Center for the Study of Intelligence
Central Intelligence Agency
Differences between stronger and weaker analytical performance are attributable in large measure to differences in the organization of data and experience in analysts' long-term memory. The contents of memory form a continuous input into the analytical process, and anything that influences what information is remembered or retrieved from memory also influences the outcome of analysis.
This chapter discusses the capabilities and limitations of several components of the memory system. Sensory information storage and short-term memory are beset by severe limitations of capacity, while long-term memory, for all practical purposes, has a virtually infinite capacity. With long-term memory, the problems concern getting information into it and retrieving information once it is there, not physical limits on the amount of information that may be stored. Understanding how memory works provides insight into several analytical strengths and weaknesses.
What is commonly called memory is not a single, simple function. It is an extraordinarily complex system of diverse components and processes. There are at least three, and very likely more, distinct memory processes. The most important from the standpoint of this discussion and best documented by scientific research are sensory information storage (SIS), short-term memory (STM), and long-term memory (LTM).29 Each differs with respect to function, the form of information held, the length of time information is retained, and the amount of information-handling capacity. Memory researchers also posit the existence of an interpretive mechanism and an overall memory monitor or control mechanism that guides interaction among various elements of the memory system.
Like SIS, short-term memory holds information temporarily, pending further processing. This processing includes judgments concerning meaning, relevance, and significance, as well as the mental actions necessary to integrate selected portions of the information into long-term memory. When a person forgets immediately the name of someone to whom he or she has just been introduced, it is because the name was not transferred from short-term to long-term memory.
A central characteristic of STM is the severe limitation on its capacity. A person who is asked to listen to and repeat a series of 10 or 20 names or numbers normally retains only five or six items. Commonly it is the last five or six. If one focuses instead on the first items, STM becomes saturated by this effort, and the person cannot concentrate on and recall the last items. People make a choice where to focus their attention. They can concentrate on remembering or interpreting or taking notes on information received moments ago, or pay attention to information currently being received. Limitations on the capacity of short-term memory often preclude doing both.
Retrieval of information from STM is direct and immediate because the information has never left the conscious mind. Information can be maintained in STM indefinitely by a process of "rehearsal"--repeating it over and over again. But while rehearsing some items to retain them in STM, people cannot simultaneously add new items. The severe limitation on the amount of information retainable in STM at any one time is physiological, and there is no way to overcome it. This is an important point that will be discussed below in connection with working memory and the utility of external memory aids.
Loss of detail as sensory stimuli are interpreted and passed from SIS into STM and then into LTM is the basis for the phenomenon of selective perception discussed in the previous chapter. It imposes limits on subsequent stages of analysis, inasmuch as the lost data can never be retrieved. People can never take their mind back to what was actually there in sensory information storage or short-term memory. They can only retrieve their interpretation of what they thought was there as stored in LTM.
There are no practical limits to the amount of information that may be stored in LTM. The limitations of LTM are the difficulty of processing information into it and retrieving information from it. These subjects are discussed below.
The three memory processes comprise the storehouse of information or database that we call memory, but the total memory system must include other features as well. Some mental process must determine what information is passed from SIS into STM and from STM into LTM; decide how to search the LTM data base and judge whether further memory search is likely to be productive; assess the relevance of retrieved information; and evaluate potentially contradictory data.
To explain the operation of the total memory system, psychologists posit the existence of an interpretive mechanism that operates on the data base and a monitor or central control mechanism that guides and oversees the operation of the whole system. Little is known of these mechanisms and how they relate to other mental processes.
Despite much research on memory, little agreement exists on many critical points. What is presented here is probably the lowest common denominator on which most researchers would agree.
Organization of Information in Long-Term Memory. Physically, the brain consists of roughly 10 billion neurons, each analogous to a computer chip capable of storing information. Each neuron has octopus-like arms called axons and dendrites. Electrical impulses flow through these arms and are ferried by neurotransmitting chemicals across what is called the synaptic gap between neurons. Memories are stored as patterns of connections between neurons. When two neurons are activated, the connections or "synapses" between them are strengthened.
