The problem of information

  In this chapter we will use the word "information" in its conventional sense as a reduction of uncertainty‹ in a set of choices defined within a stable context.  The context of information must define at least two options: 0 or 1 in the case of a computer, or "war" and "peace" in the case of a diplomatic code. Any arbitrary amount of information can be coded as a combination of binary codes. This is of course the well-established mathematical definition that has been so influential in communication engineering and computer science, except that we will be using it qualitatively, and in a somewhat broader sense. Over the past few decades, the mathematical definition has also found increasing application in psychology. It has been found useful in modeling fundamental findings about reaction time , classical conditioning, basic level categories , perceptual complexity. Thus the mathematical notion of information seems to have some psychological reality.   How is information in this sense related to consciousness? There is good evidence that we are conscious of an event only when it exists in a stable context, but not when it is so predictable that there are no conceivable alternatives to it. Conscious experience seems to exist only when there are some  degrees of freedom within a stable context. Thus the notion of reduction of uncertainty in a stable context seems appropriate. Information is inherently context-dependent, and we have presented a set of arguments before  that consciousness is also highly context-dependent.  Conscious experience of the world is not a direct function of physical stimulation. The same physical stimulus repeated over and over again will soon become less informative --- and also less conscious. But a highly significant or variable stimulus habituates more slowly. We therefore need to make a distinction between physical stimulation‹ and real information‹. On the other side of that coin, this is the difference between repetition‹ and redundancy‹. The same physical energy impinging on the same sensory receptors may be either informative or not, depending upon the reduction of uncertainty in the relevant context. Sometimes the physical absence of an expected stimulus can provide information, just as its presence may be redundant. In this sentence, we need only omit one ... to show that the absence of a stimulus can draw our attention --- and the missing item may well become conscious for the reader. Thus information and stimulation are not the same; they can vary independently.  In general, the probability of being conscious of any event

In Pavlov's conditioning experiments, when the sound of the bell signaled that food was coming (a significance beyond itself), the hungry dog was much more likely to prick up its ears, the  orienting response to the bell took longer to habituate, and learningoccurred more quickly. One way to think about significance is in terms of purposes the hungry dog is likely to have, which create goal contexts for its perceptual systems to explore. Significant information can then be seen as a reduction of uncertainty within a goal context. Thus the concept of information can be related naturally to the things that matter most to an animal or human.  We can think of information as existing  at different levels, just as we have previously suggested that contexts exist at different levels .

  The strongest argument for the close relationship between information and consciousness is the existence of what we will call Redundancy Effects. Redundancy, the absence of information, is defined in information theory as the transmission of a signal after the uncertainty at the receiver is already zero. The choice between "peace" and "war" had great information value in 1945 for most of the world, but repeating the word "peace" over and over again after that point became increasingly less informative, even though the context of subsequent events is accurately described by that word. Thus the word "peace" became increasingly redundant, but not false. There are many well-known cases in which conscious input fades with repetition --- cases like stimulus habituation, automatization of skills and mental images, perceptual adaptation, shifts in the Adaptation Level of perceptual and judgment categories, "blank-outs" in the Ganzfeld, semantic satiation, loss of access to stable conceptual knowledge, etc., etc. These phenomena allow us to do a contrastive analysis, showing a direct relationship between conscious experience and the informativeness of an event. 

     Habituation of awareness to a repeated stimulus is the most obvious example of a Redundancy Effect. At this moment the reader is likely to be habituated to the feeling of the chair, the color and intensity of the ambient light and background sounds, the language of this book, and many other predictable features of the inner and outer world. A previous chapter detailed Sokolov's arguments for the continued existence of unconscious representations of habituated stimuli. Sokolov argued a mismatch in any parameter of a habituated stimulus will elicit a new Orienting Response. To detect such mismatches, we must maintain some sort of representation of the expected input. But this representation does not elicit an Orienting Response --- or, in the terms used in this book, it is not normally conscious. Thus there must be an unconscious representation of a habituated stimulus which is similar in many respects to the conscious perceptual representation of the same stimulus when it first occurs.

