Some notes about distributed objects, technology-driven change, and the diversity of knowledge. ** But first, a correction. In my comments on Enron earlier today, I said that marketplaces are public goods. That's usually not true, and having fouled up the point, let me take a moment to explain it. According to economics, a public good is a good that has two properties: it is nonexcludable, meaning that once it exists it's impossible to keep people from using it, and nonrival, meaning that many people can use it at once. Pure public goods are rare, but many things come close. National defense is a common example: once the country has been protected, everyone gets the benefit of the protection. Ideas are nearly public goods: you can keep them secret, but once they're out there everyone can use them. The best book about public goods for noneconomists is Inge Kaul, Isabelle Grunberg, Marc Stern, eds, Global Public Goods: International Cooperation in the 21st Century, Oxford: Oxford University Press, 1999. The concept of a public good has several failings, but I'll ignore them for now. Public goods are important because markets can't be expected to provide them. Markets only provide things because someone can make a profit doing so, and if you can't exclude people from using something then you have no way of making a profit from it. For this reason, it is often proposed that public goods be provided from tax money. In fact many people prejudge the issue by conflating the economic concept with the idea of something provided by the government. This doesn't automatically follow, however, for all kinds of reasons. Some public goods are actually bad things -- we're talking about "goods" in the economic sense, not the moral sense, so that ethnic stereotypes are public goods that the government shouldn't be providing. Also, the government should only provide things that the government is likely to do a better job than the market. Many public goods are provided through indirect market mechanisms, e.g., through advertising, as side-effects of the production of non-public goods, and so on. So even though public provision of public goods is often a good idea, each case needs to be analyzed on its own terms. Marketplaces are generally not public goods, for the simple reason that if you can build a fence around something then it's not a public good. So, for example, the flea market might charge each vendor a fee to set up a stand, or it might charge each shopper a fee to enter the fairgrounds where the market is being held. When I used the phrase "public good", I did have some real issues in mind, just not the ones that correspond to that phrase. I was thinking, first, about various laws and customs that require kinds of market-makers to deal with everyone on equal terms. I was also thinking about the problematic nature of many, if not most, decisions about how marketplaces should be paid for. Marketplaces, as I mentioned, tend to exhibit network effects, meaning that the more people participate in them, the more valuable they become. Large marketplaces thus tend to crowd out smaller ones, other things being equal, so that competition among marketplaces is often unstable. A monopoly marketplace can extract rents (i.e., above-competitive prices) from its participants. This can happen through high prices imposed on vendors or sellers, but it can also happen through the marketplace's attempts to interfere with the goods being sold, the terms of sale, the nature of advertising, and so on. When a marketplace does impose such restrictions, though, it can be hard to prove that they are anticompetitive. On the other hand, even if a marketplace is a monopoly, it doesn't automatically follow that the market in marketplaces has failed. It's complicated, is the point, and it's often very unclear what the right answer is. What's worse, the mechanisms for charging people to participate in a marketplace are often determined more by the practicalities of extracting money from them than by the economically optimal answer. Thus, for example, the flea market should ideally charge a "tax" on every transaction rather than charging all participants a flat fee. The flat fee distributes the burden unfairly and thus causes some potential participants to stay home. But it's not practical for a flea market to keep track of everyone's transactions precisely enough to collect taxes. So if we come across a marketplace that pays for itself through a peculiar mechanism that nobody can explain, it need not follow that anything is wrong, though it may well follow that an opportunity exists for someone who can invent a better mousetrap. The flea market makes a good example because it's harmless; we can follow the economic concepts without getting mad. When we move along to the California electricity market, however, things get bad really fast. The complexities of electric power trading are quite amazing. To start with, you've got the power grid, which is a vast, delicate, and extremely dangerous physical object. Even on a good say the power grid only operates because of intense, highly-evolved real-time cooperation among parties spread across half the continent, any one of whom is capable of turning out the lights in Los Angeles on a few seconds' notice. The problem is hard enough when those parties all work for the same organization, but when they work for competing firms then life gets very hard. Furthermore, the economics that you learned in economics classes -- you know, the economics where they assume that the commodity is uniform in every way except maybe the one little way that a particular model is concerned with -- doesn't have much to do with real markets, even in the case of electricity, where you would think that the commodity is just about as uniform as it could possibly be. In fact, electricity markets have a diversity of different kinds of producers, each with its own attributes and thus its own agenda, and every last detail of the very complicated market mechanisms has huge implications for the strategic positions of each of these producers. That's why it helps to buy politicians, and why you want to have the resources to set the agenda and work the negotiations from day one, and get your favorite economists into the loop, and so on. It is far from clear that we know how to organize that kind of process so that it produces a rational outcome. In economics dreamworld it's easy. In this world