As you read this chapter, the experience actually causes physical changes in your brain. "In a matter of seconds, new circuits are formed that can change forever the way you think about the world."30
Memory records a lifetime of experience and thoughts. Such a massive data retrieval mechanism, like a library or computer system, must have an organizational structure; otherwise information that enters the system could never be retrieved. Imagine the Library of Congress if there were no indexing system.
There has been considerable research on how information is organized and represented in memory, but the findings remain speculative. Current research focuses on which sections of the brain process various types of information. This is determined by testing patients who have suffered brain damage from strokes and trauma or by using functional magnetic resonance imaging (fMRI) that "lights up" the active portion of the brain as a person speaks, reads, writes, or listens.
None of the current theories seems to encompass the full range or complexity of memory processes, which include memory for sights and sounds, for feelings, and for belief systems that integrate information on a large number of concepts. However useful the research has been for other purposes, analysts' needs are best served by a very simple image of the structure of memory.
Imagine memory as a massive, multidimensional spider web. This image captures what is, for the purposes of this book, perhaps the most important property of information stored in memory--its interconnectedness. One thought leads to another. It is possible to start at any one point in memory and follow a perhaps labyrinthine path to reach any other point. Information is retrieved by tracing through the network of interconnections to the place where it is stored.
Retrievability is influenced by the number of locations in which information is stored and the number and strength of pathways from this information to other concepts that might be activated by incoming information. The more frequently a path is followed, the stronger that path becomes and the more readily available the information located along that path. If one has not thought of a subject for some time, it may be difficult to recall details. After thinking our way back into the appropriate context and finding the general location in our memory, the interconnections become more readily available. We begin to remember names, places, and events that had seemed to be forgotten.
Once people have started thinking about a problem one way, the same mental circuits or pathways get activated and strengthened each time they think about it. This facilitates the retrieval of information. These same pathways, however, also become the mental ruts that make it difficult to reorganize the information mentally so as to see it from a different perspective. That explains why, in the previous chapter, once you saw the picture of the old woman it was difficult to see the young woman, or vice versa. A subsequent chapter will consider ways of breaking out of mental ruts.
One useful concept of memory organization is what some cognitive psychologists call a "schema." A schema is any pattern of relationships among data stored in memory. It is any set of nodes and links between them in the spider web of memory that hang together so strongly that they can be retrieved and used more or less as a single unit.
For example, a person may have a schema for a bar that when activated immediately makes available in memory knowledge of the properties of a bar and what distinguishes a bar, say, from a tavern. It brings back memories of specific bars that may in turn stimulate memories of thirst, guilt, or other feelings or circumstances. People also have schemata (plural for schema) for abstract concepts such as a socialist economic system and what distinguishes it from a capitalist or communist system. Schemata for phenomena such as success or failure in making an accurate intelligence estimate will include links to those elements of memory that explain typical causes and implications of success or failure. There must also be schemata for processes that link memories of the various steps involved in long division, regression analysis, or simply making inferences from evidence and writing an intelligence report.
Any given point in memory may be connected to many different overlapping schemata. This system is highly complex and not well understood.
This conception of a schema is so general that it begs many important questions of interest to memory researchers, but it is the best that can be done given the current state of knowledge. It serves the purpose of emphasizing that memory does have structure. It also shows that how knowledge is connected in memory is critically important in determining what information is retrieved in response to any stimulus and how that information is used in reasoning.
Concepts and schemata stored in memory exercise a powerful influence on the formation of perceptions from sensory data. Recall the experiment discussed in the previous chapter in which test subjects were exposed very briefly to playing cards that had been doctored so that some hearts were black and spades red. When retained in SIS for a fraction of a second, the spades were indeed red. In the course of interpreting the sensory impression and transferring it to STM, however, the spades became black because the memory system has no readily available schema for a red spade to be matched against the sensory impression. If information does not fit into what people know, or think they know, they have great difficulty processing it.
