Alph Bingham
Vice President
e.Lilly, a division of Eli Lilly and Company

"Extracting Information from Networks"

It is a privilege to be here. I can tell from the list of speakers that this is going to be an outstanding conference. I feel a little bit like a grad student coming back to give a seminar to his university professors. It is a little intimidating to present before this group of speakers. I needed to learn a lot from this group of people because it was necessary for our business to rethink its fundamentals.

In 1998, I heard Brian Arthur give the Ulam Lectures in Santa Fe. This was one of my first experiences with this area of thought. Brian distinguished between "cyberspace" and "real space" (or as a friend of mine calls it, "meat-space"). Brian drew a diagram of what goes on in cyberspace, what goes on in meat space, and the conductors between them. We thought about this model, and asked ourselves how we could keep more of our business model in cyberspace. When I tried to explain this concept to colleagues, I used the example of Amazon.com. Much of their business remains in cyberspace through the ordering process, but at some point a physical book has to be put in a real box and shipped to your house. What if they could keep it in cyberspace longer? What if you could buy not just a color inkjet printer, but also a little book binder as well? You could buy a book online, print it and bind it in your own home. That would enable Amazon.com to extend the amount of their business that is in cyberspace, which would strengthen their returns and efficiency. Amazon.com does not yet offer a $300 book binder, but you can order some books electronically, which gives you access to the text.

I have titled my remarks tonight "The Net @ Work". I will talk a little bit about network structures. Part of the inspiration for us as we rethink our business model is that our business needed to evolve and change. We have worked with Peter Drucker to explore what a post-corporate structure might look like. We have begun some experiments in which networks can replace traditional organizational hierarchies as the work structure of the future.

I would like to provide a very brief background on networks. There are three types of networks. In "Sarnoff Networks" or broadcast networks, the value of the network grows as a power of the number of agents in the network. The power of the network is simply a power of the number of viewers or listeners you have in your broadcast system.

Transaction networks are governed by Metcalfe's Law. The value of these networks grows as N-squared (N being the number of agents in the network). Most transactions on a telephone network, for instance, involve pairs of agents. The power grows as the square because of the different permutations.

Affiliation networks represent all of the different ways in which you can assemble a group of agents. This is a combinatorial problem, so the value of this network goes up at a rate of 2 to the nth.

The value of the affiliation network grows much more rapidly than the other types of networks. How do you create value in each of these network structures? How do you "win" in each of these network structures? In a broadcast network, the winner is determined by who has the best content. The best content will translate into the number of viewers, which will translate directly into the number of advertising dollars that the network can generate.

In a transactional network, membership itself will determine which network is the winner. You want to join the network (like eBay) in which you can have the greatest number of interactions with other agents.

In collaborative networks, we hope to restructure our resources and network in accordance with the demands of the moment. The network will be characterized by all of the different ways in which the network can be configured. The winning network in this environment will be the best-facilitated network - which network will offer the greatest ease of reconfiguration? How can we keep the transaction costs of reconfiguration near zero?

The fundamental question is how we can exploit networks? There are three basic levels at which networks can be exploited. The first is the simple act of mapping. I will present you with examples from outside our organization, and then I will give you examples of how we have used that type of learning to solve real-world problems. The second level of exploitation is the application of a simple rule to the network to extract value from the network. The third layer I have called "Workflow" - how can networks replace some of the traditional hierarchical organizational structures in business? How can we perform the very essence of the business itself using an alternative organizational model?

Let's talk first about mapping. The first way to extract value from a network is simply to understand what kind of network you are dealing with. Many times these networks are implicit and poorly understood. Several network maps have recently been published of protein-protein interactions and traffic on the Internet. They also mapped the relationships between the various terrorists involved in the 9/11 events. Each of these networks represents scale-free behaviors.