it's not. We read again last week about how energy trading is "inevitable", and I'm sure it seems inevitable to people who live in the dreamworld of economics. It probably also seems inevitable to people who live in the world where privatization happens whether it makes sense or not. That's a world much closer to our own. But in the real world, "inevitable" is nothing but a refusal to think or choose. I don't know what's going to happen, and I don't even have much of an opinion about it. But I do know that the Enron debacle exposes some predictable systemic difficulties with modern real-time markets. It's partly about the accounting industry, whose absolute perfidy appears to be the one area of indisputable consensus at the moment. But that's just the start. For one thing, I haven't even heard the first breath of the right answer to the problem with accounting: establish shareholder cooperatives to organize audits. We know a great deal about how to govern cooperatives, and even badly governed shareholder cooperatives will do a better job of choosing accounting firms than (for heaven's sake!) the firms that are being audited. And even when we solve that problem, we still haven't begun to think about the institutional demands of the world we're walking into. We can expect to hear a lot of fog in the next few months, because many of the same people who celebrate markets would be terrified to confront the rigors of real ones. It's much better to maintain a bunch of dark corners where they can rip people off. ** The rise of object world. One of those revolutions that keeps not quite happening is called "distributed objects". Although object-oriented programming was invented with the intention of simulating real objects in the world, in practice objects are bundles of data that interact by sending "messages" to one another. Objects are said to be "distributed" when they can reside on any Internet host, keeping track of one another and exchanging messages without any regard to their location. Distributed object standards do exist, but they are not a big part of the average user's life. There are some good reasons for this, but never mind about that. Instead, I want to imagine what the world will be like once standards for distributed objects are widely accepted and implemented, and once the creation and use of distributed objects becomes a routine part of life. Let's say we're having a conversation, and that I recommend a book to you. You'd like to follow up, so I pick up an object that represents the book (technically, a pointer to an instance of the "book" object), and I hand it to you. These actions of "picking up" and "handing" could be physically realized in several ways. For example, I could pick up a small token, wave it over a copy of the book, and click on it with my finger to lock in the book's identifier. (A Xerox PARC concept video shows something like this.) Working out an interface for this idea is a lot harder than telling a story about it, but at least it's one idea about what "picking up" and "handing" might mean. Another possibility is that we could use software that monitors our conversation for objects it can recognize, either interpreting the fluent speech or waiting for us to address it with commands. The software could show each of us a steady stream of candidate objects that it thinks we're referring to, and we could stop occasionally to poke one of those objects, indicating that we want to save it or share it. It would be fascinating to watch people develop conversational conventions for interacting with this device. It would stop being a novelty in short order, and people would conjure objects frequently without making a big deal of it. Software people think by classifying the world into object classes, some of them more abstract than others. This way of thinking is hidden from normal people, but in object world we would have to democratize it. People would become aware of object classes, and communities would have tools to come up with new object classes of their own. (Paul Dourish is doing something like this.) It would be useful to have an object class for any sort of "thing", abstract or concrete, that you want to maintain any kind of relationship with. When you make an airplane reservation, for example, an object could materialize in front of you, and you could drop it into your wallet. That object would include several attributes, all of them tracking the current state of the flight: whether it has been cancelled, when it is scheduled to leave, your seat assignment and meals, and so on. Airplane reservations are an obvious example, already famous from the early-1990s vaporware of "agents", but object world really starts when people start noticing "hey, that's a sort of 'thing' that I would like to track". Some of the "things" would be physical objects, and others would be more abstract. For example, if you are traveling to Omaha on April 22nd and 23rd, you can go to a weather site on the Web and pick up on object for the weather in Omaha on those days. Of course, there will be gnashing of teeth about whether you are going to pick up one object, or two objects, or two objects contained within an abstract "visit" object, or what, but that is the normal gnashing of teeth that programmers engage in all the time. We're just going to democratize the problem. As the date approaches, the weather objects will continually update themselves to the best available forecast. Once you reach Omaha, the objects will start tracking the weather in real time. When your visit is done, you'll have objects that record the weather statistics for Omaha in fine detail. If you care about these objects then you can include them in your diary or trip report. If not then you can toss them. Objects can be complicatedly interlinked. When you buy a car, you would also get a complex, interlinked pile of objects, one for every component of the car. The components that include electronics would be connected in real time to their object representations, so that the objects reflect the current state of the components and some of the components can be controlled remotely by manipulating the objects. When a component is replaced at the repair shop, its object would be discarded and a new object would be linked in with the others. In fact, every auto part in the world would have a unique ID embedded in its corresponding object; when the part is manufactured, the object is