Reading :: Knowledge Assets

Knowledge Assets: Securing Competitive Advantage in the Information Economy
By Max Boisot

Recently I reviewed Boisot's 1987 Information and Organizations, which—despite some really interesting ideas—borrowed a bit indiscriminately from various disciplines and used a style I found off-putting. But in 1998's Knowledge Assets, Boisot really hits his stride, producing a better structured, more coherent, and more stylistically confident book. It won the 2000 Igor Ansoff Strategy Prize, and it appears to be Boisot's most heavily cited work. 

Don't get me wrong — Boisot's work still rests on Shannon's communication theory, still thinks about information in terms of Newtonian physics, and thus still has a very thin concept of audience characterized vaguely as "populations." It still doesn't have an account of how analytic-systemic representations develop. It still seems to have no place for polysemy. It still is presented as a conceptual frame, avoiding the messiness and challenges that come from applying a framework to empirical cases. For someone who reads a lot of sociocultural theory, who thinks a lot about development and ambiguity and interpretation, and who attempts to apply theory to qualitative cases—like me—Boisot's work leaves a lot to be desired. Still, within his assumptions, he has done a great deal of work here and has produced an internally coherent, instructive account of information (and to some extent knowledge). I admire it in the same way that I might admire the theological reasoning produced by someone from a different religion: he may start from different precepts than mine, but the work he based on those precepts is internally sound and perhaps some of it can be analogically applied to mine.

So with that preface, let's get into the book. The blurb on the back of the book sums it up:

... Max Boisot provides a conceptual framework for managers and students that will help them explore and understand how knowledge and information assets differ from physical assets, and how to deal with them at a strategic level within their organizations.

In the introduction, Boisot lays the foundation by defining 

• technology: "sociotechnical systems configured so as to produce certain specific types of physical effects"
• competence: "the organizational and technical skills involved in achieving a certain level of performance in the production of such effects"
• capability: "a strategic skill in the application and integration of competences"
• complexity:  "the number of elements in interaction and the number of different states that those interactions can give rise to" (p.5)

He adds: "Clearly, the number and nature of technologies that have to be integrated into competences, and the number and variety of these that have to be mobilized to achieve a capability, will determine the level of complexity that a firm has to deal with" (p.5). That's important because complexity is increasing in firms (p.5), leading managers to cope with complexity "by developing abstract models that help them to make sense of the complexity and reduce it to manageable proportions" (p.6). At the time Boisot is writing, in the late 1990s, rising complexity has precipitated a crisis: "We are entering the information economy still firmly strapped to the paradigms of the energy economy," and managers need to move to a new paradigm to manage adequately (p.7). Specifically, the energy economy assumes linear processes, but the world is nonlinear, requiring an understanding of complexity (p.9). 

To gain that understanding, he distinguishes among three things that are often confused:

• data: "a discrimination between physical states ... that may or may not convey information to an agent"—depending on the agent's "stock of knowledge"
• information: "that subset of the data residing in things that activates an agent — it is filtered from the data by the agent's perceptual or conceptual apparatus .... [it] establishes a relationship between things and agents"
• knowledge "can be conceptualized as a set of probability distributions held by an agent and orienting his or her actions. These either consolidate or undergo modification with the arrival of new information. In contrast to information, knowledge cannot be directly observed. Its existence can only be inferred from the action of agents. It follows from this that knowledge assets cannot be directly observed either; they therefore have to be apprehended indirectly" (p.12)

Knowledge "economizes on the use of physical resources" (p.12):

• "by in-forming them—i.e., by embedding itself in physical artefacts and structures"; ex: standardized bricks (p.12)
• "by organizing them—i.e., by embedding itself as information in documents and symbolic support systems used to coordinate the creation or functioning of artefacts"; ex: house plans, read in conjunction with detailed specs and a budget (p.13)
• "by enhancing the understanding of intelligent agents that interact with physical resources—i.e., by embedding itself in the brains of individuals or organizations"; ex: the "architect draws on an accumulated stock of knowledge which reflects a collective understanding of human behavior in space" etc. etc. (p.13).

He adds:

In short, knowledge held by agents builds up the information structures latent in physical things, in documents, or in individual brains. Knowledge assets are those accumulations that yield a stream of useful services over time while economizing on the consumption of physical resources—i.e., minimizing the rate of entropy production. (p.13)

He argues that we can classify knowledge assets along two dimensions: 

• How far can they be given form, i.e., codified? (ex: mass-produced artifacts vs. discursive remarks) (p.13)
• How much can they be abstracted, i.e.,  applying to many situations? 

