NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

National Research Council (US) Chemical Sciences Roundtable. Reducing the Time from Basic Research to Innovation in the Chemical Sciences: A Workshop Report to the Chemical Sciences Roundtable. Washington (DC): National Academies Press (US); 2003.

Cover of Reducing the Time from Basic Research to Innovation in the Chemical Sciences

Reducing the Time from Basic Research to Innovation in the Chemical Sciences: A Workshop Report to the Chemical Sciences Roundtable.

Show details

10The Tacit Economics of Modeling: Indifference Curves that Should Defy Indifference

Michael Schrage1

Massachusetts Institute of Technology

The United States has had an explosion of innovation opportunities. This “opportunity glut” creates a need to explore both the meaning of innovation and the role it plays in letting firms profitably differentiate themselves in the marketplace.

As someone trained in economics, I have always been struck by organizations that had perfectly good, rational tools for getting the job done but that continued to act irrationally, counterproductively, and seemingly inexplicably. What legitimate reasons could economists develop to explain that organizations often ignore good technical solutions to their problems?

Assuming that an innovation opportunity can be modeled in some useful and meaningful way, I am interested in exploring how people behave with respect to the economics of and the tradeoffs associated with modeling. My particular emphasis is on innovation behavior, looking not at how people think but at how people behave. Actions speak louder than words. Actual behavior is more eloquent and revealing than rigorous analysis.

Explicitly, I look at how people interact around iterations of representations—or, in plain English, how people behave around versions of models. That behavior is the essence of innovation. This begs a simple proposition: when we transform the economics of modeling, prototyping, and simulation, we inherently transform the economics of innovation. Make modeling faster, cheaper, and easier and we surely change the economics of innovation. We incent different iterative and innovative behaviors.

To appreciate the real impact of this transformation, we need to annihilate some of the myths that the information technology domain has inflicted on us. We need to avoid remaining victims of a “data-driven” vocabulary.

The first myth I want to expose is what I call the Big Lie of the Information Age, which is that as we change the quantity and quality of people's information, we change the quantity and quality of people's behavior. As reality attests, that is demonstrably not true. In fact, if we change the information in organizations, we oftentimes do not change people's behavior, because people frequently ignore or dismiss the information given them.

Consider an informal gedanken experiment: I asked people to choose between a tool that offers an order of magnitude—10×!—improvement in managing all the information that goes across the desktop, phone, personal data assistant, cell phone, and Web, or a tool that offers a 20 percent—0.2×—improvement in the ability to persuade one's bosses, colleagues, and subordinates. The overwhelming majority of people consistently chose the persuasive tool. So what is the real issue most people face in their organizations? Information or persuasion? The role and rhetoric of models and simulations for persuasion in organizations have very different design emphases and sensibilities compared to the models for informing organizations.

So the common reaction is to question how to develop models, prototypes, and simulations that make a person or his group more persuasive within the organization. That is a legitimate design question. Using myself as a beta site of one, I realize I don't like being persuaded by others. However, I consider myself open minded enough that I am happy to persuade myself. The design challenge? What is a better investment? A tool that helps a person become more persuasive or a tool that helps that person's customer, internal or external, persuade themselves.

I submit that building models that enable people to persuade themselves is a different design sensibility challenge from the designing of a model, prototype, or simulation that makes a person or group more persuasive to others. Those kinds of design parameters become more important rather than less important when we talk about accelerating the pace of innovation within an organization, industry, or market segment.

Why are these kinds of design questions more important now than even 5 years ago? We are rich. We have computational wealth and power that transform both the economics and rhetoric of modeling. However, I fear that we are using the wrong unit of analysis for the assessment and measurement of computation-driven influence. People talk about Information Technology and the Information Age, but I think that modeling, simulation, and prototyping of innovation are, frankly, not an information management problem.

Instead of “Bits” management being the issue here, the real source of wealth is “Its”—Iterations. Innovation wealth is a function of a shift from Bits to Its. Instead of managing information, how do we better manage iteration? The new wealth is our ability to iterate and perform more iterations per unit time. What do we do with that as individuals? What do we do with that as teams? How do we use the opportunity to manage more iterations per unit time as a vehicle to reduce coordination costs and transactions costs? How do we create these technologies as a vehicle to facilitate communication of innovation and management of innovation within organizations? The answers to these questions determine how well—and how poorly—organizations will iterate to innovate.

Please make a conceptual leap with me. We know what financial capital is; we understand and appreciate human capital. We hear more and more about social capital. I would like you to think of the explosion of computation-driven iteration to be a form of Capital. Iterative Capital.