Information vs. novelty

                    What can be the difference between the conscious and unconscious representations of the same stimulus? Several writers have suggested that novelty or mismatch with expectations is involved in consciousness perception. It is sometimes maintained that there must be a mismatch between input and expectations for a stimulus to be conscious. This is certainly true in the case of surprise, as discussed above. But it cannot be true without qualification. Any stimulus that violates previous expectation can only do so in a context which is itself  not violated --- if input were to violate every expectation, if it were totally new, it could not be experienced at all. Therefore all understandable novelty exists within a relatively stable context that is not novel. 

 The opposite argument has also been offered. Marcel suggests that a match between input and memory is required for perceptual input to be conscious. But this cannot becompletely true either: if there is a perfect match between input and expectation, we have the case of habituation and loss of consciousness. We find ourselves in a middling position with respect to both the match and the mismatch hypothesis: yes, there should be some sort of match, but not too much. Both the mismatch and the match hypothesis capture some, but not all, of the evidence.    We will develop the argument that the notion of information is more attractive than either simple match or mismatch. Information involves both a match of context and a mismatch of the stimulus details.  It further suggests that the input must be useful, in the sense that many systems can reduce their uncertainty relative to it. It also allows us to include the notion of significance as a reduction of uncertainty in a relevant goal context. And it seems to explain the well-established Redundancy Effects.   If the concept of information is indeed preferable, what about the case of surprise, which is indeed a mismatch of input and expectations? Mismatch reduces to a special case of information --- it is the case where the context of the expected input is itself violated. This context then becomes "decontextualized" , and its components must be reassembled in the stable framework of a higher-level context. The point here is that the notion of information seems wellcsuited to handle a number of important properties of conscious input; it can also explain mismatch phenomena like surprise.

 Much of the argument depends upon the Redundancy Effects, those cases where repetition leads to a loss of conscious experience. There are some apparent counter-examples to the Redundancy Effects: cases where repeated events do not fade, or where they seem to become more consciously available with practice. For instance, conscious access to highly significant or unpredictable events is lost only slowly. In the case of chronic pain people do not lose conscious access at all. We suggest that these facts reflect the special role of signficant information. But as we have mentioned, significance can be treated as a reduction of uncertainty in a higher-level goal context. Further, there are cases in which practice seems to access to conscious events. For instance, practicing recall will bring memorized material to mind more readily and practicing visual search will cause the search target to "pop" into consciousness quite involuntarily . Notice however, that what is being practiced here is not the visual or memory target, but the process of search or recall‹. That is, in these cases we gain automaticity in the skill of controlling access to  consciousness, but the input that will become conscious is not predictable, and may be quite novel and informative.  Thus, these facts do not contradict the claim that consciousness requires informative input. Indeed, the process of recall or search itself does become automatic and unconscious with practice. Only its result remain informative and conscious.

 There are also cases where repeated stimuli fade from consciousness, only to return in a new guise. We conclude that Redundancy Effects are both powerful and pervasive, while apparent counter-examples can be explained plausibly in an extended information-theoretic framework. All this supports the idea that informativeness may be a necessary condition‹ for conscious experience of some event. This viewpoint also suggest a new perspective on context: In a sense, context consists of those things to which the nervous system has already adapted; it is the ground against which new information is defined.

   Our use of the term "information" is similar to the classical mathematical definition developed by Shannon and others  but we should make note of some possible differences. Psychological contexts are highly complex knowledge structures that have many more dimensions than the simple, binary, one-dimensional message contexts of classical information theory. But of course, we know that knowledge structures of any dimensionality and complexity can be reduced in principle to binary choices. Further, the classical definition presumes that reduction of uncertainty takes place in a stable context of choices; but psychologically, we know that contexts are not totally stable, but adapt to informative input whenever possible. The nervous system learns about predictable inputs; it is not passive like the contexts of conventional information theory. We will argue below, however, that conscious experience is associated with a range of phenomena in which the context of informative choices is relatively stable. Within these relatively stable contexts, the classical definition is quite useful.  Finally, the formal definition of information is quantitative, but we will not develop a quantitative approach here. Quantification at this stage can apply only to a small, experimentally defined subset of the full range of phenomena. Others have already done this. We will focus here on the making a case for the special relationship between consciousness and information in general.