Codification and abstraction are two interrelated ways to economize on information processing, lowering the cost of converting usable knowledge to knowledge assets (p.14). 

In the next chapter, Boisot says, "we shall propose a way of integrating physical and information phenomena in a single unified representation," using the production function from neoclassical economics (p.19). To do so, he reiterates the differences among data, information, and knowledge:

• Data: "a discernable difference between alternative states of a system" (p.19)
• Information: "Data that modifies the expectations or the conditional readiness of an observer" (p.20)
• Knowledge: "the set of expectations that an observer holds with respect to an event" (p.20). 

So knowledge assets are "that subset of dispositions to act that is embedded in individuals, groups, or artefacts that have value-adding potential" (p.20). 

Boisot charges that Marx thought of capital as congealed labor; he "held a strictly energy-based view of economic processes" (p.21). But, he says, in industrial societies we rely more on information—and that means that we must "economize on the consumption of data as well as that of physical resources. How do we do this? By extracting information from data and then junking the latter" (p.29, his emphasis). We abstract patterns from the data, then focus on the patterns.

This tendency allows us to do more, but it also involves complexity reduction, and that carries dangers. "[A]bove a certain level of complexity, we face chaos" as we cannot "effectively process the amount of data we are confronted with at the speed it requires" (p.37). At the lower bound, we are faced with "excessive order—characterized by an undersupply of data" (p.37). Effective learning happens in between these, at "the edge of chaos," which is "a region that complex systems are drawn to in their quest for dynamic stability" (p.37).

This brings us to the third chapter, on the I-Space. Boisot refreshes us on codification, which "can usefully be thought of as a process of giving form to phenomena or to experience" (p.41, his emphasis), and on abstraction, "the process of discerning the structures that underlie the forms" (p.41). Working together, these reduce an agent's "data-processing load" and "facilitate communication processes and hence the diffusion of information" (p.41 — my emphasis this time). 

Codification, he says, "constitutes a selection from competing perceptual and conceptual alternatives," and this selection "is often conflict-laden" (p.44). Higher codification economizes on data-processing resources, but it also results in lost flexibility and options (p.47). It fossilizes hard-won skills, and thus also deskills skilled craftspeople (p.47). 

Whereas codification gives form to phenomena, abstraction gives them structure (p.48). Abstraction saves on data-processing resources by "minimizing the number of categories that we need to draw on for a given task. Abstraction then works by teasing out the underlying structure of phenomena relevant to our purpose" (p.49). "Abstraction, in effect, is a form of reductionism: it works by letting the one stand for the many" (p.50). 

Diffusion is the availability of data and information. It's different from deliberate uptake: information can be diffused (made available more broadly), but not taken up by agents (not adopted) (p.52). Citing Shannon and Weaver, Boisot notes three problems that can threaten diffusion:

1. "Is the message received the same as the message sent?" (technical level)
2. "Is the message received understood?" (semantic level) 
3. "Is the message received acted upon as intended?" (pragmatic level) (p.53)

We then get to the I-Space, the cube diagram with the axes of codification, abstraction, and diffusion (p.56). Boisot contrasts it with Nonaka and Takeuchi, with one differentiator being that I-Space posits three types of tacit knowledge: things that are not said because

• everybody understands them
• nobody understands them
• some people understand them but cannot articulate them without cost (p.57)

He draws on the example of military technologies that a French firm transferred to the Iraqi military, assuming that the knowledge explicit in specifications and diagrams was adequate. They were incorrect (p.57). Reading this example, I thought of the ANT studies on the Zimbabwe Bush Pump and the gazogene—and I wondered where the "outside" of the I-Space cube was. In other words, what bounds the cube and the analysis? Who is sharing knowledge assets? Clearly not all agents have the same knowledge stock (we'll use that loose term since Boisot does), so the same information asset will not occupy the same space for all of them. Unfortunately, Boisot only addresses this (critical) bounding question in an offhand way—mentioning "individuals" and "public knowledge" here, and "populations" elsewhere. This becomes a real problem here, because he posits a "social learning cycle" (pp.58-66) without specifying the bounds of the social. 

