The Challenge Network

   back   menu   next   

Valuing information

Valuing information

This web site discusses the knowledge economy. The presumption is made that there is a hierarchy usefulness that runs, roughly, up from the void of noise through data to information, and thence to knowledge, wisdom and other good things to have. We have made much about the loose, unspecified and social nature of knowledge and the tight, fixed nature of specified fact. The section on art, communication and brand discusses the transitions between these forms of knowledge elsewhere. However, over half of the market value of firms is unattributable, save to 'knowledge' and its handmaiden, 'potential'. The tacit understanding by which countries and communities live together is, equally, made up of loosely coupled relationships. How we are to think about these topics is of fundamental, if perhaps less-than-immediate, importance.

Biologists have found that the size of a genome - that is, the amount of complexity which can be encoded within an organism - is directly related to the fidelity with which it is able to copy information and transmit this down the generations. In rough terms, the numbers of units of information that can be 'managed' is inversely proportional to the error rate with which they are transcribed. Manfred Eigen noted that if an organism had a 1% error rate in replicating its genetic material, then it could hold around 100 units of information. Viruses have around 10,000 such units and have an error rate of about 1 in 10,000 per replication. Simple multicellular organisms have around a million units of genetic information and an error rate of one in a million, and complex organisms such as humans are coded in a few billions of units, and have error rates of one in a billion. Faithful management of information therefore defines how complex you can be; and the complexity of a functioning entity - biological or no - is usually equivalent to increased resilience and enhanced adaptability. Such capabilities have a value, and can be assessed as options which one could value were they open to purchase from the marketplace.

In an ideal world, we would be able to calculate the value of an item of information, or the potential value of having such insight, and compare this to the cost of acquiring it. This would allow us to answer the perennial management problem of "how much is enough?" Organisations could be sure that they were spending their resources on just enough of accessing the most useful kinds of knowledge.

We can, in fact, do almost exactly this - but only in extremely defined domains. A financial option, for example, puts a value on resilience or risk reduction. Another way of seeing this is that if there was perfect knowledge of the future - if we could buy the future dollar-yen exchange rate - then this is the value that we would put on that knowledge. (Of course, in the real world, such knowledge would change a great deal of things and the outcome of this would make our original problem look trivial! These are thought experiments, conducted in very limited conditions.) If we are running a project such as drilling for oil, or developing a new pharmaceutical, then we can estimate the risk that new information would abate, and calculate what this knowledge is worth by comparing this to the market-determined cost of insuring against a bad outcome. If we are concerned with generic uncertainty, then we can turn to the capital asset pricing model, where the relative volatility of stock in the market place determines their relative value. It is therefore possible to assess how much a given reduction in volatility in earnings is worth to the marketplace, and so put a value on a measure which would deliver this reduction.

These are, however, artificially closed domains. Most of life is not cast in this form. Further, it is not always - or even often - clear what question we are trying to ask, or what an answer would look like. If we ask how much it is worth for a company to spend on new product development, we can estimate how much it would cost its shareholders if it did not do so. Beyond these grand simplifications, however, matters become extremely fuzzy. Spending this money would not guarantee a successful outcome. Equally, there may be many other ways of attaining the same ends through acquisition or by playing follow-the-leader. There is no implicit guide as to how to spend the money, or how to know whether the outcome is optimal, or even good.

The word "information" is a convenience, but it is not an analytical tool. Many different concepts lie buried within it. One may be valuing the framework within which an item of new knowledge is assessed, or one may be valuing the item itself. Information is contextual, such that freely available data are valuable to organisations which have a contextual "engine" through which to make something of them, and valueless to others who do not.

We need to dig into what we mean by "information". The first section is, unfortunately, heavy going. It tries to synthesise something of what is known about information. We need to understand what we mean by data, information and the system that is transducing information before we can say useful new things. This takes us to some strange outcomes.

The second section draws on this analysis. It extends what has been learned into risk analysis, option planning and accounting for value. What we mean by 'information' is contextual. What is essential data to a mating moth is mere passing air to a currency trader. The value of information is thus also contextual.

What is "information"?

We are accustomed to the clear presentation of facts, embodied as words, data, bits, switch positions, indicator lights. These are, of course, artefacts which we have created in order to clarify the otherwise obscure. We have designed them specifically to reduce the influence of noise and distraction. Physicists and information scientists realise that data and noise merge seamlessly in nature, however, and that the amount of data that can be sent through a conduit is related both to the amount that can be forced down a given pipe, and to the noise in the transmitter, pipeline and receiver.