Now, I would like to talk about an application that we are using. This is an application that some of you may be familiar with called "gene chips". Genes are placed within a small micro-cell, and that array is visualized under various conditions. I might establish a baseline array for an individual, and I will see the behavior of each gene that is mapped out on this chip. Is that gene being expressed (or turned on) or suppressed (not turned on and therefore not manufacturing the protein that it encodes for)? I might then take an image of that individual after they have acquired a disease. I will look for changes in gene expression, and I might subject that individual to treatment and then look again for changes in gene expression. For the most part, this is done as a complex N by M array, and by looking at these images, I am trying to decode what those changes are. I can tell when a gene turns off or on. It is more difficult to determine what the relationships are between the different genes.

Eli Lilly has been building an interactive network map showing how all of these genes relate to other genes. Using this new map, we can see the relationships between all of the genes. This map is based on an established hierarchy developed by gene oncology. We can now read these maps to determine which genes show dramatic changes during therapies. This can be used to show not only the efficacy of a drug, but also potential side effects. It is rare for a single gene to be affected by a treatment - it is often a collection of genes that responds to a therapy. It helps us zero in on which are the key metabolic processes. This is the kind of scientific application one would expect from a pharmaceutical company.

Let's move on to a sales application of networks. We know that the practice behavior of physicians is a complex interaction between the knowledge they possess, the regimens of the institutions in which they practice, and the interactions they have with other physicians both professionally and socially. Long ago we hypothesized ways in which we can rationally deploy a sales force to maximize the distribution of new information about new treatment regimens. We decided to test whether those strategies were optimal in the Boston market. We tested how physicians were influenced in the ways they practiced medicine.

The reigning hypothesis is that there are global thought leaders - individuals who publish and make declarations in meeting - whom most other physicians follow. The second rationale for deploying a sales force is that anyone who has prescribed a lot of medicine in an area will continue to prescribe a lot of medicine in this arena. This is the kind of physicians to whom you would provide details about medical breakthroughs in this arena.

As it turns out, that's not quite the way it works. We constructed the influence diagrams to discover that there was far more local effect than we had believed. We also found that the act of prescribing a medication is not necessarily an indicator of influence - it is an indicator of practice. Perhaps counterintuitively, those who were less active in seeing patients and prescribing tended to be more active in research and were therefore more influential than we had previously believed. In fact when we built the map, we discovered that there were relatively few hubs of influence and at least one of those hubs was a doctor to whom we had never talked before because they didn't meet our prescribing criteria. Having built the network map, we discovered that there was a far more efficient way to structure our sales force to disseminate information.

Let's move to the second level of extracting value from a network - applying a simple rule. The simple rule that you are all familiar with is "buy low, sell high." This is a reasonably easy rule to remember. From this simple rule, however, has evolved a very complex set of interrelated behaviors on the part of investors. We have tried to apply this rule to our own business as well. We began to experiment with knowledge markets. We created a trading floor, not for the purpose of finding the "right price" for a good, commodity or security, but instead to achieve information efficiency. We looked at www.hsx.com, which trades shares on Hollywood movies and actors in order to predict how successful a movie might be. We looked at the Iowa Electronic Markets (www.biz.uiowa.edu/iem), and I will discuss those in more detail later. We had several important partners in our early experiments.

When we talk about markets, we are looking for information efficiency and allocation efficiency. In order to create a knowledge market, we create a security around an event that we would like to understand better. Let's take a hypothetical example. I will pay $100 if there is a military coup in Argentina before January 1, 2008. Most people will say that they do not know much about Argentina's political situation, but there are some things that we know. None of you would pay $99.99 for this security. There is too little upside potential and too much downside risk in that proposition. You claim to know nothing about Argentina, but you know enough to create some kind of bracket around a "reasonable" price. If we started trading these shares in this room, the price would hit stability surprisingly fast. With other groups, this price has settled quickly to $6 or $13. This reflects not only the probability of the event, but also some measure of the risk tolerance of the population. We can still extract a lot of information from this kind of trading.