The content of schemata in memory is a principal factor distinguishing stronger from weaker analytical ability. This is aptly illustrated by an experiment with chess players. When chess grandmasters and masters and ordinary chess players were given five to 10 seconds to note the position of 20 to 25 chess pieces placed randomly on a chess board, the masters and ordinary players were alike in being able to remember the places of only about six pieces. If the positions of the pieces were taken from an actual game (unknown to the test subjects), however, the grandmasters and masters were usually able to reproduce almost all the positions without error, while the ordinary players were still able to place correctly only a half-dozen pieces.31
That the unique ability of the chess masters did not result from a pure feat of memory is indicated by the masters' inability to perform better than ordinary players in remembering randomly placed positions. Their exceptional performance in remembering positions from actual games stems from their ability to immediately perceive patterns that enable them to process many bits of information together as a single chunk or schema. The chess master has available in long-term memory many schemata that connect individual positions together in coherent patterns. When the position of chess pieces on the board corresponds to a recognized schema, it is very easy for the master to remember not only the positions of the pieces, but the outcomes of previous games in which the pieces were in these positions. Similarly, the unique abilities of the master analyst are attributable to the schemata in long-term memory that enable the analyst to perceive patterns in data that pass undetected by the average observer.
Getting Information Into and Out of Long-Term Memory. It used to be that how well a person learned something was thought to depend upon how long it was kept in short-term memory or the number of times they repeated it to themselves. Research evidence now suggests that neither of these factors plays the critical role. Continuous repetition does not necessarily guarantee that something will be remembered. The key factor in transferring information from short-term to long-term memory is the development of associations between the new information and schemata already available in memory. This, in turn, depends upon two variables: the extent to which the information to be learned relates to an already existing schema, and the level of processing given to the new information.
Take one minute to try to memorize the following items from a shopping list: bread, eggs, butter, salami, corn, lettuce, soap, jelly, chicken, and coffee. Chances are, you will try to burn the words into your mind by repeating them over and over. Such repetition, or maintenance rehearsal, is effective for maintaining the information in STM, but is an inefficient and often ineffective means of transferring it to LTM. The list is difficult to memorize because it does not correspond with any schema already in memory.
The words are familiar, but you do not have available in memory a schema that connects the words in this particular group to each other. If the list were changed to juice, cereal, milk, sugar, bacon, eggs, toast, butter, jelly, and coffee, the task would be much easier because the data would then correspond with an existing schema--items commonly eaten for breakfast. Such a list can be assimilated to your existing store of knowledge with little difficulty, just as the chess master rapidly assimilates the positions of many chessmen.
Depth of processing is the second important variable in determining how well information is retained. Depth of processing refers to the amount of effort and cognitive capacity employed to process information, and the number and strength of associations that are thereby forged between the data to be learned and knowledge already in memory. In experiments to test how well people remember a list of words, test subjects might be asked to perform different tasks that reflect different levels of processing. The following illustrative tasks are listed in order of the depth of mental processing required: say how many letters there are in each word on the list, give a word that rhymes with each word, make a mental image of each word, make up a story that incorporates each word.
It turns out that the greater the depth of processing, the greater the ability to recall words on a list. This result holds true regardless of whether the test subjects are informed in advance that the purpose of the experiment is to test them on their memory. Advising test subjects to expect a test makes almost no difference in their performance, presumably because it only leads them to rehearse the information in short-term memory, which is ineffective as compared with other forms of processing.
There are three ways in which information may be learned or committed to memory: by rote, assimilation, or use of a mnemonic device. Each of these procedures is discussed below.32
By Rote. Material to be learned is repeated verbally with sufficient frequency that it can later be repeated from memory without use of any memory aids. When information is learned by rote, it forms a separate schema not closely interwoven with previously held knowledge. That is, the mental processing adds little by way of elaboration to the new information, and the new information adds little to the elaboration of existing schemata. Learning by rote is a brute force technique. It seems to be the least efficient way of remembering.
By Assimilation. Information is learned by assimilation when the structure or substance of the information fits into some memory schema already possessed by the learner. The new information is assimilated to or linked to the existing schema and can be retrieved readily by first accessing the existing schema and then reconstructing the new information. Assimilation involves learning by comprehension and is, therefore, a desirable method, but it can only be used to learn information that is somehow related to our previous experience.
By Using A Mnemonic Device. A mnemonic device is any means of organizing or encoding information for the purpose of making it easier to remember. A high school student cramming for a geography test might use the acronym "HOMES" as a device for remembering the first letter of each of the Great Lakes--Huron, Ontario, etc.