A clear datum - a nod of the head, a letter on a page - is not information. It is, rather, a token that is exchanged between two systems that wish to exchange information. To an outside observer, data can be said to be made into information by context. If I am looking for my hotel room number, then "42" will means something. If I am looking for a patient's age, then it will mean something different. It will not be data for a passing bird, or for a person seeking an eight-digit telephone number.

It may, however, be information, but be incorrect. My room number may in fact be 76, but this does not stop the transaction from being one in which information is created. Equally, it may be partly heard or vaguely recalled as 'forty-something', a datum partly obscured by noise.

This is all rather complex. At its most abstract, data are encoded over a carrier which 'knows' nothing about what it carries. The information is almost always encoded across many such carriers, and there are firm rules which say how many are needed for a message of a given level of complexity in an environment of a given level of noise. The nature of the carrier is practically important but theoretically irrelevant. A currency trader does not care whether signals are carried by light, electrons or a mixture of both. Terms which describe each of these carriers perfectly may, however, convey nothing about data which they may carry. The only decoder of this is the receiver, where data create information.

This preamble begins to bring us close to being able to say what information 'is'. It is a transaction. The key transactional agent is a system, something that we define more closely below. The data which embody the transaction - patterns arranged across carriers of data - act upon this system so as to change its properties. There are two kinds of change, which we review below. The receiving system has an established relationship with the carriers of the data. Its internal machinery is tuned to filter and otherwise refine the relevant carrier streams. This excludes as non-informational one-off events, as when lightning strikes a tree.

The concept of the 'system' also needs to be analysed. By a system we mean an established set of dynamical relationships between discernible sub-units. How we are able to discern these boundaries is discussed in a moment. Such a system might be a mind or a quantum structure undergoing decoherence, some complex machinery or a marketplace. It contains agents, and these agents interact with each other in ways that create complex properties for the ensemble as a whole. The interaction of simple things so as to create complex outcomes is a commonplace phenomenon termed 'emergence'. Examples abound: consider the joint role of organisms making up an ecology, of gas molecules generating the regularities which we call thermodynamics, or of people cohering into a team or making a market. The system or systems in which they, the individual components are embedded changes what happens to them, and the system which they together comprise has properties which the member components of it do not. No one gas molecule has a 'pressure', for example. As a consequence, a system which is undergoing emergence needs a more complicated description than do the individual parts that make it up, considered before the system forms. However, once the system is in place, the properties of the parts can only be understood if the systems itself is taken into account. This is the case because system has acquired new degrees of freedom, ways of varying, which the parts of it in individual isolation do not possess, but by which they are influenced. Markets affect the participants of markets, and change their behaviour. The greater complexity is directly measurable, as with the discussion of bandwidth needs with which we began.

These peaks and troughs of descriptive complexity allow us to draw objective boundaries around systems and the component parts of systems. Aware minds project boundaries onto the world: we call this a table and that a goldfish bowl. Some of these boundaries of convenience are directly equivalent to those set by objective measures of descriptive complexity.

This brings us to four major implications.

First and as we have already discussed, systems are impinged upon by both brute forces and by information. Brute forces impinge on the system one by one, and they deliver whatever it is that they do to it in isolation from whatever peer forces might also do. We used the example of lightning hitting a tree, above. By contrast, data are delivered when a pattern that is superimposed on these carriers is delivered by the joint consequences of these individual brute-force affects. The one is discriminated from the other by the recipient system. Brute impacts do nothing which could not be expected, ex ante, from a simple inspection of the agent and the target that it will impact. A system which receives data, by contrast, modifies the interactions amongst its constituent parts as a result of the pattern imposed on a number of these impacts. (Of course, just one impact can convey data, but only to a heavily prepared mind and then with great risk of false signals.) Two kinds of modification to the systems can come about, of which more below.

Second, the agents within a system may themselves be made up of systems as well as being contained within a system. They can be 'nested'. A system is likely to be made up of many forms of information transaction amongst its sub-agents; and complex sub-agents may themselves be systems. Nested hierarchies of systems seem to be the norm, with various kinds of information-based transactions occurring at many levels, often in ways which are detected only by the agents immediately involved. For example, the cells in the liver of a monkey carry out all manner of internal and peer-to-peer information-based transactions, whilst simultaneously being a part of an organism and and ecology in which quite distinct transactions are being undertaken. The information exchanged by one part of the system is not information to another, and the data are often not detectable as such.

Third, therefore, we must recognise that information cannot exist without its transducing system. That is, data can be stored if properly encoded, but data are not information. A book is just dirty cellulose without someone to read it. Information exists, therefore, only when systems are reacting to data and only in those systems. Further, one system's data is another system's noise. A whisper in the ear of a currency trader is requires both the whisper and the ear to add value.