manufactured along with it, and the object will then shadow the part forever. When the part is discarded or destroyed, the object will hang around as a record of it. This may sound complicated, but once the necessary standards and infrastructure are in place it will be perfectly routine. Besides, it's already happening on a large scale, and the resulting numbers are data-mined for quality-control and insurance purposes. Once this system goes public, children will grow up assuming that everything has a digital shadow. Objects can be designed to provide different kinds of access to different sorts of people. You could have a "home page" or other public space (perhaps by then the whole idea of a home page will be quaint) where you post some objects for anyone to take, such as your latest publication or the upcoming performance of your theater company. You could also distribute objects that make a wider range of attributes available to some people and a narrower range available to others. Decent graphical tools will allow regular non-programmers to design such things. We are a long way from object world. We could build demonstration versions of it today, and someone surely has. (Do send URL's for papers about any demonstrations.) To make the idea scale, though, real problems need to be solved on several levels. One problem is middleware. Lots of work has been done on distributed object databases, but now we're talking about making billions of objects available in billions of locations, exchanging them over wireless links, and all sorts of hard things. If I have a link to my car's objects in my wallet, where are the objects really stored? How widely are they replicated, and how is that replication managed? Are they really going to be kept constantly up to date while the car is being driven? Will they be updated ten times a second? That's a lot of data. Or will the updates wait until I use the object? How does the object know it is being used? And what if I associate policies with the object that it should take such-and-such an action if such-and-such predicate on the attributes of the car starts holding, for example to detect theft? Once we ask these questions, we come face-to-face with the problem of middleware, the intermediate layers of software that are capable of supporting the full range of answers to these questions that might be reasonable in one context or another. To design middleware you need some idea of the range of applications and the practical problems that arise in them, and that in turn requires a fair amount of painful iteration between the applications level (which wants the middleware to stand still) and the middleware level (which wants to keep issuing new releases forever). Other issues arise at the interface level. The idea is to insinuate objects into the finest corners of everyday life. If we interact with objects only by mousing them on desktops then we have failed. The voice-recognition interface that I sketched above is half an idea of what it would be like to weave object-manipulation into the most ordinary activities. No doubt other interface models will work best in particular settings, and these interfaces will have to work for a wide variety of people. Interface design is relatively easy if you have the user's complete attention, which has been the case for nearly all interfaces so far, but interfaces for people who are mainly focused on something else are harder. Likewise, interfaces are easy to design if you have immobilized the user's body, but they are harder to design if three-quarters of the user's bodily orientation is driven by a practical task in the real world, such as fixing a car. The hardest problems, perhaps, are semantic. It is already a serious problem that people in different communities use words in different ways and thus recognize different categories of things. An object like "the day's weather" can probably mean very different things to athletes, meteorologists, hikers, event planners, and disaster relief workers. The negotiation of those semantic differences has heretofore happened behind the scenes, but in object world it will become more public. The world is full of standards organizations dead-set on what they call "semantic interoperability", meaning that language gets its meaning standardized for the convenience of computers. The whole idea makes my flesh crawl, but in object world we'll definitely have to decide if we want it. ** Logarithmic change. One of the many unexamined assumptions of the cyberfuturists is that exponential growth in the power of computers (which nobody doubts) automatically implies exponential growth in the impacts on society. If you really believe this, and it's implicit in a great deal of public prognostication, for example in the work of Ray Kurzweil, then society is headed for a revolution of inconceivable proportions. But what if it's not true? I have already suggested one reason why social impacts might not be proportional to computing power: there may simply be a limit to the amount of computing that society has any use for. But that argument is no more convincing than the opposite; its purpose is really to balance one implausibility with another. Let me try another counterargument -- or really a counterintuition -- that might work better. The counterintuition is that computing has its "impacts" on society logarithmically. The technological changes might be exponential, but the consequences might be linear. If you list the potential applications of computing, they fall along a spectrum, and that spectrum has a logarithmic feeling to it. One category of applications requires 100 units of computational zorch, the next category requires 1,000 units, the next category requires 10,000 units, and so on, and each category of applications is equally easy for society to absorb. Think, for example, of the computational differences between audio and video. We can now do quite general audio processing in cheap, ubiquitous embedded processors, but the same is not yet true with video. The result is a natural kind of pacing. Society can digest the technological, entrepreneurial, cultural, criminal, and aesthetic possibilities of digital audio processing, and then it can move along to digest the possibilities of video processing. Society will be digesting new digital technologies for a long time, no doubt about it, and we are always getting better at living in a world of continual