Web hosting Custom Email SiteBuilder

Journal of Bionomics
Edited by Steve Waite

• Bernardo Huberman's Biography

JOB: Your latest paper, "The Dynamics of Organizational Learning", presents a dynamic theory of organizational learning, which refers to the increases in productivity observed as firms gain experience in production. We suspect many CEO’s are not well versed in the literature on organizational learning, though they should be. What attracted you to this issue?

BAH: That is an interesting question. I am attracted to phenomena that display strong regularities. That is, observations that hold independently of the nature of, say, the business, the organization, and so on. As you know, the social sciences, including economics, sometimes suffer from the fact that although there are lots of theories and lots of opinions, they are not necessarily corroborated by empirical facts, across a wide spectrum of cases.

JOB: Uh huh.

BAH: Now, the phenomenon of organizational learning and the aspect that I studied -- namely, that the cost in constant dollars of producing a particular good or service as a function of the number of units you have already produced tends to decrease in a very regular way is a remarkably solid regularity. If you read Michael Rothschild’s book, Bionomics, there is a whole chapter on the universal curve. He makes a point, and I think he is correct, that this is an extremely important phenomenon that affects the way we think about growth and economics. From my point of view, the fact that there are no real theories on this phenomenon made me very interested.

JOB: So you are interested in things that have a regularity, a pattern to them?

BAH: Yes, I like patterns. I like things that are actually observed. The universal curve in organizational learning has been documented since the turn of the century and it invites all sorts of reflections and questions about its nature.

JOB: Let’s talk about the power law of organizational learning, which is discussed at length in your paper.

BAH: OK. If you were to look at the cost of making something, like tanks, airplanes, umbrellas, paper or even financial services, you will notice that the cost of production as a function of the number of units of the good or service produced already tends to go down in constant dollars. And the decrease in price is not just arbitrary. It follows a pattern, a so-called power law, because it is the inverse of the number of units produced raised to some power. The power, or exponent, tends to vary across industries but it is typically observed to have a value near 1/3. It is this value that underlies the common observation that each doubling of cumulative output leads to a roughly 20% reduction in unit cost.

JOB: That is indeed powerful.

BAH: Yes, very. There has been a lot of empirical work done on this subject over the years. The interesting thing is that the theories on this issue are largely unsatisfactory. It is an interesting problem, so I thought I would try to explain it.

JOB: If you are running a business, how would you measure organizational learning?

BAH: You would plot how much it would cost to produce a unit of the widget you are producing as a function of the number of units produced. If you were to do this, you would notice that adjusting for inflation, the price of the good you produce goes down as you manufacture more and more widgets. You could then project, based on the power law, how much it will cost you to produce the thing in the future, if the constant dollar value of learning continues. Of course, there occur at times technological discontinuities -- when new technologies are introduced. As has been pointed out many times, the existence of the learning curve allows a business to set up its pricing. This tool has been used in other instances as well, for example, in international trade with respect to the issue of dumping.

JOB: If you look at the history of business through the lens of organizational learning, does it necessarily follow that companies get smarter over time.

BAH: Absolutely. That is the whole idea. Businesses find new ways of saving money and time -- usually time in the production process. That generates efficiency gains. I don’t really know how to define in general the notion of being smart. But if you measure the smartness of a company by how cheaply it can produce something, presumably they are getting smarter because they are learning how to make the same thing in a shorter time and at lower cost.

JOB: The government statisticians tell us that growth in output per hour in the service sector hasn’t really changed in the past two decades despite the tremendous advances in communications and information technology. I see powerful applications of your theory in the service sector. The problem is you are not producing widgets, you are producing things government statisticians don’t measure properly. I suspect that the power law applies to the service sector as well as the manufacturing sector. Is that correct?

BAH: I am not an expert in this area, but I feel that they are measuring the wrong thing. Take the extreme limit where you look at a financial services company such as an insurance firm or a commercial bank. If you unplugged all of the computers, the productivity of these firms would suffer, don't you think?

JOB: Yeah, and the government statisticians would never know! Robert Solow once quipped that we see computers everywhere but in the productivity statistics.

BAH: I don’t really get it.

JOB: I don’t either. Organizational learning is a function of technology, right?