     Adaptation                             

                    After information, the second major concept in this chapter is adaptation. Here we will use it in a narrow sense, as the process of learning to represent some input --- to know and understand it to the point of automatic predictability. Learning to represent something involves, of course, a reduction of uncertainty (i.e., information). When there is a perfect match between input and its representation, the input is redundant with respect to its representation. Thus  redundancy is the end-product of successful adaptation.  We can borrow Piagetian terms here to represent different ends of the adaptation continuum (Piaget, 1952). When confronted with a situation that is new and strange, people need to find new contexts for experiencing the input; the result resembles Piagetian accomodation. In other words, accomodation has to do with the discovery of usable contexts. On the other end of the continuum, when the input is highly familiar and predictable, minimal adaptation is required, so that we can assimilate it into readily available contexts. In the extreme case of redundancy, context and input match exactly.              

                    Conscious experience of an event seems to occur midway between the outer poles of assimilation and accomodation. If we can automatically predict something completely, we are not conscious of it. But if the input requires a deep revision of our current contexts, we do not  experience it either --- it is too confusing or disorganized to experience as such, though we may experience fragments and tentative interpretations of the input. Somewhere between these two extremes, between the assimilation and accomodation poles of the continuum, we may have an accurate conscious experience of the event. From the adaptation point of view, an informative conscious event translates into a demand for adaptation. This is of course the claim stated in the chapter title: that conscious experience is informative --- it always demands some degree of adaptation.                  

                    In sum, there is a close web of  connections between certain fundamental ideas: information, consciousness, reduction of uncertainty, a drop in contextually defined alternatives,  a demand for adaptation and learning, a gain in predictability and redundancy, and the creation of new contexts. Adaptation takes place over time, and we develop now the notion that conscious experience corresponds to a certain stage of the "adaptation cycle" --- namely, the stage where there is a relatively stable context for understanding the input, but there is still uncertainty to be reduced within that context. Many processors can cooperate in reducing the uncertainty. A fundamental point is that the nervous system is always in dynamic adaptive activity --- it is always working to adapt to conscious input --- even when we seem to be conscious of only a single thing. Conscious experience cannot be understood apart from this process of dynamic adaptation. We turn to this issue next.                 

               The adaptation cycle: Any learnable task goes from context- 

        creation to conscious information to redundancy. 

                    In learning about a new source of knowledge we often start with considerable uncertainty and confusion. By paying attention to the problem, a sense of clarity is often gained, as we become more conscious of what is to be learned. Finally, with practice, the material becomes highly predictable and fades from consciousness. These three stages make up what we will call the adaptation cycle: starting only  with the knowledge that there is something to be learned, the first stage of context creation is resolved as the elements to be learned are defined; in the second stage we have a working context for understanding the new material, which is now informative --- that is, input now serves to reduce uncertainty within the working context. In the third stage we have adapted completely, and lose conscious access to the learned material. Consciousness is primarily involved in the first two stages, but in the first, the input is so ill-defined that we are not even truly conscious of it as such. Consciousness of the input as such is confined to the second stage, which we call the stage of informativeness.          

                    Below we present a number of empirical findings  that support these points.   

                   Context-creation.                         

It is in fact a coherent picture of something; but in order to experience it we need some context. Some may be provided by the picture's title, "Dalmatian in the Park"; some percentage of obervers will find this hint helpful. (Note that if this helps, you may not be conscious of the title as such at the moment it seems to help --- i.e., the effect is contextual, since influences that are unconscious at the moment in  question help shape the conscious experience.) Other observers find that it helps to notice that the diagonal black "lines" converging toward the center are the edges of a sidewalk in a park. Knowing this may help to reconstruct  the three-dimensional properties of the picture. But once again, depth information created by the converging sidewalks will be unconscious at the moment when it constrains and helps reveal the conscious object. Other conscious hints that help to create context include the dog's nose, the tree above, the circular planter in which the tree stands, and the realization that a black-and-white photograph of a spotted Dalmatian in a shadow-flecked park can indeed look like this.                