We can look back on knowledge markets like the one for the Bush-Dukakis election in 1988. The Iowa Market was created in which a share was established for Dukakis and a share was established for Bush. The price of the share would be equal to the percentage of the popular vote that that candidate would receive on Election Day. Taking the difference between these two share prices, we can find the margin by which victory would be achieved.

This market is going on for the 2004 Presidential Election as well. Over the past several months, there has generally been some preference for Bush, but it has not been exclusively so. If you ask the question about the distribution of votes, you will get one kind of response. If you ask how confident we are that one person will win, then you have created a "winner take all market" in which the winning share will pay out, say, $100 and the losing share will pay $0. This question creates different dynamics and provides you with different kinds of information. The confidence that Bush would win began to soar after the Republican Convention. This trend peaked at the end of the first debate, and once Kerry seemed to demonstrate some capability as Commander-in-Chief, the certainty of a Bush win began to drop. It continued to drop throughout the debate period. This market therefore highlights very different information. Incidentally, this market is focused on fraction of the popular vote. It does not try to predict the outcome of the Electoral College.

How would we apply this to our own business? We created six hypothetical drug stocks and a winner-take-all market. Because these were hypothetical drugs, we knew in advance which of them would be successful through Phase II clinical trials. This is a very critical point in the decision-making process for pharmaceutical companies. This is the point at which the investment required jumps up to the tens or hundreds of millions of dollars. We randomized the information about the drugs. We created mock drug substances that were actually assemblages of real case studies. We randomized the way in which the information about the drugs would be released. We asked the traders to buy stock in the drugs based on the simple rule of buy low and sell high. We would release tidbits of information about each of the drugs. This would let the traders respond to the new information and it would be aggregated across a large number of scientists in this marketplace. Normally we would assemble a limited number of experts (about twenty) who would receive all of the information filtered through others and have to make a live-or-die decision on each of the drugs.

If you look at the end of the trading, there are clearly three strong stocks and three weak stocks. Naturally, once the answer was known, then no further trading was allowed. The three strong stocks turned out to be the three successful drugs, and the weaker stocks were indeed duds. The trading mechanism knew the answers to these questions for the most part about halfway through the exercise. The information was pulled together so efficiently by this mechanism that the answers were clear well before the end of the activity. If this were translated into practice, it would represent literally hundreds of millions of dollars in cost savings for our business.

Now, today the confidence in this kind of trading mechanism does not surpass the confidence that we have in the traditional decision-making process. This is a dynamic that is still being worked out in our company. This is very much like the resistance of Hollywood executives to look at the Hollywood Stock Exchange to make opening day predictions.

Once we had run such an experiment, we were able to pick it apart to seek out new information. We were able to compare the "consensus opinion" with the trading price of each drug. Even as individual traders were refusing to say that there had been any consensus reached, the trading prices were converging pretty convincingly.

We can compare this to a restaurant. If you go to a restaurant and have a good meal, what is the likelihood that you will have another good meal the next time you go there? Based on a sample of one, you might think that the odds are still 50-50 or maybe 60-40 that you'll have another good meal. The "odds" won't be that strong. On the other hand, if you started trading a stock, then if a lot of people had had one good meal there, then the trading price of that stock would be much higher much faster.

There is still some reluctance to publicly disclose all of the places in which we are working on this. We continue to run these kinds of markets. We are looking at valuations of our portfolio. We are using markets to predict sales forecasts. We try to understand potential legislative outcomes. We are generally not trying to identify ourselves in these markets, in part to avoid being tarred with the same brush that Poindexter and company got tarred with a few months ago, but also not to distort the markets themselves. We want the markets to operate independently from our sponsorship. The early returns on these experiments are encouraging.