The social learning cycle posits six movements:

1. Scanning for patterns
2. Problem-Solving by giving structure and coherence to the insights from scanning
3. Abstraction or "generalizing the application of newly codified insights to a wider range of situations"
4. Diffusion or "sharing the newly created insights with a target population" (notice that the assumption seems to be that an individual agent is performing these steps — not necessarily an individual person, perhaps an individual organization)
5. Absorption or "applying the newly codified insights to different situations in a 'learning-by-doing' or a 'learning-by-using' fashion," eventually acquiring uncodified "context" around it
6. Impacting, in which abstract knowledge is embedded in concrete practices (pp.59-61)

Boisot adds that these steps can run concurrently (p.61). And as mentioned, he characterizes the I-Space as applying to a "population" (p.62), acknowleging that sometimes "data can enter the I-Space from the outside." A few quick notes of caution here:

• The social learning cycle is portrayed as individual (although that individual can be a collective agent); we don't get a sense of how social or cultural knowledge might weigh across boundaries.
• The social learning cycle is also portrayed as intentional, in which problems lead to problem-solving and deliberate steps to improve the situation. We don't get a good account of, for instance, genre evolution in which small, collective, largely undirected changes result in the emergence of a more coherent problem space—cf. Bazerman's Shaping Written Knowledge. This point is striking in light of Boisot's many appeals to complexity theory!
• The social learning cycle is depicted as occurring "within" I-Space, specifically the I-Space of a given "population." That population is not discussed further, but I don't view it as nearly enough of a qualifier. My field (writing studies) made a similar handwaving generalization in the early 1990s, talking about "discourse communities" or populations that shared the same discourse. But the fact that people talk similarly or share similar characteristics is not enough to bind them together—to be meaningful in terms of social learning, they have to cross-reference each other. Social theories provide various ways to do this: for instance, activity theory bounds the case by identifying a shared object that people are laboring to transform, while qualitative case studies bound their cases by looking for formal organizations or identifiable types of interactions. This appeal to "populations" does not do that, instead waving the hard problem away. Without solving this problem of bounding, the I-Space really can't address social learning effectively.

Moving on, Boisot proposes to use the I-Space to represent knowledge assets. He states that "assets are stocks rather than flows and we have seen that knowledge assets can be stocked in people's heads, in documents, or in artefacts" (p.63). He provides a scaling guide (a table) to help us understand what high, medium, or low codification, abstraction, and diffusion might look like (p.65). He ends the chapter by noting that organizations seek minimum entropy, but they can't stay in that state (p.67).

The next chapter is on the paradox of value. Here, he doubles down on his previous assertion, arguing that information goods are naturally scarce only when "they are deeply embedded in some physical substrate that is limited in space and time," such as "individual brains and certain physical products" (p.71). This understanding of brains as a substrate for information is quite problematic from a sociocultural standpoint. Among other things, he sees knowledge as assets —something people have and save in "stocks"—rather than something we collectively do.  

In any case, he uses the I-Space cube to map maximum and minimum value. The minimal value is concrete, uncodified, and highly diffused—it's broadly known, but can't really be put into words or applied beyond a single concrete domain. Think in terms of lore. The max value is abstract, codified, and undiffused (p.79)—think of a formula that is a trade secret. But consider an issue that Boisot isn't mapping here, which is that any conceivable I-Space—let's take a company, since companies have trade secrets—will necessarily have multiple information or knowledge assets at different levels of codification, abstraction, and diffusion, and these will necessarily interact with each other (a point he makes earlier in the book when talking about reference documents), and that different agents in the same space will have different uptakes. Companies have both trade secrets and lore, and everything in between, and value can (and I'd argue, usually does) emerge from the relationships among them rather than from one specific asset. Boisot has earlier conceded that different levels of abstraction and codification yield different benefits, but seems to see social learning as a series of transformations of one piece of information rather than creating connections across several different representations oriented to different aspects of a problem space. Again, he's hindered by the vague understanding of what makes up the cube depicted in the I-Space.

He goes on to use the I-Space cube to describe base, key, and emergent technologies (p.85). Truthfully, Boisot ends up using this cube for everything: the social learning cycle, value, technology, governance, etc. The cube is so flexible partly because the axes are abstract, the points of reference are unspecified, and the space is vaguely bounded by 'populations." 