                    A good deal of learning begins in confusion. Until the confusion is dispelled, until guidance is provided either by the material itself, by a parent, guide or teacher, or by the use of knowledge and strategies from the past, we do not fully experience the material that is to be learned. This point is not limited to perception. The is point that linguistic topics are  often ambiguous. Providing a title ("washing clothes") creates enough context for the paragraph, so that we can become conscious of its meaning as a whole, rather than as a fragmented and incoherent set of words and sentences.  Often we have only a goal context in confronting new material. Someone tells us that here is something interesting or important --- pay attention to it and you will become aware of it. This is how the reader was guided in the Dalmatian demonstration above. It is how psychologists usually get people to do things in experiments. We tell subjects which goals to pursue, and see how they do: "Please pay attention to this, try to memorize that, tell us what you see there." Even just providing a goal context helps narrow down the possibilities.               

                    Of course, the context of the experience itself is evoked by conscious events. Context-creation may involve tentative conscious interpretations of the input, or conscious fragments, or consciousness of a different level of analysis than the desired experience. In the Dalmatian demonstration, the reader was surely conscious of black and white spots even at the beginning. But in the stage of context-creation one is not conscious of the material to be learned as such. The Dalmatian becomes conscious only after context has been created for it.

                Conscious experience corresponds to the information  stage of adaptation. 

Harking back to the Dalmatian as a case in point, once we have the appropriate contextual constraints for a figure, we can evidently become conscious of it. But what does that mean?  Inherent in the notion of conscious experience is the existence of features of the conscious event. The Dalmatian has size, location in space, color, texture, and so on. But features involve discriminations of some sort: at the very least judgments of presence or absence, and implicit comparisons to other features. The Psychophysical Law implies that perception of intensity always involves an implicit comparison to previous intensities , and this point is not limited only to intensity. Many aspects of a perceptual event, such as color, brightness, and pattern, are known to involve implicit comparisons. Judgments of conceptual qualities are also thought to involve implicit comparisons. Research on person perception clearly shows that we perceive people in comparison to others. In sum, conscious experiences can often be shown to involve not one, but at least two alternative representations, one of which is implicit. Whenever we become conscious of something, internal processes are saying, in effect, "Aha! It's a dog with black and white spots, not a brown dog, or  cat, or some other object." Needless to say, this kind of implicit comparison must take place at many levels of analysis at the same time.               

                    Another way of saying this is that conscious events are object-like: they have many correlated features, which are implicitly differentiated from alternatives. This is quite different from habituated representations, which are not experienced as object-like. Since habituated representations are highly predictable, we may presume that their alternatives are represented, if at all, as highly unlikely. In a later chapter, the object-like nature of conscious experience will become very important .               

                    A direct connection between conscious events and quantitative measures of information has been established in several cases. Take for example the notion of "basic level categories." In a number of studies Rosch and her colleagues have shown that people tend to think of the world in terms of prototypical members of categories. Thus we think of "furniture" not so much as an abstract class, but in terms of its typical members like chairs, tables, and beds; other members like ashtrays, drapes, and vases are much less likely to be thought of spontaneously (Rosch, 1975). Further, objects fit into a hierarchy of abstraction, with the most typical objects occupying a middle level in the hierarchy between great specificity and great generality. Thus a Colonial kitchen chair‹ is quite specific; kitchen chair is more general, followed by chair, furniture, artifact, inanimate object‹ and physical object. The word chair has been thought to be the most typical member of the hierarchy: it is easier to describe, recognize, draw, recall, and the like . Typical objects are probably easiest to bring to consciousness. However, there is now evidence that the level of an object hierarchy that is easiest to use at any given time depends upon the alternatives that are being entertained. If we are considering the esthetic pros and cons of man-made objects vs. natural scenery, "chairs" are not necessarily the best example of man-made objects. Similarly, if we are debating the merits of Colonial vs. modern kitchen chairs, very specific differentiating features are likely to come to mind. Depending on our purposes and implicit comparisons, different levels of the hierarchy of abstraction are likely to come to mind. Along these lines, Barsalou and Murphy & Medin have shown that conceptual structures are highly unstable and vary with the context of alternatives. Gluck and Corter have developed a set of quantitative predictions based on this reasoning. They have modified the well-known mathematical formula for information to measure Œrelative informativeness‹, and find that the resulting model accurately  predicts which level of a category hierarchy will be chosen under different circumstances.               