change. But the technologies will appear in manageable bunches, and we will learn to manage them. It would be helpful to have a model of the space of forthcoming applications of computing. One approach is to think about computing applications in terms of their input sources and output sinks. As to inputs: you can't compute without data, and the data has to come from somewhere. So the first question is how the data is produced. Sometimes this isn't a problem. Nature provides an infinite supply of data, and natural sciences applications (the paradigm case is environmental monitoring) will be able to absorb fresh supplies of computational capacity forever. But what are the output sinks for natural sciences computing? That is, who uses the results? In order to change the world, the outputs need to feed into the social system someplace where they create new opportunities or destabilize old ones. If scientists write research papers based on experiments that employ petabytes of data rather than terabytes, does the world change at all? The world does change somewhat because of the organization required to capture petabytes of data; someone has to install and maintain the sensor arrays. But the institutions of science are already geared to maintaining substantial distributed infrastructures. The picture of the lone scientist in the lab coat was old-fashioned a generation ago. Other examples can be multiplied. Video games that employ gigaflops rather than megaflops are still video games. They fit into the world in the same way, they appeal to the same audience, and they require the same amount of time and attention. People likewise only have a certain number of hours in a day that they can spend talking on the telephone. We can do a much better job of delivering print material to people -- there's no reason why the average citizen can't skim the equivalent of a several hundred books a year -- but again people's capacity maxes out after a while. In my view, the applications of computing that most clearly change the world are the ones that involve the "object world" that I described above. (David Gelernter refers to something similar as the "mirror world", but I've explained elsewhere the many problems with the mirror metaphor.) Consider again the case of the digital car object. That case is perhaps a little frivolous, given that few people need to track the deep workings of their cars. Even so, there's a quiet, profound revolution going on: everything in the world is growing a digital shadow, that is, a data object that reflects its attributes in real time, not to mention its entire history and its various simulated futures. It's a quiet revolution because it's hard to do. It's not going to be an overnight success like the Web. It requires whole layers of existing practices to be ploughed up and replanted, and that means trashing decades of tacit knowledge and innumerable small power bases. In the long run, however, it really does change the world. It allows various activities to unbundle themselves geographically and organizationally and then reassemble themselves in new combinations. (It doesn't mean that the whole world disassembles irreversibly into fragments, however, no matter what you've heard.) It creates whole new power arrangements based on access to objects; it cuts people loose from old social arrangements while binding them into new ones. The question is how significant the social effects of the object world are as a function of the magnitude of the data they involve. Clearly, it will be a long time before everything in the world acquires a full-blown digital shadow. The computing power required to make real-time data on every device in the world available to every other device in the world is several orders of magnitude greater than what's available now. It will happen, given exponential improvements in computing, but it will happen slowly. Assuming that organizational change processes can keep up with technological improvements (perhaps we'll have a lag of undigested technological capacity at some point), we can imagine the object world being built and taken for granted within a few decades. The process has two kinds of terminus: when we've exhausted the input side by mirroring every object in the world, and when we've exhausted the output side by doing everything with the mirrored data that we can possibly (I mean, profitably) imagine doing. Be that as it may, it would still be helpful to have intuitions about the magnitude of the social impact as a function of the magnitude of the inputs. If we track ten times as many parameters of the world, does that cause ten times the consequences? It seems unlikely. For one thing, we can expect people to objectify the high-impact data first -- meaning, the data that produces the highest payoff relative to the effort of capturing it. And some types of data are harder to capture than others, relative to the infrastructure and manufacturing techniques that are available at a given time. It is relatively easy to capture the operating parameters of an auto engine; the engine is already full of computers, all of which will soon be on a general-purpose network. Embedding sensors and processors in every bolt will take longer and result in orders-of-magnitude increases in the amount of available data, but it's hard to imagine that digital bolts will change the structure of the auto industry much more than the first round of objectifying that's under way now. The first round really *will* change the industry, and not just the manufacturers but the insurers, repairers, hobbyists, regulators, and cops. But the next round? I suspect that much of the institutional framework to deal with those data flows will already be in place. We will see. ** Networks and problems. Different fields produce different kinds of knowledge. The idea of a diversity of knowledge, however, intimidates many people; it sounds to them like relativism, as if *anything* can count as knowledge if someone simply says so. That's silly; no such thing follows. Even so, it *is* a hard problem to understand how knowledge functions in society if knowledge is diverse, for example how to tell the difference between quality-control and censorship. The scholars who have argued for the diversity of knowledge, despite the quality of their research, have often been unconcerned with the public-relations problem that their insights suffer. They can win the argument about