In the next chapter, he contrasts Newtonian learning (an equilibriating process with change coming from the outside) with Schumpterian learning (which explores the potential of nonlinear phenomena). He provides the example of an "industry-level SLC with a handful of players" (p.105), underscoring (in my opinion) the unit-of-analysis problem that he has fallen into. This is the weakest chapter in the book, in my view.

The next chapter addresses culture as a knowledge asset. Boisot argues that "only a small part of what we call cultural knowledge gets itself embedded in technologies and artefacts. A large part is embodied in social processes, institutional practices, and traditions, many of which are carried around in people's heads. For this reason most cultural knowledge has tended to be taken for granted rather than treated as an asset to be prized and exploited" (p.117). He characterizes culture as "a kind of collective memory" (p.120) and cautions that all substrates are subject to entropy, including the substrate of a human brain, which can go senile (p.120). Fortunately the I-Space — I'm sure you didn't see this coming, dear reader — lends itself well to the study of cultural transmission in Boisot's view (p.122). He suggests different I-Spaces for different subcultures (p.123). For cultural action, he suggests taking transactions as our unit of analysis (p.124). And that gets us to what he characterizes as types of transactions but what others have characterized as organization types or institutional logics: fiefs, clans, bureaucracies, and markets (pp.126-127). 

Later, he returns to the question of population raised earlier, arguing that one population is the environment for another, and that the larger a population is, the more abstract is the level that binds them together (p.136). He goes on to map natural cultures in the I-Space (p.142). 

Moving on, in the next chapter, he discusses products, technologies, and organizations. Here, he says that although we can "choose to treat firms as data-processing agents for I-Space purposes, but this is a convenient fiction that should not blind us to the fact that human agents are where the action is" (p.155). Ultimately, agents are either individual humans or aggregations of individual humans, possessing knowledge in the same way. For aggregations, knowledge assets are located in substrates including physical objects, data in documents, and human heads (p.155); this assertion is illustrated on p.156, fig.7.1, where one knowledge substrate is labeled "Heads." Later, Boisot pictures the migration of knowledge assets from head to documents and objects over time (p.165). 

Skipping a bit: In Chapter 10, Boisot asserts that the conceptual framework of the I-Space has been "tested in the field by users" and is undergoing "laboratory-like tests" (p.231). The field testing appears to be done in two-day workshops in which managers use the I-Space to map out knowledge assets (p.231). In a sense, then, these are like Engestrom's Change Labs, in which the conceptual framework is used to spark conversations and collective problem-solving — although Engestrom typically tries to involve a cross-section of stakeholders rather than just managers. 

In sum, this was a fascinating book, though as my notes suggest, it is heterodox to a sociocultural understanding of knowledge and information. I-Space has great potential for strategically mapping assets, but I'm not sure its fundamental understanding of information can allow it to reach that potential.

Let's illustrate with a simple case: a recipe you find online, connected to a homespun narrative about making the food, and illustrated with a simple video. Specifically, I see these problems:

• Frame of reference. What's the bound case in which information and knowledge are at play? Are we examining how end consumers will experience the assets, or are we looking at how the recipe genre, the HTML, the ASCII, and the MOV encodings are decoded by the computer? Is the video considered highly codified (MOV format) or relatively uncodified (conveying gestures)?
• Audience. "Population" isn't enough because any agent can be a member of multiple overlapping populations. Consider the COO of a family business. How is the I-Space bounded?
• Knowledge. It's externalized, putting off interpretation as outside the system. Even "uptake" is only mentioned briefly and offhandedly at the beginning.
• Knowledge is treated as a noun, an asset, rather than a verb, a practice.
• Knowledge is additionally treated as embedded in substrates, with one of those substrates being people's heads. This characterization is hugely problematic for cultural psychologists, sociologists, and others.
• Learning is treated as movement from one point in the I-Space to another. But these assets coexist and relate to each other,
• Information is treated as anchored within the I-Space, but in empirical studies, we often see information artifacts that share associations across multiple frames of reference (ex: a worker learns a calendaring system at her university, then continues using it in her first job even though that job's environment has different assumptions and constraints). We don't get an account of such associations or how they conflict. 
• Information is treated as distinct within the I-Space, but in empirical studies, we typically see people relating different information sources together. Knowledge doesn't just come from transforming individual representations but also from relating multiple representations.

Still, I find the I-Space to be incredibly interesting and potentially very fecund. I'll keep thinking about it and seeing if I can find ways to productively transform these insights.