So we need to ask ourselves what is our ROI—not Return on Investment but Return on Iteration. What do we want to accomplish as innovators? What kinds of attributes are we iterating around? What is it that we are really trying to learn as we iterate to innovate? Are we interested in the development of a particular structure or material or that structure or material in a certain kind of a context? How do we manage the return on iterations? Ultimately, the more choices you have, the more your values matter.

The Capital Asset Pricing Model and other financial theories around diversification offer useful insights. If I give you a million dollars, you are a fool to invest in venture capital. But if you have a billion dollars, you are a fool if you don't invest in venture capital.

As we transform the costs of iterations, we have to transform our investment profile. Being wealthy means you have more choices. If you have more choices, your values matter more. In the end you have a portfolio of iterations. How do we want to manage those iterations?

All the previous information makes sense on a “rational” analysis basis, but as we know, for every buyer in the stock market, there is a seller. These individuals and institutions have different risk profiles—they have different expectations of the future.

We have to look at real-world positive behavior as opposed to normative behavior. For organizations the true test of a model is not how well it works, but how well it is used. There are a number of very good reasons why organizations behave very irrationally with respect to their investment in models. There are three examples of this poor behavior: the golden goose, the magic mirror, and the stone soup.

Chemists are inevitably asked by organizations for better models for problem solving. In theory all employees in an organization understand the problem and are therefore capable of building better models that produce better answers. In reality everyone does not understand the problem and its underlying issues, so a model can be improved in two ways: by changing it to produce a better answer, and by changing it so that the model is more accessible to all employees. The modeler chooses from which direction the greater return for the effort comes. Ph.D. chemists are very good at communicating with other chemists; however, communication with business people is difficult. This poor communication usually results in increasingly better models that become increasingly less accessible to nonscientists.

If I offer people the choice of either golden goose eggs or the goose that lays the golden eggs, most choose the goose. Unfortunately, that is the wrong choice because there is no information on how much the goose costs, whether it's a mean goose, or whether it costs more to take care of the goose than the value of the eggs.

There are many organizations with many managers, particularly on the business side, who don't want models, prototypes, or simulations (the goose)—they want the answers to their questions (the golden egg). The models that have been built in the past are generally not engines of innovation. They have often been treated as a necessary evil and are merely overhead for finding the answers to the organization's questions. Which organizations are just building models that are technically interesting but don't generate answers about the economics of production for an innovative material, and which are investing in models to get answers?

The issue is similar to the goose and the golden eggs because it is usually unclear in an organization what the investment in a model is actually for and whose needs it meets. For example, is the investment in the model to get answers for only the next 6 months, or is the investment to build a more robust model able to meet changing needs and economics?

There is no reconciliation or discussion of that because the answer-driven people don't care about the model; they only care about the answers—they want the golden eggs. The chemists or modelers are only interested in the technical elegance of better models and do not understand the business aspects— they want the goose. There is no honest discussion of the distorted economics because one side of the house thinks they are investing in answers while the other side of the house thinks they are investing in a medium. So are you a golden goose organization or are you an eggs organization?

Then there is the magic mirror. You can stand in front of the magic mirror and ask it to make predictions: What am I going to look like 5 years from now if my lifestyle remains the same? What am I going to look like 5 years from now if I exercise every day and eat right? What am I going to look like if I let myself go? What do I look like in a hat? What do I look like in used clothes? What will I look like if I get liposuction?

The mirror is smart enough to understand the questions. Ordinarily the interfaces of modeling and innovation are not so well defined. How much time would you spend in front of that mirror? What questions would you ask the mirror? What questions would you always ask the mirror? What questions would you never ask the mirror? Which images would you archive and preserve? Which images would you make sure once you saw them you never generated again?

But that is the gutless question set. The gutsy question is as follows: Do you take your significant other in to see your images on the mirror with you? Do you give your significant other the right to ask questions of the mirror about your future appearance, based on suggested changes? Some people would never want their significant other to see the worst-case scenario because it might frighten them away.

Conversely, some people would never want their significant other to see the best-case scenario because that is an unachievable ideal that they do not want to live up to. It requires no huge conceptual leap to realize that this is exactly the situation we are approaching in computational chemistry. Do we want our key suppliers to look in the mirror? Do we want our customers to look? What questions do we want our customers to ask or not ask? What are we prepared to do collaboratively and what not? The technologies for doing all of these things exist. The purpose of the information highway is not just to drive to the solution of a problem, but it is to drive to a location to share with other people, to ship goods to other people. There is a related social context, which leads to the stone soup story.