BAH: Yes, of course. It is a function of technology, but there is also a measurement issue. My sense is that when people like Solow talk about these issues they are using a productivity measure that does not take into account the vast amount of information that people have on their desks. From a manufacturing point of view or even when you look at the number of mistakes made at a nuclear plant, you see a decrease in the number of mistakes, which essentially amounts to learning. Everything in the manufacturing sector that has a complicated set of procedures eventually exhibits the behavior captured by the power law. I haven’t extensively studied learning curves for the service sector, so I cannot really discuss this. But I agree with you that the government is measuring the wrong thing.

JOB: Let’s discuss for a moment a concept in your paper called "organizational forgetting."

BAH: Sure. Let me first tell you about the data. It is a fact that usually after periods when a plant has gone idle, whether because of a strike or retooling, the cost of producing widgets after the shutdown tends to be higher than it was before the shutdown. If you were to look at the cost curve, you would see that superimposed on the ever decreasing costs of manufacturing there are bumps corresponding to rising costs which eventually decay back into the learning curve. The bumps coincide with the period following the shutdown. Many people call this organizational forgetting.

JOB: Interesting.

BAH: There is another phenomenon associated with organizational forgetting. This has to do with the fact that many times, when the procedures are not well documented and people leave an organization, you tend to see that learning actually slows down. This is because it takes time for the new people who join the firm to learn the procedures that were done before but for which there is no documentation of how to perform the job. Organizational forgetting has to do with the fact that there is knowledge embedded in the organization that sometimes is not captured correctly. As a result, as old employees leave and are replaced with new employees, knowledge in the firm evaporates. This typically results in things getting done less efficiently than before.

JOB: Makes a lot of sense.

BAH: If you look at the fast food industry, for example, there is a good rate of learning. And yet, as you know, the turnover in that industry is incredibly high. The reason new employees do so well is because these procedures are well documented.

JOB: They are standardized.

BAH: Exactly. Whereas in many other situations this is not the case. There is a famous example that was published in Science magazine several years ago. The article stated that the design of nuclear weapons at Los Alamos National Laboratories was not well documented, in the sense that a lot of the underlying assumptions behind computer codes were in the heads of the scientists and the engineers involved. This turned out to be a concern because many of the people involved in nuclear weapons design are retiring. There are a lot of details that go into the design of nuclear weapons, as you can imagine. So now the folks at Los Alamos are interviewing people to try to document how they did what they did.

JOB: You mentioned in your paper that as new routines and shortcuts are introduced into an organization, care must be exerted so that the associated learning of new procedures compensates for the increase in novel ways of working.

BAH: One way of thinking about all this is to imagine wandering down a maze. The exit of the maze is essentially the product you are producing. When you walk through the maze, every once in a while you find shortcuts. It is also true that when you walk through the maze, there can be many more shortcuts but then there are many more ways of getting lost. It is like adding more streets to a city, thus increasing the possibility of getting lost. When this happens in the context of organizational learning you see a novel phenomenon which I discuss in my paper, namely production costs increase rather than decrease with the number of units produced. This was observed in the production of the Lockheed Tri-Star L-1011. This is one case where unlearning was so rapid that it almost broke Lockheed Corporation.

JOB: Do you call that the power law of unlearning?

BAH: You could. It is not very often when you see this, but it does happen. My theory on organizational learning gives you both cases. I believe that learning in a organizational setting has two components: 1) discovering new shortcuts, routines or procedures, and 2) selecting, out of all these procedures that you are presented with, which ones to really follow. Every time you look at a production system, you have to decide which way to go. Most of the time organizations do not spend much time thinking what should be the appropriate next step. Most people find shortcuts that get implemented and that's it. Most of the time you just see the power law of learning associated with the introduction of shortcuts. Rarely do you see an organization shutting down to reflect on what it is really doing and how it could be doing it better.

JOB: I am fascinated by the L-1011 story. Are there other examples?

BAH: There are other examples in a different situation, which I will call the limit of production by a single individual. The power law of learning is also seen in individual cases, and as a matter of fact, it was first observed in that context. If you look at how people roll cigars in a factory, or learn how to play with Rubik's cube, for example, they also exhibit a speed up as they practice a task. Now it turns out that these anomalies that are seen in the case of the L-1011 have a funny counterpart in individual learning. Take the example of a person learning to repair computers. At the beginning, sometimes you see that the person gets more and more confused the more details they learn about what they are supposed to repair. That is because there are so many possibilities that it is easy to get lost among them, and it is only with practice that they discover the shortcuts that lead to fast repairs.

JOB: Interesting.