To learn the first grocery list of disconnected words, you would create some structure for linking the words to each other and/or to information already in LTM. You might imagine yourself shopping or putting the items away and mentally picture where they are located on the shelves at the market or in the kitchen. Or you might imagine a story concerning one or more meals that include all these items. Any form of processing information in this manner is a more effective aid to retention than rote repetition. Even more effective systems for quickly memorizing lists of names or words have been devised by various memory experts, but these require some study and practice in their use.
Mnemonic devices are useful for remembering information that does not fit any appropriate conceptual structure or schema already in memory. They work by providing a simple, artificial structure to which the information to be learned is then linked. The mnemonic device supplies the mental "file categories" that ensure retrievability of information. To remember, first recall the mnemonic device, then access the desired information.
An analyst's memory provides continuous input into the analytical process. This input is of two types--additional factual information on historical background and context, and schemata the analyst uses to determine the meaning of newly acquired information. Information from memory may force itself on the analyst's awareness without any deliberate effort by the analyst to remember; or, recall of the information may require considerable time and strain. In either case, anything that influences what information is remembered or retrieved from memory also influences intelligence analysis.
Judgment is the joint product of the available information and what the analyst brings to the analysis of this information. An experiment documenting differences between chess masters and ordinary chess players was noted earlier. Similar research with medical doctors diagnosing illness indicates that differences between stronger and weaker performers are to be found in the organization of information and experience in long-term memory.33 The same presumably holds true for intelligence analysts. Substantive knowledge and analytical experience determine the store of memories and schemata the analyst draws upon to generate and evaluate hypotheses. The key is not a simple ability to recall facts, but the ability to recall patterns that relate facts to each other and to broader concepts--and to employ procedures that facilitate this process.
A well-known article written over 40 years ago, titled "The Magic Number Seven--Plus or Minus Two," contends that seven--plus or minus two--is the number of things people can keep in their head all at once. 34 That limitation on working memory is the source of many problems. People have difficulty grasping a problem in all its complexity. This is why we sometimes have trouble making up our minds. For example, we think first about the arguments in favor, and then about the arguments against, and we can't keep all those pros and cons in our head at the same time to get an overview of how they balance off against each other.
The recommended technique for coping with this limitation of working memory is called externalizing the problem--getting it out of one's head and down on paper in some simplified form that shows the main elements of the problem and how they relate to each other. Chapter 7, "Structuring Analytical Problems," discusses ways of doing this. They all involve breaking down a problem into its component parts and then preparing a simple "model" that shows how the parts relate to the whole. When working on a small part of the problem, the model keeps one from losing sight of the whole.
A simple model of an analytical problem facilitates the assimilation of new information into long-term memory; it provides a structure to which bits and pieces of information can be related. The model defines the categories for filing information in memory and retrieving it on demand. In other words, it serves as a mnemonic device that provides the hooks on which to hang information so that it can be found when needed.
The model is initially an artificial construct, like the previously noted acronym "HOMES." With usage, however, it rapidly becomes an integral part of one's conceptual structure--the set of schemata used in processing information. At this point, remembering new information occurs by assimilation rather than by mnemonics. This enhances the ability to recall and make inferences from a larger volume of information in a greater variety of ways than would otherwise be possible.
"Hardening of the Categories". Memory processes tend to work with generalized categories. If people do not have an appropriate category for something, they are unlikely to perceive it, store it in memory, or be able to retrieve it from memory later. If categories are drawn incorrectly, people are likely to perceive and remember things inaccurately. When information about phenomena that are different in important respects nonetheless gets stored in memory under a single concept, errors of analysis may result. For example, many observers of international affairs had the impression that Communism was a monolithic movement, that it was the same everywhere and controlled from Moscow. All Communist countries were grouped together in a single, undifferentiated category called "international Communism" or "the Communist bloc." In 1948, this led many in the United States to downplay the importance of the Stalin-Tito split. According to one authority, it "may help explain why many Western minds, including scholars, remained relatively blind to the existence and significance of Sino-Soviet differences long after they had been made manifest in the realm of ideological formulae."35
"Hardening of the categories" is a common analytical weakness. Fine distinctions among categories and tolerance for ambiguity contribute to more effective analysis.