Fourth, systems operate to a set of relationships: to what an outside observer would call its rules. Information transactions can change these rules. They can make the system unpredictable. To understand this, we have to take a small diversion.

The nature of dynamic systems has, of course, been much studied. Individual researchers, such as Kauffman, have suggested that there are three broad class of system: those that settle down to inactivity or to frozen patterns of repetition; those with chaotic dynamics and those which lie on a narrow fringe between these states. The middle-ground state 'on the edge of chaos' is very labile to small forces. It responds to perturbations, not least to data. Data, turned into information, can have a disproportionate - indeed, dominant - affect in this region.

The behaviour of a system can be represented by trajectories through a space. This can be the familiar three dimensional space through which we all move and in which we locate objects. It may, however, also be a movement along more abstract dimensions: changing price rather than location, for example, or altering consumer satisfaction. The dimensions in which the system is observed to vary make up what is called the 'state space' of the system. The variously frozen, chaotic or labile characteristics of the behaviour of systems that we have just discussed may, therefore, be confined to limited parts of the state space of a given system, may occupy all of it or none of it. However, where there are labile systems - as above - then a shift in behaviour by these can be thought of a shunting the locus of events from one part of the state space to another. Often, trajectories are confined as if within a bowl, and a shift moves them to a new bowl or bowls. Such a focus for a system is termed an attractor. If one drives a rattling car, one can hear dynamical systems shift from one attractor to another as the frequency of vibration rises. However, in a wide range of systems, the evolution between attractor basins is not smooth, but lagged and sudden.

The emergence of fields of variation, and the existence of attractors within these, gives rise to much of the predictability that we encounter in large objects. The weather varies, but does so within reasonably defined limits. The locus representing the current weather stays within a circumscribed volume of the entire state space of which the atmosphere is capable. However, as the seasons shift, so the rules of what we may expect to encounter alter. Major exceptions tend to be quenched, or tend to lead to a new but equally circumscribed set of behaviours.

This self-healing of the system means that its behaviour remains predictable in terms of a common set of rules that apply everywhere in the space. In the prevalent jargon, the rules are said to be 'symmetrical'. Symmetrical rules are, of course, affected by data flows, which is to say that the system changes what it is doing in response to data inputs. Such changes can be symmetrical, which is to say that they change the behaviour of the systems from one well-understood part of the state space to another, where the same rules apply, symmetrically. Such a response does not change the model that an outside observer would have of the system. It remains predictable; symmetry is not broken.

We have, however, repeated dropped hints that there is another kind of change that can occur to systems. This occurs when the state space itself is altered or extended, such that symmetry is broken. Imagine that an insect, hitherto capable of hopping and crawling can now suddenly fly. A range of new 'ways of being' have to be taken into account, and the result is an altered and probably more complex state space. Matter, hitherto confined to one set of responses and properties, has acquired new ones for a limited, symmetry-breaking configuration that is encoded in systems information.

Evolution, innovation, competition and learning all break symmetry in this way. The means by which state spaces are changed is very often due to in-flowing data, which is turned into information that changes the very properties of the systems on which it impacts. This is both obvious - when one has thought about it - and also quite extraordinary. It means that a predictable system can rendered unpredictable, in that it breaks or extends the rules which one constrained it, and allowed for predictability. Once the change has occurred, one can 'go back down' the system hierarchy, to see what this means for lower order components in the system. However, the atoms now flying around with our insect are doing so by virtue of information changes in the system, not their innate properties.

Laplace suggested that if one knew the position, mass and motion of all the particles in the universe, then one could predict every thing that the universe could do. Aside from the non-classical nature of the universe at very small scales, however, we now know that ordering flows from many sources and in many directions. Currency trading, and changes in currency trading rules, drive electrons through wires; but the rules innate to being an electron in such a system define an overlapping set of reactions and limits. Save in specialised or very simple systems, there is no one governing perspective that allows one to answer 'why?' an event has occurred. How a goldfish got into its bowl in a specific kitchen at a specific time conjures a potentially-endless regress of explanations. Many of these lines of explanation do not knit together, but are mutually necessary. Children like pets. Kitchens are social centres in many households. Goldfish are shiny and golden for complex physiological (and evolutionary, and commercial) reasons. These assorted frameworks of explanation do not 'add together' but are orthogonal variables, which a visitor from Mars would have to assemble before answering to the question to its satisfaction and predictive utility. Lines of explanation intersect in the fact of this fish, however, here and now. Or this share price. Or this brand premium.