A vagabond puts a stone into a pot in the middle of a town. When asked what he is making, he replies, “Stone soup.” The townspeople wonder how he could be making stone soup. “It is wonderful soup, but it can use some carrots every now and then,” the vagabond says. So people volunteer to bring the carrots in. The stone soup is a way to get other people to participate and invest in the meal. One of the major problems with modeling infrastructures today is that they do not invite others to participate, and they do not run simulations that are capable of embracing and integrating other people's data. An invitation for them to cook and collaborate with you is needed.

What is so ironic is that collaboration can occur even with bad models. Often times people with very good models have more trouble collaborating than people with lower-quality models because the former group feels that their model is good without additional perspectives or input—it is presented as a fait accompli. The modelers have the answer, and they are trying to persuade you that their answer is right. The people with the mediocre models have a greater incentive to collaborate and pay more attention to a customer, a supplier, or other participant. In any collaborative context, business or academic, a less efficient model often invites greater participation by others than a superior model.

At the MIT Media Lab, the really skilled graduate students and professors stand out because they don't “show and tell.” They “show and ask.” They use a demonstration, model, or prototype to elicit the ideas and insights of others as opposed to convincing others of their perspective.

The people who really manage to build funding, create community, and obtain the interest of venture capitalists all use participatory styles. With this in mind, what are some actions to take that will manage some of the pathological issues associated with models, the things that you could actually use in an organization to leverage the modeling simulation, and the prototyping infrastructures that you have to manage for the innovation process?

First, is the model or simulation designed in a way that it can be used by a key customer or key supplier? Are you designing for accessibility? The use of your model by others is an excellent test of the model's accessibility. When one of your key customers or key suppliers is using your model in a way that you never expected it to be used, your model has succeeded. It is tremendous to learn that your research and development has become market research.

A model can be thought of not just as an interaction engine or an information engine but as a marketplace, where people trade ideas. For example, one group in the organization can only have 30 iterations, another group can have 60 iterations, and they swap and trade between them.

Second, a good model, prototype, or algorithm should attract the attention of other people. The modeler does not need to be charismatic; the work should be charismatic. This is one of the advantages of very good models. The sign of a superstar is not scoring a lot of points but making everyone else on the team play better. In an organization, models and simulations make people think better. They may not even be particularly good models, but they may force a more creative, intriguing, provocative, and useful kind of thinking. Look at how people behave around models, rather than just how models themselves behave.

Finally, the “80/20” rule is absolutely true for models as it is for most things in economics: 20 percent of the model generates 80 percent of the usage. For any model it would be fascinating to know what functionality of the model, what 20 percent, generated 80 percent of the usage. Look at any piece of computer software—only a tiny fraction of the functionality gets used. As you evolve a model, what is the 20/80 ratio? Such auditing is not just targeted experimentation but targeted evolution of how the model is actually used relative to the declared potential of the model.

This kind of introspection will give you tremendous insight into the economics and culture of modeling within an organization, because the real problem is how the models are actually used rather than how they are actually built. There is a utility problem with models more than a design problem. At the core of that is the issue of the economics—distorted economics that lead to distorted behavior. As we become smarter about the internal economics, we will become smarter about the usage of our models and our tools.


Joseph S. Francisco, Purdue University: There is an area of economics that is very exciting called experimental economics. Some of the people here from industry are thinking about taking modeling results and utilizing your suggestion of trades by playing it out within the experimental economics scenario of looking at the outcomes. This could help to guide their investment decisions after seeing the outcome of their initial choice. Is that what you are trying to suggest?

Michael Schrage: That is one of the things. In fact, I am shocked and impressed that you raise that question because there are two lines of thought. There are experimental economics and behavioral economics and finance,2 which have begun to converge. There is work currently being done on the modeling of experimental economics in finance and investment. It would be very interesting to pick those same economics principles and apply them to the planning of a chemical or chemical engineering experiment, treating that as the marketplace. Such research could discover the tradeoff between rational investment decisions and distortions in behavior because of prospect, theory, biases, cognitive biases, and the like. The convergence of experimental economics and behavioral finance, I think, is going to play an enormous role in ultimately shaping how research and development institutions both make and justify their investments.

Kenneth A. Pickar, California Institute of Technology: As a manager, I find what you say to be one of the great attributes of terrific ideas: it is blindingly simple and has the virtue of matching my own prejudices. I will give you two examples. First, the standardization of tools is extremely difficult. This does not apply only to models but also to computer tools for mechanical engineers or a piece of software. The standardization of tools is in everyone's best interest once it is done, but to convince someone that the tools they are using are not as good as the one you want them to use is nearly impossible and requires almost an autocratic approach. Sometimes we will find ways of making it not work.