relativism when they are arguing with people equally as erudite as themselves, but they have historically not done a good job of translating the arguments into a rhetoric that wins public debates. That's partly because they are so concerned to defeat the mythology of unitary knowledge that they emphasize heterogeneity more than they emphasize the limits to heterogeneity. That's too bad, because the diversity of knowledge actually turns out to be related to the Internet's place in society. Let me suggest an intuitive way to think about the differences between different kinds of knowledge. To simplify, I'll stick with academic fields. Every academic field, I will suggest, has two dimensions: problem and network. By the "problem" dimension of knowledge I mean the ways in which research topics are framed as discrete and separable, so that researchers -- whether individuals or teams -- can dig into them and produce publishable results without enaging in far-flung collaborations. By the "network" dimension of knowledge I mean the ways in which researchers organize themselves across geographical and organizational boundaries to integrate experience from many different sites. Every field has its own complexity in both of these dimensions, but often the emphasis is on one dimension or another. As a result, we can roughly and provisionally categorize academic fields as "problem" fields and "network" fields. The prototype of a "problem" field is mathematics. Think of Andrew Wiles, who disappeared into his study for several years to prove Fermat's Last Theorem. The hallmark of "problem" fields is that a research topic has a great deal of internal depth and complexity. The math in Wiles' proof may seem like vast overkill for something so simple as the statement of Fermat's Last Theorem, but you can think of it as an engineering project that finished building a bridge over a conceptual canyon. Publicity value aside, the mathematicians value the bridge because they hope that it's going to carry heavier traffic in the future. Even so, it's not clear that Wiles' type of math represents the future. Math papers are more likely to be coauthored than in the old days, as mathematicians work increasingly by bringing different skills together. This is partly a legacy of the major math project of the 20th century, which aimed at the grand unification of fields rather than producing heavier theorems in a single area. That unification project opened up many seams of potential results along the edges between different areas of math. The increasing practical applicability of even very abstruse areas of math (e.g., in cryptography) didn't hurt either. Even so, math is still weighted toward the "problem" dimension. Math people do form professional networks like anyone else, but the purpose of these networks is not so much to produce the knowledge as to ensure a market for it. The same thing is true in computer science, where professional networks also help with funding. And those are not the only problem fields. Cultural anthropology is a good example. The anthropologist goes to a distant island, spends two years learning the culture, and writes a book that uses it as raw material to explore a particular theoretical problem in depth. The "problem" nature of cultural anthropology is partially an artefact of technology; if long-distance communication is hard then it's easier to uphold the myth that humanity comes sorted into discrete cultures, and a fieldworker who travels great distances to study a culture has no choice but to define a large, solitary research project. But that doesn't change the fact that the best anthropology (and there's a lot of good anthropology being written) has intellectual depth to rival anything being done in computer science, even if the conceptual and methodological foundations of the research could hardly be more different. Contrast these fields to some others: medicine, business, and library science. Medicine, business, and library science may not seem similar on the surface, but they have something important in common: they are all network-oriented. Because they study something that is complex and diverse (illnesses, businesses, and information), they build their knowledge largely by comparing and contrasting cases that arise in professional practice. Physicians don't make their careers by solving deep problems or having profound ideas; they make their careers by building networks that allow them to gather in one central location the phenomenology of a syndrome that has not yet been systematically described. Medical knowledge is all about experience-based patterns. It says, we've seen several hundred people with this problem, we've tried such-and-such treatments on them, and this is what happens. Business is the same way: we've investigated such-and-such an issue in the context of several businesses, and this is the pattern we've discerned. Library science, likewise, is concerned to bring order to the diversity of information as it turns up in the collections of library institutions worldwide. When mathematicians look at business or computer scientists look at library science, they often scoff. They have been taught to value "problems", and they are looking for the particular kind of "depth" that signifies "good work", "real results", and so on. When they don't find what they are looking for, they often become disdainful. The problem is that they are looking in the wrong place. The don't realize that the "problems" that they are familiar with are largely artificial constructions. To fashion those kinds of problems, you need to take several steps back from reality. You're abstracting and simplifying, or more accurately someone else is abstracting and simplifying for you. Many job categories are devoted to suppressing the messy details that threaten to falsify the abstractions of computer science, starting with the clerks whose computer terminals demand that they classify things that refuse to be classified. The dividing-line between computer science and the business-school discipline of "MIS" is especially interesting from this point of view, since the MIS managers are much closer to the intrinsic complexity and diversity of day-to-day business. Computer scientists, as a broad generalization, have little feeling for the complexity and diversity of the real world. That's not to say that they are bad people or defective intellects, only that the field of computer