BAH: Since firms do not like to publish their learning curves, as they tend to be secretive about the data, and second, failures don't get documented that well, this anomaly is not the kind of thing you see in literature.

JOB: Wouldn’t it be fascinating to go into Microsoft or Intel and study the learning that goes on at these organizations?

BAH: Most likely, they would never let an outsider do it, but that is an interesting issue. Microsoft is a huge success story, and I would be fascinated to see what the cost of software production is and how it declines with experience. There have been a few studies on this, but usually they are done in academic settings and not in business settings, so they have less relevance.

JOB: I am wondering whether as the information age progresses, we will uncover new laws of organizational learning?

BAH: There may be a new law. Most likely we will keep confirming the existince of the learning curve and establish how the exponent will vary across industries. You have to remember that the literature on learning curves has been evolving since the 1920s.

JOB: Let’s talk about the connection between individual and organizational learning, which you said in your paper is an issue that remains largely unresolved.

BAH: That is a tough one, and for many reasons, but not necessarily insoluble. I have grappled with this one for quite a while but without success. We have done some research and we have even done some computer simulations. As I said earlier, it is a fact that individual learning also follows this power law. A very nice experiment was performed at PARC by P.E. Agre and J. Shrager where they videotaped a subject photocopying a two-sided document. They noted that there was a power law of learning. The most interesting thing was that they were able to observe how the subject was finding lots of new shortcuts, so I believe that the basic mechanism of my theory is correct. But in connecting the individual and organizational level of learning is another story, for individual performance is not the whole story. This is because an individual can learn but if that knowledge doesn’t get propagated through the organization it doesn’t go anywhere.

JOB: Right.

BAH: So there are two levels at this point, and it is difficult to tell what connects them. Obviously, if everybody were to leave an organization the organization would learn nothing. So we know that people make a difference. How it is that they make a difference is a complicated problem because a lot of the knowledge is already outside people’s heads -- it is in books, procedures, etc. So you bring in a new person to a company, and that person doesn’t know much about the job, but they can perform the job very well. The most famous example, by the way, was documented during World War II and involved the production of Liberty ships. They were making one a week I believe. The turnover was incredible, and yet the learning was phenomenally straight. People came on board and there were such well documented procedures you could teach someone right away what to do. Over time, people found new shortcuts which were documented and production costs continued to fall.

JOB: Sounds like organizational communication is important to the process.

BAH: Yes, you are right. But it is more than that. It is also how well other people generalize a discovery into something else. If you are working in manufacturing and somebody discovers something in accounting, why should you bother learning it? These things are more of an art than a science at this point.

JOB: Perhaps an Intranet would help solve this problem. For example, you could have a site that tells employees the most efficient way to copy two-sided documents.

BAH: Perhaps. That is actually happening in some cases, but it also leads to lots of noise in the system, as most of the information you receive on a daily basis is irrelevant to you. But we still don’t have an explanation of how it is that discoveries by individuals in organizations lead to organizational learning in a systematic way.

JOB: But it seems to me there is something about the concept of networking computers that enhances organizational learning.

BAH: I think the idea of networking really makes a big difference. It allows for the propagation of new discoveries and new routines throughout the company.

JOB: I read recently that some companies are employing anthropologists.

BAH: Yes. We do at PARC. What they do is trying to understand at a very, very local level, what is it that really gets involved when people say they perform work. And with the new technologies, how is it that these new technologies truly affect the way people do work. I think there are some very useful insights that come out of this.

JOB: Should these anthropologists observe and record how organizations learn?

BAH: They do. The problem with the methodology they use is that it tends not to be very generalizable and it is not very quantitative. It is, nonetheless, very important.

JOB: Do you think more companies will be hiring more anthropologists to observe organizational learning?

BAH: I think companies should be hiring people, anthropologists or whatever, who understand all of these new information technologies to find out how they can leverage them inside the corporation.

JOB: Learning is at the very heart of Bionomics and is a critical determinant of corporate profitability. The most profitable companies are typically those highest on the learning curve.

BAH: That is correct. Now the question is, how do you translate the findings of my theory into practice, so that it can be used by organizations to increase the rate of learning? I agree with you that the notion of learning and the notion of discovering new ways of producing things is important.

JOB: You have put together a dynamic theory of organizational learning. Is there a static theory of organizational learning?

BAH: I think that static theory of learning is a contradiction of terms. Equally bad, many people use the word dynamic, but when you look at what they write, there is no variable called time. Time is absent from the analysis.