The second great example is the election in Florida. Here you have intelligent people given the same amount of information and, depending on their predilections, they were convinced that they were being robbed by the other side in a dishonest underhanded way. They were equally passionate about their side of the conflict, and yet in other ways they seemed like reasonable people.

This fits into the context of accepting new information. The idea of making tools very useful, as distinguished from very cool, is important. The part with which I had some trouble is the optimization around bits because my rudimentary knowledge of economics says that you look for scarce resources and try to make them go the farthest. You could iterate a thousand times.

Michael Schrage: No, whenever you have an abundance of a resource, lots of time, lots of people, there is an issue of waste, and when we are dealing in a competitive environment, you need to know when you have hit the point of diminishing returns.

We have got to be careful—you can go past the point of diminishing returns by making sure that every contingency is planned for with Monte Carlo everything. We must be aware that there is such a thing as diminishing returns and that, when a resource is growing, we want to make sure that we don't go down a groove or a rut and that we are at least aware of the tradeoffs.

Kenneth A. Pickar: I will give you a perfect anecdote. Part of my job is to see how engineers are actually using their tools. There was one engineer who was using a particular mechanical engineering tool and had just completed 150 iterations. It turned out that the results had hardly changed between the 20th iteration and the 150th iteration. So I asked him why he had continued to perform iterations. He answered that the time it takes to finish 150 iterations was the amount of time we had allocated in the program to do this.

David E. Nikles, University of Alabama: I have a management problem. I run a team of faculty, which is like herding cats. While I am a grubby experimentalist, we have modelers who talk about the elegant math they will use to do this and that and model real-world phenomena. The models are never finished in time to be useful for the experimental project. I see a fundamental disconnect because I cannot use the model myself for an application. On the other hand, maybe the modelers should be doing some experimenting. How do you manage that? I think the experimentalists have to meet them halfway somehow.

Michael Schrage: Part of the problem goes back to Ph.D. chemists' lack of ability to communicate with people in their subdiscipline. This is problematic, particularly since there is a trend toward funding collaborative and multidisciplinary projects.

If modelers cannot design an interface for the model that other people in the group can use, they fail because they have not produced a model; they have made a black box. Opacity should not be acceptable for any models designed for an interdisciplinary or multidisciplinary setting.

David E. Nikles: I think there are many things we can learn from modeling. However, the modelers always model what I already know and they never tell me something that I didn't know.

Michael Schrage: There are two important characteristics of models: accessibility and generation of novel, sometimes counterintuitive, results.

Mary L. Mandich, Lucent Technologies: I know the employee at Bell Laboratories who wrote Troff, which was an early but very powerful word-processing program. It was so complicated that every secretary who came onboard had to spend weeks in training learning it. There was a computer consultant in every building who you went to when you had trouble. One time I asked this man why they did not take some time to make the program easy to use, and he said that that would have been harder than building the original program. It seems to me that you are telling us not to spend time abandoning the complexity but to spend time on the equally hard stuff.

Michael Schrage: Right. It may be not be appropriate for certain kinds of model builders to focus on user interfaces because that is not how modelers are trained. There are two things going on. There is designing for the better model versus better access. An experimentalist might get useful information from what the modeler would consider to be junk.

There needs to be that kind of negotiation between modeler and experimentalist. Additionally, there are interesting real-world situations that will make the modelers think twice about what the meaning of elegance in the model actually is.

It is necessary to use the model as a medium to manage interaction between the positive and the normative folks. That is one of the underplayed aspects of the problem because the modelers are optimizing it for their community rather than optimizing the models for interaction. You can do this in academe. In business it fails completely. The real question is how do we want this community of collaboration to evolve, and we can adjust our investment in the modeling infrastructures based on that answer.

Richard C. Alkire, University of Illinois at Urbana: Your economic examples here are good ones because they take us out of our box and make us think from another perspective. The impact of all of this information, not only on science and engineering but on our ability to determine the way we will live in the future, is so great that, like economics, it goes beyond economic principles. It gets into issues of ethics.

People understand this in the economics where economic issues become transformed into questions about the capabilities of economic instruments to do good things, especially how they pass from one culture to another and empower people to assemble the resources they value. Architects also think about these things when they create structures that shape the way people live their lives day to day, moment to moment. There are architectural features that have a sense of comfort, whether it is an interesting entrance to a house or a garden bench next to a wall. We all know when we are happy and when we are not happy, not because of economic decisions or architectural designs but because we feel comfortable. Turning now to the Information Age, could you comment on the process by which we might learn to design our living space so that we can feel comfortable in the presence of so much information?