                    There are other connections between likely conscious contents, implicit comparisons,  and mathematical information theory. Garner has shown that people prefer to select stimuli that are neither very high nor very low in information content, suggesting that we tend to pay attention to events that do not require enormous adaptation, but that do require some. There is also good evidence that choice reaction time, the time needed to choose between explicit alternatives, is a function of the mathematical bit-rate, the quantitative measure of information. Even classical conditioning is a function of the amount of information carried by the signalling stimulus. In communication, we select the most informative features to convey to the listener ccc that is, the features with the most valuable comparison. We have already mentioned Olson's revealing experiment in children's speech, in which the context of alternatives determines whether the child will say, "The gold star is under the square block or "The gold star is under the white block" --- even though the block in question is the same one. But we can even find this reasoning going back to Aristotle, who defined a "definition" as a statement of a general category plus a differentiating feature in the category. "A mammal is an animal that suckles its young." "A chair is a piece of furniture for people to sit on." The general category provides a context of alternatives, and the differentiating feature reduces those alternatives to the correct subset.               

                    This pattern is fundamental in linguistics. Conversations depend upon the "given-new" contract, the shared understanding that certain things are true, so that they do not need to be repeated, allowing new information to be brought to the fore. Similarly, individual sentences have a "focus" and a "presupposition": In the sentence, "It was the book that I lost yesterday", "the book" is brought forward in the sentence and made more focal. For comparison,  "Yesterday I lost the book" does not have this special focus. There are numerous linguistic techniques for bringing some information to the fore, and backgrounding other messages that are already shared. When hearing a sentence, people seem to pay attention primarily to whatever is new, focal, topical, and emphasized. There is implicit deTemphasis of anything that  is known, peripheral, and irrelevant in the moment.              

                    If conscious events indeed exist in a context of implicit alternatives, how do we know that these alternatives are being reduced? After all, information is defined as a reduction of alternatives. Part of the answer depends, of course, on the Redundancy Effects discussed below. But consider one of the most obvious aspects of conscious events: the fact that they are so fleeting. It is very difficult to keep a conscious image in mind more than a moment. With rehearsal, we can refresh inner speech to an extent, but even rehearsed words tend to fade and satiate rather quickly. Perceptual events can be renewed in various ways, but we tend to stay with them only as long as needed to achieve some specific goal, and no longer. Conscious sensory memory, such as iconic or acoustical storage, lasts only a few hundred milliseconds. One explanation of the fleetingness  of conscious events is that adaptive processes are continually learning from conscious events, reducing Œtheir alternatives, and nibbling away at the information provided. If that is true, then most conscious events lose most of their information value very quickly, as uncertainty is reduced by adaptive mechanisms. The evanescence of conscious contents is consistent with the notion of informativeness as a demand for adaptation.               

                    Finally, notice that habituated or automatic events can become conscious again when their predictions are violated. If they become unconscious due to redundancy, one way to make them conscious again is to make them informative. And indeed, that is what happens when the foot pedal of a bicycle falls off, and we become conscious again of aspects of bicycle riding that were largely automatic before. Steering the bicycle, balancing it and the like, now become informative again, as they become conscious. Again, consciousness is correlated with information content. 

---

Be first to comment this article

Only registered users can write comments.
Please login or register.

Powered by AkoComment Tweaked Special Edition v.1.4.6
AkoComment © Copyright 2004 by Arthur Konze - www.mamboportal.com
All right reserved