science frames its knowledge in certain ways. It takes all kinds to make a world, and that goes for knowledge as well. We should encourage the creative tension between problem field and network fields, rather than arguing over who is best. Medicine is an interesting case for another reason. Even though problem fields are higher-status than network fields as a broad generalization, medicine is an exception to the rule. If my theory is right, then, why doesn't medicine fall into the same undeservedly low-status bin as business and library science? The reasons are obvious enough. Medicine is a business unto itself -- at UCLA it's half the university's budget -- and it brings money in through patient fees, insurance reimbursements, and Medicare, as well as through research grants and student tuition. Money brings respect, all things being equal, although the increasingly problematic finances of teaching hospitals will test this dynamic in the near future. Medicine is also very aggressive in the way it wields symbols -- it's hard to beat life and death for symbolic value. What's more, business and library schools have stronger competitors than medical schools, so they have a greater incentive to speak in plain English. Precisely because they rely so heavily on symbols, medical schools have never had to explain how their knowledge works in ways that normal people can understand. Professional schools in general tend to produce knowledge that is more network-like than problem-like, but historically they have very often responded to the disdain of the more problem-oriented fields by trying to become more problem-oriented themselves. This strategy is very old; in fact Merton described it perhaps fifty years ago. Unfortunately, it doesn't always work. You end up with professional schools whose faculties are trained in research methods that are disconnected from the needs of their students, or else you end up with factionalized schools that are divided between the scientists and the fieldworkers, or with people whose skills lie in network methods trying to solve problems because that's what the university wants. I think this is all very unfortunate. I'm not saying that every field should be homogenous, and even if everyone does the research they ought to be doing we'll still have the problem of how scholars with incommensurable outlooks can get along. Still, the asymmetry of respect between network knowledge and problem knowledge is most unfortunate. I think the world would be better off if network knowledge were just as venerated as problem knowledge. Before this can happen, we need better metaphors. We are full of metaphors for talking about the wonders if problem knowledge, as we ought to be. When Andrew Wiles can go off in his room and prove Fermat's Last Theorem, that's a good thing, and there's nothing wrong with using the metaphor of "depth" to describe it. It's just that we need metaphors on the other side. So here's a metaphor. I propose that we view the university as the beating heart of the knowledge society. The heart, as we all know, pulls in blue blood from all over the body, sends it over to the lungs until it's nice and red with oxygen, and then pumps it back out into the body. The university does something similar, and the predominant working method of business schools can serve as a good way to explain it. If you read business journals, especially journals such as the Harvard Business Review that are largely aimed at a practitioner audience, you will often see two-by-two matrices with words written in them. These sorts of simple conceptual frameworks (which I've talked about before) are a form of knowledge, but it's not widely understood what form of knowledge they are. Once we understand it, we'll be able to see how the university is like a heart. So let's observe that there are at least two purposes that knowledge can serve: call them abstraction and mediation. Abstraction is the type of knowledge that the West has always venerated from Plato's day forward. It is something that rises above concrete particulars; in fact, it carries the implicit suggestion that concrete particulars are contaminants -- "accidents" is the medieval word -- compared to the fixed, permanent, perfect, essentially mathematical nature of the abstractions. Abstractions generalize; they extract the essence from things. They are an end in themselves. In Plato's theory we were all born literally knowing all possible knowledge already, since access to the ideals (as he called them) was innate. That made questions of epistemology (i.e., the study of the conditions of knowledge) not so urgent as they became subsequently, as the West began to recognize the absurdity of a conception of knowledge that is so completely detached from the material world. But if knowledge can abstract, it can also mediate. The purpose of the two-by-two matrices in the business journals is not to embody any great depth in themselves, the way a theorem or an ethnnography might. Instead, their purpose is to facilitate the creation of new knowledge in situ. Choose a simple conceptual framework (transaction costs, core competencies, structural holes, portfolio effects), and take it out into real cases -- two or more, preferably more. Study what each conceptual framework "picks out" in each case; that is, use the conceptual framework to ask questions, and keep asking questions until you can construct a story that makes sense within the logic of that particular case. That's important: each case has its details, and each case is filled with smart people who have a great deal of practical knowledge of how to make a particular enterprise more or less work. So work up a story that makes sense to them, that fits with their understandings, yet that is framed in terms of the concepts you've brought in. Of course, that might not be possible; your new concepts may not pick out anything real in a particular case, in which you need to get new concepts. But once you've found concepts that let you make sense of several cases, now you can compare and contrast. And that's where the real learning happens. Even with the concepts held constant, each case will tend to foreground some issues while leaving others in the background. Take the issues that are foreground in case A, and translate those issues over to cases B, C, D, and E, asking for each of them what's going on that might correspond to the issue from case A. It doesn't