Let's talk next about Workflow. We are first and foremost in the innovation business. We do innovation the way innovation should be done. David Runyan once said, "The race does not always go to the swift, or the victory to the strong, but that is how you bet." We are good betters. We recruit the swift and the strong. We pick the best talent out of the universities. We bring them into an environment where they have resources and they can interact with other bright people. We ask them to produce a steady stream of innovation that can keep our earnings and stock price growing. In the area of innovation and creativity and invention, this is a pretty tall order. It is not exactly clear why that would work. If you look at history, you see that the race does not always go to the strong and the swift. If you were to put money on who would discover the structure of DNA, you would support Linus Pauling. You would not put your money on a couple of upstarts named Watson and Crick. So it goes with innovation.

If you're not going to use this betting approach, what other approaches are available? One of the non-traditional ones that we explored is talk radio. If you think about what scientific research is, there are some parallels to talk radio, believe it or not. Talk radio is an environment in which a query can be formulated and a response generated. This is fundamentally what scientific research is. Scientists develop their query as a hypothesis, and they perform experiments as a way to craft a response.

I live in Indianapolis. When May rolls around, you can guarantee that radio talk shows are talking about the Indianapolis 500. They will analyze tires, rules, drivers and everything else. As with any talk radio program, they are bound to go off on a tangent. Let's imagine that their tangent is about who came in third at the 1938 running of the Indianapolis 500. If that were a scientific query, we would know what to do - we would go to the literature and look through books until we found the answer. This is not what happens on talk radio. The answer comes back in about thirty seconds. The answer comes from a listener who knows who came in third in the 1938 race. In fact this guy worships that driver as a hero. That is a bizarre phenomenon. This gets practiced in a lot of different ways in chat rooms and on help lines.

We asked whether there was a way to build a business around this phenomenon. Can we actually execute the innovation agenda of the company using this mechanism as opposed to the "smart way to bet" approach? To create our business model, we coupled the talk radio model with the old bounty hunter model. One of the dilemmas we face in the pharmaceutical industry is the amount of risk that we assume - that group of twenty really smart people makes a lot of mistakes. They make a lot of false positive errors and false negative errors. All of this error making incurs costs for the company, which translates to higher risks and higher margins required for survival. We need to find a way to strip out the costs from the system.

By combining the talk radio and bounty hunting models, we shift some of our risk out into the marketplace. At first blush, it strikes people as arrogant and audacious for a wealthy industry to shift their risk onto these independent researchers. Risk is not symmetrical. The risk that we move out of our business is not necessarily equal to the risk that the researcher takes on. We are no longer assigning challenges to whatever researchers we have in our labs. These researchers can self-select the challenges that they take on based on their own knowledge of their capability to come up with a solution. In this manner, some of the risk is already trimmed out of the system.

Risk is also trimmed out in a second manner - these researchers have multiple utilities for even working on the problem in the first place. Tom Wandless is a professor of chemistry at Stanford. He told me that he didn't know how much a business model like Innocentive would influence him. He had dedicated his life to finding new ways to synthesize hydro-amino acids, but he would be happy to include our compounds in his table. He was already doing this kind of work. He looked at any award as frosting on the cake. This represents a completely different risk than I would hold inside a pharmaceutical company. The risk is not symmetrical.

We posted some of our problems on our website. Scientists can look at a picture of a chemical compound and instantly know whether or not they are a player. They are required to navigate through several layers of the problem, and we are required to protect intellectual property in both directions. Furthermore, we need to also provide protections for the internal IP of the seeker companies. There were a number of IP issues that needed to be worked out in order for us to combine the talk radio model with the bounty-hunting model.

Facilitation is one of the ways in which some of these networks can be competitive. We have established a workroom where people can execute work. Watson and Crick were scientists working near the bottom of the "iceberg" - there were other scientists who were much more visible and above the water line. Nevertheless, we have been able to tap into the tremendous potential of the network beneath the surface. We have been able to solve some very difficult problems using this mechanism.

Our network of solvers continues to grow, and has reached almost every country. Solvers are required to provide some personal information when they sign up, so we can watch for trends. We now have two registrations in Mongolia for the first time. I would love to swap t-shirts with them - "Mongolian Department of Chemistry"!

