker the investigator learns the value of a. This example corresponds to a situation where the investigator can separate invention from the other activities of the firm. The investigator can design the most efficient experiment to learning based on cost-benefit considerations to learn a relationship needed in the inventive process .

2. Estimation and control (Innovation)¹⁸: Suppose an investigator wants to minimize the expected value of the sum of the square deviations of y_t from B as follows: minE[Sum_t(B - y_t)²] subject to y_t = ax_t + e_t where t = 1,2,...,T where e_t is a uniform distribution and a is an unknown coefficient for which the investigator has a prior belief as to its value. The investigator can not set x_t arbitrarily large in order to efficiently learn the value of a because he or she will incur very large costs if y_t deviates from B. How much should the investigator deviate from best current performance in order to improve future performance is the focus of the estimation and control literature. This model corresponds to the usual situation in innovation where a firm with a new organization or incentive system is limited in its ability to experiment from profit maximization considerations. (B represents the behavior that maximizes profits and any deviation from this behavior reduces profits) Consequently, the rate of learning needed to improve the performance of the change to make it an innovation is reduced. The first economist to recognize that innovation involved solving difficult problems was F. Knight, whose work can be made consistent with Basyesian statistics.

Estimation and control problems are mathematically intractable even in very simple models. Unlike pure experimental design, there is no extensive literature of estimation and control models to be applied in real world situations. Let us consider the simplest strategy for innovation learning an improvisatory strategy that means that the innovators intuitively improvise as necessary to achieve better performance. To the extent that theoretical knowledge can be applied to eliminate unpromising ideas, the learning process is improved. Also, to the extent the good scientific experimental methodology based on an estimation and control model can be employed, the learning process is improved.

Consider, for instance, the innovation in automation in manufacturing. In the 1980s General Motors built a sequence of new and renovated plants such as Hamtramck, Saginaw Vanguard, and Saturn. Each plant was a pilot project for a new technological advance such as MAP, machine automation protocol, a specification to link machines in manufacturing¹⁹. Furthermore, these plants have been operated to produce products for sale not simply as laboratories to test automation advances.

The fact that much innovation is based upon a single alternative without experimentation creates a problem: how can the innovator perform the applied discovery necessary to perfect the implementation of the new alternative? What makes applied discovery possible at all is that individuals, firms, and government have much greater incentives to innovate in repetitious tasks than unique tasks. Consequently, applied discovery takes place each time the new alternative is employed to perform the task. This basic strategy for innovation used by most innovators will be defined an improvisatory strategy.

With this strategy the innovator uses this acquired knowledge to revise the new alternative to improve the performance. Without the constant stream of revisions the performance might not reach the level of efficiency necessary for it to be classified as an innovation. Indeed, in the GM example above, GM seriously underestimated the amount of applied discovery needed to make the advances in automation an innovation and scaled back the technological advances implemented in the Saturn plant.

From the perspective of the previously discussed task model of goal-directed behavior, an innovator can make applied discoveries in all aspects of the hierarchy of tasks subordinate to the new alternative. For a concrete example, consider the implementation of a new, more automated plant to assemble a product. The process of achieving improved performance through discovery and revision is variously called the experience curve or learning by doing. Specifically, performance improves as workers sharpen their skills at performing their tasks in the new plant. Design defects are discovered and corrected. Bottlenecks, for example, in the delivery of parts are discovered and remedied. In Japan, this learning by doing has been institutionalized into quality circles.

While individuals do indeed improve their performance by learning by doing and imitation, man is not an optimal learner²⁰. Besides his previously mentioned limitations in combining data, man tends to make spurious correlations and to seek data to confirm hypotheses rather than to test them. Given the limitations of man's cognitive capabilities, the prospect of achieving near-optimal performance would appear slight. For repetitive production tasks with a given technology under competitive conditions, however, man appears to have the learning capacity to approximate optimal behavior, given sufficient time²¹.

The rate at which performance in a repeated task improves depends on the strategy for implementing an innovation. With the improvisatory strategy most learning takes place through experience or through the need to solve obvious problems. Innovators perceive little need for systematic variation of variables to obtain empirical discoveries that improve performance. Performance can be improved in many cases by the adoption of an innovation-implementation strategy which incorporates a more systematic approach to learning.

To illustrate the limitations of the improvisatory strategy, we may suppose an innovator uses a cutoff, such as a minimum score on a test or a minimum grade point average as a condition of employment at his plant. As a consequence, the innovator will never observe the performance of potential employees below that cutoff and hence cannot draw any conclusions concerning the validity of the cutoff²². With an improvisatory strategy, this cutoff remains fixed until the innovator perceives the cutoff to be a problem. With no policy for systematic variation of such a variable, the innovator cannot develop a systematic innovation-implementation strategy to improve the performance of the chosen alternative.

Another factor which influences the rate of applied discovery in implementing an innovation is the number of imitators involved. With competition any new alternative which produces much better results than current practice will be imitated immediately. Because imitators are likely to evaluate the factors contributing to the success of an innovation differently, imitators are likely to make numerous modifications in imitating an innovator. The more competing imitators generate greater variation, the more empirical knowledge is gained contributing to refinements to a successful innovation.

The creation of a more systematic approach to learning for innovation is far from simple. With a single implementation of an alternative, most aspects of innovation do not constitute a large enough sample for a statistical design. Moreover, variation through imitation is not controlled, and variation in variables over time may require too long a time horizon to be considered worth the effort. Finally, mathematical models of learning subordinate to performance are very difficult to solve. This means that the innovator mentioned above has no simple way of determining how hiring a sample of workers below the cutoff will improve the performance of his chosen alternative. Because systematic variation presents so many problems, it is assumed in this book that generally the improvisatory strategy, even with imitation, introduces far too little variation for optimal innovation performance.

Nevertheless, the strategy for implementing an innovation can be improved in some circumstances. In tasks common to a large number of individuals or institutions with similar goals, there are economies of scale²³ in specializing the learning activity. This strategy will be called a separation strategy. In other words, learning for innovation, like invention, becomes a separate task. Where learning is a separate activity, performance in learning can be achieved through improved methodologies such as statistical procedures and experimental designs. Once the performance of an innovation has been established through experimentation, the innovation will be adopted to improve performance. This strategy is used in agriculture in advanced nations, where research stations experiment with new types of seeds and production techniques and transmit the results to the farmers.


Overall in spite of the difficulties, the activities of invention and innovation have a multiplier effect on knowledge accumulation. As was previously mentioned, the drive to obtain the full economic impact of an invention stimulates discovery. In addition, by definition an invention requires an innovation to become profitable. This innovation generally includes a significant increase in practical knowledge about how to produce, use, and repair the new invention. But, as a new technology displaces an old technology some of the practical knowledge of the old technology is lost.

This knowledge multiplier can be very large, as inventions and innovations frequently have a cascading effect. For example, industrialization started with inventions in textile machinery and innovations in textile production. Subsequently, by attracting a large number of workers to a central location, industrialization created an urban workforce. By extension, industrialization generated innovations both in lifestyles and in city government in the sense that it created the need for providing public services such as clean water. Fully realizing the social implications and economic advances of a discovery in basic knowledge, then, generally requires a major increase in practical knowledge.