matter whether the other cases are all directly analogous to case A; even if the issue sorts out differently in those other cases, the simple fact that you've thought to ask the question will provoke new thoughts that may never have occurred to anybody before. That's what I mean by the mediating role of knowledge: it mediates the transfer of ideas back and forth between situations in the real world that might not seem at all comparable on the surface. And that's the beating heart: what the university does is fashion concepts that allow ideas to be transferred from one setting to another. Each setting has its own language, so the university invents a lingua franca that gets conversation started among them. At first the ideas will pass through the doors of the university. A researcher will go out to several different sites, gather ideas, bring them home, think about them, and then scatter them in other sites. Eventually the concepts themselves will be exported, so that students who graduate into companies or consulting firms will become beating hearts on their own account. (That's a place where the analogy falters: maybe the university is more like a manufacturer of hearts.) We in modern society take for granted something remarkable: that nearly every site of practice is on both the donating and the receiving end of these mediated transfers of ideas. Often we don't realize it because the people who import ideas by mediation from other fields will often present them full-blown, without bothering to explain where they got them. Other times, a kind of movement will get going whereby researchers and practitioners unite across disciplinary lines around a particular metaphor that they find useful for mediating transfers among themselves: self-organization is one of the fashionable metaphors of the moment. Mediating concepts can be used in various ways, but in general what you see is a mixture of two approaches: explicit comparing/contrasting of particular cases and something that looks more like abstraction. The resulting abstractions, however, usually have no great depth in themselves; their purpose is simply to summarize all of the issues and ideas and themes that have come up in the various cases, so that all of them can be transferred to new situations en masse. This is what "best practices" research is. It's also what physicians do when they codify the knowledge in a particular area of medicine; the human body is too complicated, variable, and inscrutable to really understand in any great depth, and so codified medical knowledge seeks to overwhelm it with a mass of experience loosely organized within some operational concepts and boiled down into procedures that can be taught, and whose results can be further monitored. This is the important thing about network knowledge: it really does operate in networks -- meaning both social networks and infrastructures -- and networks are institutions that have to be built and maintained. In a sense, network knowledge is about surveillance, and mediating concepts exist to render the results of surveillance useful in other places. The mediating role of concepts can help us to explain many things. It is a useful exercise, for example, to deliberately stretch the idea of mediation to situations where its relevance is not obvious. Philosophy, for example, has long been understand as the ultimate abstraction, something very distant from real practice. This is partly a side-effect of the unfortunate professionalization of philosophy that led to the hegemony of analytical philosophy in the English-speaking world perhaps a century ago, but really it dates much further back into the ancient Greek mythologies of ancient times. The popular conception of philosophy as the discipline of asking questions with profound personal meaning is almost completely unrelated to the real practice of philosophy at any time or place in history. There are exceptions. One of Heidegger's motivations, especially in his earliest days, was to reconstruct philosophy around the kinds of profound meanings that he knew from Catholic mysticism. Some political philosophers have tried to make themselves useful to actual concrete social movements. But for the most part, philosophy has been terribly abstract from any real practice. Yet, if we take seriously the mediational role of concepts, then maybe the situation is more complicated. One role of the university is precisely to create concepts that are so abstract that they can mediate transfers of ideas between fields that are very distant indeed. Perhaps we could go back and write a history of the actual sources of scholars' ideas, and maybe we would find that the very abstract concepts that scholars learned in philosophy often helped them to notice analogies that inspire new theories. Analogies have long been recognized as an important source of inspiration for new discoveries, especially in science but in other fields as well, and nothing facilitates the noticing of analogies so efficiently as an abstract idea that can be used to describe many disparate things. I would like to see the university take the mediating role of concepts more seriously. I would like every student to be taught a good-sized repertoire of abstract concepts that have historically proven useful for talking about things in several disparate fields -- examples might include positive and negative feedback, hermeneutics, proof by contradiction, dialectical relationships, equilibrium concepts from physics, evolution by natural selection, and so on -- and teach them not as knowledge from particular fields, but as schemata that help in noticing analogies and mediating the transfer of ideas from one topic to another. The students would be drilled on the use of these concepts to analyze diverse cases, and on comparing and contrasting whatever the analyses turn up, and then they be sent off to take classes in their chosen majors. After a while we could do some intellectual epidemiology to see which of the concepts actually prove useful to the students, and we could gradually evolve the curriculum until we've identified the most powerful concepts. I do realize the problem with this proposal: it is bound to set off power struggles along political lines, and between the sciences and humanities, over the best repertoire of concepts to teach. But that's life. The mediating role of concepts, and network knowledge generally, are also a useful way to