If I wanted to broadly expose a problem in chemistry or biology within our organization, I could get at most 700 minds looking at the problem. What are the odds that the solution will lie among those 700 minds? Using this new mechanism, I can expose the problem to one hundred times that number of minds. I can increase the likelihood of finding a solution by 100-fold. Over time, the demographic composition of the solver community has shifted. We started with a large North American solver community since word spreads quickly here. Over time, word has spread to more out-of-the-way places, and we have seen the community grow in other areas. The community now roughly correlates with the worldwide availability of scientific minds. China is our largest community. The US is second largest. England and Russia are third and fourth, respectively. Our community continues to grow strongly, but in our fourth year we seem to be stabilizing our rate of growth.

This is not a Lilly phenomenon. We realized that this was not a business we wanted to be in. Instead, we wanted to be a customer. Other companies have joined as customers as well. The initial reaction from most companies is "How can I be sure I'll get my problem solved?" Some companies feel that we are fundamentally threatening their identity. These people have PhD's because they are trained problem-solvers, not problem-askers. It turns out that even if you do have a PhD, you may not be the best problem-solver out there. When you engage a large number of minds, a lot of different things can happen. When we had 700 solvers, unusual things did not happen very often. Even at 5000, we got good scientific responses. But between 5,000-10,000, something very interesting began to happen. This is best demonstrated by the time we received a detailed solution to a challenge from Kazakhstan within 72 hours of posting it. This person in Kazakhstan realized that based on work he had already done on other problems in his career, he was already very close to a solution. He did not have to solve the problem from scratch. This phenomenon has happened frequently since then, but it does not happen in all cases.

These seeker companies are prone to post intractable problems. They are not yet convinced that this approach will work on their core business, so they are experimenting with problems on the periphery of their businesses that their scientists do not have time to focus on. We are working with Michael Raynor to publish a paper on the basis of innovation and competition in the future. Given a large network, problems that are resolvable will be resolved. The challenge for innovators is to ask the right question at the right time. This is like the old software mantra - given enough eyeballs, all bugs are shallow. This is the scientific version of that.

I will wrap up with a story about one particular solution. A seeker company (not Lilly) posted a problem looking for a polymer that had a very unique and unfamiliar set of properties. Under certain conditions, this polymer will absorb cations (ions with positive charges). Under other circumstances, the polymer needed to release those ions. It had to resist sheer force. There was a very long list of properties that this polymer needed to exhibit. At the end of the day, the senior client agreed to pay five awards. There were five solutions that were of interest to them. They wanted all five to be within their rights-to-practice state. They paid awards to a person who studies carbohydrates in Sweden, a small agribusiness leader, a retired aerospace engineer, a veterinarian, and a transdermal drug delivery systems specialist. I guarantee that they would have found none of those people within their own company. They would have found none of those people if they had done a literature search in the field of interest. They would have found none of them by soliciting input from their consultants.

We have not discussed this next example publicly, and it involves more of the facilitation element that I discussed earlier. We have developed a network of medical professionals, and we have built a core group within our company to work alongside this network. This network configures itself to the clinical study that is under investigation. Members of that network are compensated only for contributions that they make to the network. We are executing our sixth clinical trial using this approach. Our clinical designs have met and exceeded the expectations of our internal review board in terms of elements of ethics, safety, information acquisition and scientific rigor.

Part of the secret goes back to an experience that Kevin Kelly described in his book Out of Control, in which a conference audience is each given a paddle that is green on one side and red on the other. There is a camera in the ceiling pointed at the audience, and this shot is broadcast to the audience. At some point during the conference, the speaker asks the audience to create a block letter M. The audience can see themselves on-screen, and it starts to shimmer between greens and reds. An outline of the letter appears, and then the rest fills in almost instantly. This is the power of giving the participants a global feedback on the state of the system that they are cooperating in. This feedback is the facilitation mechanism. In our medical network, the action of developing a clinical candidate has been translated into imagery that has meaning for participants - they can tell how their individual actions can enhance the progression of that endeavor.