Michael Schrage: That is an extraordinarily difficult question to answer; therefore, I am going to oversimplify it to a level where I am comfortable answering it. There is a major ideological battle going on between the normativists and the positivists. That battle is fundamentally based on the core of economics—how people should behave—which exemplifies rational choice versus actual behavior. Theories like experimental economics and behavioral finance represent some effort to arbitrate normative expectations with positive observations.

I have made a slow and eventually accelerating migration to the positive side because people don't behave rationally, how they are expected to. There are two superb books addressing this. Image of the City by Kevin Lynch discusses the mental maps that people have. In this book, cognitive maps and spatial representations are tied into the architecture of city planning, which is directly relevant to many issues in design. The other book is How Buildings Learn: What Happens After They're Built by Stewart Brand. Serious Play: How the World's Best Companies Simulate to Innovate, authored by me and Tom Peters, looks at the issues associated with models, prototypes, and simulations and the culture of models, prototypes, and simulations in organizations.

Richard C. Alkire: I would like to mention that my own remarks are based on a book called A Pattern Language: Towns, Buildings, Construction by Christopher Alexander.

Michael Schrage: His work is superb. Incidentally, the ideas in A Pattern Language have been reappropriated by the object-oriented software designers (architectural design rules, laws of form, and the like)—truly technology transfer.

David J. Soderberg, BP Chemicals: I believe that the psychology of model application, technology management, and the business interface are more important than economics. One of the challenges technology managers have is interfacing with businesses. For example, in process modeling a group of modelers will talk to a group of process engineers about a model, but because of their common backgrounds, they drive each other for a better model, rather than one that is meaningful to the business.

Michael Schrage: The model is often designed to coordinate how the company behaves as opposed to how it interacts with clients or customers. Your example reinforces the fact that there are different design emphases.

Unless they are designing the model as a medium for communication, in addition to a medium to improve their problem-solving capabilities, they have failed in their professional responsibilities.

Walter G. Copan, Lubrizol Corporation: I wanted to thank you for your insight on the subject. Your comments on these issues have resonated with many of us. We have certainly begun to see the power of the potential represented by the interactions between human beings and models. This impact can be seen in new kinds of customer-supplier relationships as well as partner interactions. Because shared models create a deep, rich dialogue that provides profound insights, a much higher level of understanding is possible.

I would like to ask you a couple of questions about the power that models have to increase interaction and understanding. First, what are the elements of success you believe are critical to achieving the ultimate benefits of interplay between organizations and the full use of the power of their models? Also, what do you believe will happen to the future of customer-supplier or partner interactions as a result of having models available that are commonly built?

Michael Schrage: Well, those are both great questions. For models to make the maximum impact possible, we need the different incentive structures to encourage model usage. There is a very simple rule: using a model should be easier than not using a model. The real question, then, is if I learn to use this model, how long will it take me to feel like I am getting a real benefit from it? That deals directly with user interface.

Employees need an incentive, a reward for using the model. I believe that most large organizations should set up two kinds of prizes, one for the group or team whose model gets used the most by other people in the organization, the other for the person or group that does the best job of stealing somebody else's model.

Walter G. Copan: The second question is “Where could modeling lead us ultimately in terms of customer-supplier and -partner interactions?”

Michael Schrage: To answer the second question, I believe that business intercourse and design intercourse will increasingly be mediated by models. A wealth of doctoral theses will be done along the dimension of in which industries do the vendors use the customers' models, and we are seeing the supply chain management, and in what industries do the customers use the vendor models, which we oftentimes see in aerospace or high-tech industries, where the suppliers' competitive advantages are disproportionately more sophisticated than their customers'.

There are information asymmetries in certain industries regarding whether the supplier or the customer has a competitive advantage by using models. These information asymmetries will be reflected in the shared spaces, and models will become the media for collaboration. They will become the bridges and the glue between disparate organizations.



Michael Schrage is co-director of the MIT Media Lab's eMarkets Initiative and a senior advisor to the MIT Security Studies Program. His research focuses on the role of models, prototypes, and simulations as essential media for managing innovation and risk. His book, Serious Play (Harvard Business School Press, 2000), explores the economics and ethology of modeling within organizations.


Shortly after this workshop, the 2002 Nobel Prize in economics was awarded to Daniel Kahneman and Vernon L. Smith for research in these two areas.

Copyright © 2003, National Academy of Sciences.
Bookshelf ID: NBK36340


  • PubReader
  • Print View
  • Cite this Page
  • PDF version of this title (5.5M)

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...