re-understand fields that we normally understand mostly in terms of their problem knowledge. (You'll recall that my classification of fields as "network fields" and "problem fields" is a heuristic simplification, and that every field has both dimensions.) What is the network-knowledge dimension of math or computer science? I've already described one role of professional networking in each field, which is to provide an audience for one's work. All research depends on peer review, so it's in your interest to get out there and explain the important of your research to everyone who might be asked to evaluate it. Likewise, if you need funding for your research then you'll probably want to assemble a broad coalition of researchers who explain the significance of their proposed research in similar ways, so that you can approach NSF or the military with a proposition they can understand. But none of that speaks to the network nature of the knowledge itself. What is network-like about knowledge in math and computing? It's true that neither field employs anything like the case method. But they do have something else, which is the effort to build foundations. Much of math during the 20th century, as I mentioned, was organized by the attempt to unify different fields, and that means building networks of people with deep knowledge in different areas. Only then can proposed foundations be tested for their ability to reconstruct the existing knowledge in each area. In computing, the search for foundations takes the form of layering: designing generic computer functionality that can support diverse applications. In that kind of research, it's necessary to work on applications and platforms simultaneously, with the inevitable tensions that I also mentioned above. So in that sense math and computer science have a network dimension, and I think that each field would profit by drawing out and formalizing its network aspects more systematically. Even though anthropology is built on deep case studies, the network nature of its knowledge becomes clearer as you speak with the more sophisticated of its practitioners. Anyone who engages seriously with the depths of real societies is aware that theoretical categories apply differently to different societies, and that there's a limit to how much you can accomplish by spinning theories in abstraction from the particulars of an ethnographic case. I am basically a theorist myself, but I realize that my research -- that is, the theoretical constructs I describe -- is only valuable for the sense it makes of particular cases. So I read case studies, and I try to apply my half-formed concepts to those cases, or else I draw on concepts that have emerged from particular cases, and then I try to do some useful work with them. My work is also influence by personal experience, usually in ways that I don't write about. But I can only go so far before it's time to start testing the concepts against real cases again, and that's why I often move from one topic to another, contributing what I can until I feel like I'm out on a limb, beyond what I can confidently say based on existing case studies and common knowledge. It *is* possible to useful things without being directly engaged with cases, for example pointing out internal inconsistencies in existing theories, sketching new areas of research that other people haven't gotten around to inventing concepts for, noticing patterns that have emerged in the cases so far, or comparing and contrasting theoretical notions that have arisen in different contexts. But if you believe that theory can blast off into space without any mooring in real cases then you're likely to do the sort of pretentious big-T Theory that gives us all a bad name. Anthropologists are thoroughly infused with that understanding, and so the best ones really do refuse abstraction. They see their theoretical constructs very much as ways of mediating between different sites. Their concern is not practical, so they are not interested in moving ideas from one site to another on a material level. They are usually not trying to help the people they study. Rather, they are interested in describing the fullness of the social reality they find in a given place, and like the business people they understand that the real test is the extent to which their story about a particular case makes internal sense. Granted, they are less concerned than the business people to be understandable to the people they are studying, although that too is changing as the "natives" become more worldly themselves, and as it becomes more acceptable by slow degrees to study "us" as well as "them". In any case, I think that the anthropologists' relationship to theory is healthy, and I wish I could teach it to people in other fields. Anthropology is also becoming more network- like as reality becomes more network-like, and as the myth of discrete cultures becomes more and more of an anachronism, but that's a topic for another time. Knowledge is diverse because reality is diverse. In fact, reality is diverse on two levels. A field like medicine, business, or library science derives knowledge by working across the diversity of illnesses, businesses, and information, gathering more or less commensurable examples of each under relatively useful headings that can be used as to codify and monitor practice. And then the various fields themselves are diverse: they are diverse in diverse ways. Fields that pride themselves on abstraction operate by suppressing and ignoring diversity. That can be okay as a heuristic means of producing one kind of knowledge -- knowledge that edits the world in one particular way, and that can be useful when recombined with knowledge that edits the world in other ways. But it's harmful when abstraction is mistaken for truth, and when fields that refuse to abstract away crucial aspects of reality are disparaged as superficial compared to the artificial depth at the other end of campus. Let's keep inventing metaphors that make network-oriented fields sound just as prestigious and heroic as problem-oriented fields. The point, of course, is not just to mindlessly praise the work, since bad research can be done anywhere. The point, rather, is to render intuitive the standards that can and should guide us in evaluating research of diverse types. If we don't, then we will disserve ourselves by applying standards that don't fit, or else no standards at all. end