Peter Drucker once said that "The corporation as we know it, which is now 120 years old, is not likely to survive the next 25 years." It will still be an entity on the books, but it will not exist structurally as it is today. We are endeavoring to guess how the corporation of the future will operate economically and structurally in the future. We are experimenting with how business can be even more productive in the future.

Q: When you allocate research dollars, you probably have high-probability targets with low pay-outs and low-probability targets with high pay-outs. Can you use these kinds of markets to sort through the low-probability targets to allocate dollars more effectively?

A: That is our hypothesis. There is a growing literature of articles about the conflict between market mechanisms and our cultural bias towards experts. We are running both methods in parallel for now. Our hope is that in the future, markets will allow us to allocate research dollars much more effectively.

Q: When do you think that the tipping point will come when you tell the experts that you're not going to use them anymore?

A: The realities of productivity and cost pressure on our research model are pushing us to find better ways of making those bets. Early in the research process, there are a large number of opportunities, but most of them will be failures. You will execute them at a relatively modest cost per asset, but there are a lot of assets. You will have to filter out nine for every one that makes it through the first phase. This involves a lot of expenditure. Later on, the process resolves at the end of Phase II into fewer assets, but the cost per asset is much larger. It is in this realm of uncertainty that we incur most of our expenditures. As we feel greater cost pressures, we will have to take a very hard look at cutting down the costs in these early phases. We need to be better students of probability. We need better systems to extract the knowledge that may be distributed across 1,000 minds in the research program.

Q: Who participates in the marketplace?

A: In the example I gave, the participants in the marketplace were about 100 people with deep scientific training, PhD's in most cases, who were knowledgeable about science, but not necessarily experts in the asset being traded. Call them "semi-experts". This is an ideal composition because we're not looking for an individual to lead the group. Instead, we want their trading behavior to bring out their collective knowledge and expertise. In some other examples, we have opened up the markets very broadly. There does not appear to be very much dilution of a market if it has non-expert traders involved. Buy high, sell low behavior does not seem to affect the overall tracking of the marketplace. It should not affect your results to open up the market broadly, but the logistics may be more difficult.

Q: Can you talk about the organizational push-back associated with actually trying to implement a non-hierarchical, collective decision-making mechanism in an environment that has traditionally been quite hierarchical?

A: There has not been an executive who has not thought our idea is cool. But there has also not been an executive who has immediately told their organization that this is the new way of doing things. Everybody wants to watch. Just because it works in Hollywood does not mean that it will work in the pharmaceutical industry or in a sales context. There is a reluctance to transfer what they see to be working in other areas, although they are usually pretty intrigued by it.

We are in the middle of some very deep experimentation. We are creating markets in a variety of contexts, and we are extracting a lot of information about trends, patterns, submarkets and behaviors. We are gathering a lot of knowledge, but I do not think we will be able to walk into an executive with this and convince him to switch. What I hope to achieve is the next step - parallel processing. We want to run our markets and provide the panel of experts with our results as yet another input into their decision-making. The experts will know that they will be held accountable for decisions that they make that contradict the market results. This will help to enroll the experts adopting the market-based approach. This transition will take several years.

Q: I have a three-part question. Is there a certain class of problems that represent more powerful applications of this approach? Second, can you describe the logistics of giving awards to researchers and the dollar-amounts involved? Third, can you describe the nature of IP in this environment and what it will look like in the future?

A: The answers to the first two questions are linked. You need to synchronize the compensation for a problem with the nature of the problem. We have defined the target for Innocentive as the "bounded problem" - a problem for which I can describe the solution in advance, before I have it in hand. Solvers on the site see this description at the third level of the site, and they generally know when they submit their solution whether or not they have really solved it. We receive a lot of answers that admit that they do not solve the complete challenge, but they wish to share their results anyway in case the seekers find them interesting. Scientists are like that sometimes! They know full well that there won't be any compensation, but they want to share everything that they've discovered anyway.