Data are encoded at many levels in a system. If the goldfish had inadvertently been placed in the kitchen blender and consequently made into soup, then the quantum information that described its parts - the hydrogen, carbon and so forth - would be conserved, and the blending would consist of an exchange of quantum information. However, the systems information would not be conserved. This systems information is very 'dense', as a thought experiment will suggest. If one wished to reverse events - so making a goldfish from soup - then would require exactly-defined forces to sweep in from infinity, doing so in perfect synchrony so as to collide in ways that exactly reversed the blending. This is both improbable and impossible to orchestrate, given the limiting speed of light. One of the ways of thinking about systems complexity - the creator of systems boundaries, as discussed above - is that it is the sum of systems interactions that got the structure to where it is today. Goldfish are the outcome of something over a billion years of interplay, encoded in genes. Their presence in glass bowls is, however, due to equally complex systems that make people value pets, and which have brought a given particular instance of goldfish information encoded over matter to a given place and time. The blender disperses all of this embedded complexity. This complexity maps onto physical structures, and allows us, as observers to assign boundaries. That is, where a kind of explanation or the complexity of explanation changes, there we draw the boundary that makes the structure in question a 'thing', different from its surrounds as the goldfish is different from the water in which it swims. These boundaries are real, as defined by the repertoire of behaviour that the thing displays, and real in quantifiable terms. The consist, however, entirely of embedded information.

Let us consolidate what we have learned. Information is real - it acts and much be seen as an agency, an independent actor - even if it is intangible. It is expressed through the action of systems. It is transmitted between systems, or stored ready for use by systems, as patterns imposed over carriers. The nature of the carrier is not relevant to the data that it encodes.

Systems operate within a state space that is defined by their ability to act. Their position within this state space is defined by internal transactions, within which data exchanges are often important. Data which they emit to other systems from which they are otherwise decoupled depend on their locus in this state space. The nature of the state space can be changed by information-based events, including data transfer, and the system can change its innate properties in unpredictable ways (although these ways are not mysterious in retrospect, but only not open to prediction from the terms of reference of the system before the change occurred.)

System boundaries are marked by boundaries of the state space - perhaps more easily thought of as the bounds to the validity of the model that makes sense of the system to an observer - and this boundary is also measured by changes in complexity. Complexity can be measured in terms of how complex a path had to be followed to get to the state in question, how complex a set of independent models must be evoked to 'explain' what is happening, how much information is needed to completely map the structure and its properties, or - the same thing, restated, how complex a computer program would be needed to exactly emulate the properties of the system. The Turing test, in which a sentient machine cannot be distinguished from a human respondent by an observer would, therefore, be expected to need a very complex program.

Information is, therefore, a nested hierarchy of relationships that are informed by patterns of interchange, themselves encoded on tokens. Information can act upon itself, changing its rules. Useful data changes information systems either by changing its rules, or by changing what the rules unchanged may nevertheless dictate. Data which are acutely meaningful to one system are meaningless to another. It is pointless to count informational beans, for volume is no pointer to quality. Accounting for value in information is, therefore, a vexed issue.

Valuation of information, contexts, potential

Our gallop through information theory may have left some thrown and others with saddle sores. However, there are some real gains to be made.

This said, a good system can recognise useful data and act upon it. That is, the collective prepared mind can spot opportunities and options. It is more likely to find value in a given tranche of data than a disorganised peer. It is less likely to miss an opportunity altogether or for lack of agility; and it is less likely to walk into traps. This should allow one to put a direct cash value on oversight and preparedness.

There are, however, three essential issues to be faced. First, we do not know how to value knowledge-utilising organisations, and we do not know how to value knowledge acquisition. This leads to enthusiasms for generics - "high tech" - to wild guesses and to the systematic misapplication of resource. There is, for example, no relationship between the growth of sector factor productivity growth in the US and the growth in sector market valuation. Investment in IT has been wildly over-rewarded, whilst investment in people - in training, for example - is hardly rewarded at all.

Second, we do not know how to tell systematically when a human information processing system is working well or badly; and we do not know how to manage or motive for better performance in this. Small firms handle this through intuition and personalised feedback. Large companies have passed through many fashions, but have yet to learn how to do much more than let teams of professionals inspire each other, and to back this up with performance-related incentives. So much, so Sloane.

Third, we do not have the machinery to pick, assess, capture relevant data as information in most large organisations. This malaise extends from environmental scanning and strategic map-making to options picking and partnership management. Above, we have suggested that it is possible to put a formal value on an ideal system, and to estimate the degree to which current situations are less than ideal.

These are fairly profound areas of ignorance. The value that can be placed will be rough, vague, uncertain. It does not have the traditional precision of conventional accountancy. However, this is data seen through noise. We may learn to do better, as discussed elsewhere. Even if we do not, the general state and relative excellence of what is in place can be estimated. One knows which end of the tunnel has the light in it.

 to the top 

The Challenge Network supports the Trek Peru charity.