JOB: It seems to me that all this information technology and computational power we have today allows us to study dynamic systems in a way that we could never study before. My economics courses in graduate school focused mainly on comparative statics. But it seems to me that if you are an economist or money manager on Wall Street, or a business strategist working with Andy Grove at Intel or Bill Gates at Microsoft, you are mainly interested in how to navigate through disequilibriums. I believe Andy Grove calls these Strategic Inflection Points.

BAH: Yes, and although I don't know about these Inflection Points there are important advances in finance are associated with dynamics. Think about the pricing of options.

JOB: The Black/Scholes options pricing model is based on a model of heat dissipation.

BAH: Correct.

JOB: Speaking of dynamic versus static systems, when you think of governments, and you think of organization learning, what comes to mind? It seems to me that there is a whole bunch of organizational forgetting that goes on in government. Administrations come and go. Some political parties remain in power for decades, and then get replaced. I think of the Labour Party in Great Britain that has been out of power for almost twenty years. I wonder what their learning curve looks like, assuming they win the election next spring.

BAH: You have to remember that in Europe the bureaucracies stay in place for years, regardless of the party in power. Only the top layers of government changes. In many of these countries, the procedures don’t change, right?

JOB: Apparently so.

BAH: My friends tell me that the procedure to get a telephone in Italy today is as long and contrived as it was many years ago. The civil servants in these countries are very permanent and they are tremendous agents of resistance to change. I think in the United States it is much more complex, but overall I think the infrastructure of government doesn’t change that much. Look at the State Department or the Department of Defense. The people that actually do something are there all of the time. It is true that you get a new Secretary of State or Assistant Secretary of State, but the core remains fairly stable.

JOB: Let’s go back to Italy for a moment. Assume that the bureaucracy in Italy stays in place, and that they continue to do things the way they did twenty years ago. Let’s now introduce new information technology. The bureaucracy resists, but if the new information technology revolutionizes the way things get done in the world, something radical will have to happen to shake up the status quo.

BAH: That is absolutely correct. There will be a discontinuity in the system.

JOB: If you look at competing systems, say a public and private postal system. One is run by the federal government and the other by Federal Express. Give them the same capital base and then allow them to do business. It seems to me that it is a no-brainer who will become the most efficient provider of that service.

BAH: Of course, the one that has to live off of its profits, not the one that is subsidized.

JOB: I believe we can safely predict this outcome because of what we know about organizational learning in the private sector versus the public sector. This is very important.

BAH: Or perhaps what we know about incentives. This is one of the reasons why federal governments everywhere around the world are divesting themselves of manufacturing processes and giving them to the private sectors.

JOB: It all boils down to organizational learning, right?

BAH: That is correct. Learning means that the cost goes down. You have to be interested in having costs decline. If you are a government employee and you don’t care about costs, why should you bring costs down?

JOB: I am glad we got all of that cleared up. What’s next on your research agenda?

BAH: There are several issues I am working on. At the most general level, I'm trying to understand the relationship between local procedures and the global dynamics of distributed systems. This also involves what in computer science we denote as the dynamics of multi-agent systems, which means understanding the behavior of intentional agents and their aggregates. The problem is always one of learning at which level you can make progress. You know, there is a good thing and a bad thing about my paper on organizational learning. The good thing was that I realized at what level I should study, and that is the level of routines and procedures. That was my insight. The bad thing is that the theory doesn’t connect to the underlying people who are doing the actual work in an organization. You gave a very good example of this in the theory of heat diffusion. People talked about heat diffusion in the past century without discussing atoms and molecules, but they got the basic phenomenon and wrote the correct equation that describes it. Eventually, though, we learned about atoms and molecules, and had to explain heat diffusion in terms of the motion of those atoms and molecules. It wasn't easy but it was done. Here we have a situation where there is a possible explanation of organizational learning in terms of procedures and routines at the level of the organization, but I still don’t know how to connect this back to the individual level. This is a problem I would like to solve.

JOB: I think it would be great if you would look at the service sector productivity issue. This is a huge issue for economic policy makers and I think your insights on the topic are right on.

BAH: Well, I would certainly like to explore the issue. As the service sector continues to grow, we have to know how to measure efficiency curves.

JOB: Agreed. Thank you very much, Bernardo.

Related Links:

• Xerox Parc
• Bernardo's Web Page

[ Home