Because we're using a bounded problem, the compensation is very straightforward. The seeker will offer a reward to a single solver. If two solutions arrive, then the award goes to the first one that we receive. Sometimes a seeker will offer multiple awards if they see value in more than one of the solutions. How a group of solvers divide an award is not our problem. If a group of colleagues solves the problem, then we will pay the award to the name of the person who uploaded the solution to our site.

Contrast that with a system we have put in place in our medical organization. We actually do some medical design and protocols. We open it up to both problem identification and solving, and we want to reward both identifiers and solvers. We used a piece of software from a company that is no longer in business. You can work in two ways in this software. You can either make a contribution towards the solution, or you can evaluate someone else's solution. The population of respondents are not only making suggestions but also evaluating the value of everyone else's suggestions. The software then sits back and evaluates what portion of the value of the solution each solver has contributed. If you offer a solution that everyone thinks is a great idea, your value will go up. If your idea is lousy, then your value will go down. The administrator will seed the challenge with an award of, say, $1,000. They would then watch the quality of the dialogue on the site, and as the quality increased, the administrator would add money to the award. The participant could see at any moment what percentage of the award they were entitled to receive based on their contribution.

The last question was about Intellectual Property. One of the executives in our company early on was an IP attorney, and he thought through most of the issues we needed to address. The first people we talk to in each of our seeker companies are the IP attorneys. In all of those cases, we have convinced them that the risks of working with us are equal to or less than the risks of the traditional approach. Invariably, at the end of a certain amount of discussion, they will tell us that they are completely comfortable with our approach.

IP was one of the biggest challenges of turning this into a business. This is quite different than the Nikon chat room where people will tell you for free how to fix your camera. I do not want to downplay how important these issues are.

Q: Do the solvers know who the seekers are and what they're working on?

A: The identities of the seekers are hidden from the solvers. This is one of the ways that we protect the IP of the seekers. When a seeker proposes a problem, we will often go back to them with suggestions for how to mask their problem even better. For example in the drug industry, if you are pretty sure that if you had Molecule A, then you could create Molecule B, then post the challenge for A, not B. Given A, no one could guess that your real goal is to achieve B. We are often surprised by the notion of "hiding in plain sight". People who are trained as experts often think they know things that they really don't. An expert at one conference confronted me with the claim that he knew what every challenge on our site was for. He got about 0.6% of them right! The seeker companies can be discrete about what they choose to disclose.

Q: To what extent are members of the network employed by other big pharmaceutical companies?

A: There is a whole set of IP issues around working for pharmaceutical companies. Most scientists at other firms would have their IP so encumbered that they would not actually be able to participate. We have had a very low fraction of misbehavior on the system. We have had some interesting reactions from other pharma companies, though. They demanded to know how we were going to prevent their employees from solving our problems. In a very polite way, we asked why that was our problem at all.

If I can extrapolate your question, then you are also asking if this detracts from the value of the researcher as a basis for competition. This is a very interesting question. We were talking to another pharmaceutical company. When they heard that we were going to talk about a mechanism by which they could improve the reliability for solving synthetic organic problems, they tried to cancel the meeting. They told us that we were wasting their time. They told us that they had the best staff on the planet. No one could do this better than they could. We asked them for the courtesy of finishing the presentation.

At the end of the presentation, one of their team sat back and said, "I see something scary about this. If you are right (and I'm not convinced that you are), then all of those things I said at the beginning are irrelevant. I am no longer competing on whether I can assemble the best 200 scientists. I am competing with everyone else in the community, and I will never have a fraction that can outperform the collective." They had always traded on the basis that they had the smartest people. So why would we pursue this strategy? For one thing, we believe that this model represents the basis for competition in the future.

The comments, opinions and any forward predictions presented about any particular security, the economy and "the market" are based on the analysis of the speaker. These are not necessarily the opinion of, and should not be construed as a recommendation on the part of Legg Mason Capital Management or any of its affiliates.