Things That Influence What Is Remembered. Factors that influence how information is stored in memory and that affect future retrievability include: being the first-stored information on a given topic, the amount of attention focused on the information, the credibility of the information, and the importance attributed to the information at the moment of storage. By influencing the content of memory, all of these factors also influence the outcome of intelligence analysis.
Chapter 12 on "Biases in Estimating Probabilities" describes how availability in memory influences judgments of probability. The more instances a person can recall of a phenomenon, the more probable that phenomenon seems to be. This is true even though ability to recall past examples is influenced by vividness of the information, how recently something occurred, its impact upon one's personal welfare, and many other factors unrelated to the actual probability of the phenomenon.
Memory Rarely Changes Retroactively. Analysts often receive new information that should, logically, cause them to reevaluate the credibility or significance of previous information. Ideally, the earlier information should then become either more salient and readily available in memory, or less so. But it does not work that way. Unfortunately, memories are seldom reassessed or reorganized retroactively in response to new information. For example, information that is dismissed as unimportant or irrelevant because it did not fit an analyst's expectations does not become more memorable even if the analyst changes his or her thinking to the point where the same information, received today, would be recognized as very significant.
Training courses for intelligence analysts sometimes focus on trying to open up an analyst's established mind-set, to get him or her to see problems from different perspectives in order to give a fairer shake to alternative explanations. More often than not, the reaction of experienced analysts is that they have devoted 20 years to developing their present mind-set, that it has served them well, and that they see no need to change it. Such analysts view themselves, often accurately, as comparable to the chess masters. They believe the information embedded in their long-term memory permits them to perceive patterns and make inferences that are beyond the reach of other observers. In one sense, they are quite correct in not wanting to change; it is, indeed, their existing schemata or mind-set that enables them to achieve whatever success they enjoy in making analytical judgments.
There is, however, a crucial difference between the chess master and the master intelligence analyst. Although the chess master faces a different opponent in each match, the environment in which each contest takes place remains stable and unchanging: the permissible moves of the diverse pieces are rigidly determined, and the rules cannot be changed without the master's knowledge. Once the chess master develops an accurate schema, there is no need to change it. The intelligence analyst, however, must cope with a rapidly changing world. Many countries that previously were US adversaries are now our formal or de facto allies. The American and Russian governments and societies are not the same today as they were 20 or even 10 or five years ago. Schemata that were valid yesterday may no longer be functional tomorrow.
Learning new schemata often requires the unlearning of existing ones, and this is exceedingly difficult. It is always easier to learn a new habit than to unlearn an old one. Schemata in long-term memory that are so essential to effective analysis are also the principal source of inertia in recognizing and adapting to a changing environment. Chapter 6, "Keeping an Open Mind," identifies tools for dealing with this problem.
29Memory researchers do not employ uniform terminology. Sensory information storage is also known as sensory register, sensory store, and eidetic and echoic memory. Short- and long-term memory are also referred to as primary and secondary memory. A variety of other terms are in use as well. I have adopted the terminology used by Peter H. Lindsay and Donald A. Norman in their text on Human Information Processing (New York: Academic Press, 1977). This entire chapter draws heavily from Chapters 8 through 11 of the Lindsay and Norman book.
30George Johnson, In the Palaces of Memory: How We Build the Worlds Inside Our Heads. Vintage Books, 1992, p. xi.
31A. D. deGroot, Thought and Choice in Chess (The Hague: Mouton, 1965) cited by Herbert A. Simon, "How Big Is a Chunk?" Science, Vol. 183 (1974), p. 487.
32This discussion draws on Francis S. Bellezza, "Mnemonic Devices: Classification, Characteristics, and Criteria" (Athens, Ohio: Ohio University, pre-publication manuscript, January 1980).
33Arthur S. Elstein, Lee S. Shulman & Sarah A. Sprafka, Medical Problem Solving: An Analysis of Clinical Reasoning (Cambridge, MA: Harvard University Press, 1978), p. 276.
34George A. Miller, "The Magical Number Seven--Plus or Minus Two: Some Limits on our Capacity for Processing Information." The Psychological Review, Vol. 63, No. 2 (March 1956).
35Robert Tucker, "Communist Revolutions, National Cultures, and the Divided Nations," Studies in Comparative Communism (Autumn 1974), 235-245.