By looking extensively at the literature on human cognition and deception, a model was formed of human cognition with specific focus on its application to deception. This includes Lambert's data collection and mapping into his model of human deception.

Lambert's Cognitive Model

We begin with Lambert's model of human cognition. [2] This model is linked to the history of psychological models of brain function and cognition and, as such, does not represent so much the physiology of the brain as the things it is generally believed to do and the manner in which it is generally believed to operate. There is no sense that this model will be found to match physiology in the long run, however, it is useful because it relates to a great deal of other experimental work that has been done on deception and the limits of human perception. It may also be related to perceptual control theory's notions of orders of control and, through that mechanistic view, to physiology. [31]

Lambert's Model of Cognition

This model identifies integers as labels for major brain functions. Within this model, Lambert has created href=Lambert.html>a structure of sub processes identified with behavior in general and deception in particular. This structure is broken down into subsections as follows. In addition to the structural association, Lambert created a detailed mapping of how cognitive function was thought to work. The structure can be interpreted as a stimulus response network but there is an isomorphism to a model-referenced adaptive control system. The components consist of (1) the global executive, (2) a controller with limited processing resources and buffer memory, (3) shot-term memory and working memory which includes visual acoustic, motor, and coded memories, (4) the local manager which does problem solving, learning, and procedures, (5) buffer memories for both input and output, (6) sensors, which include transducers for the senses, (7) affecters, which includes transducers for all outputs, and (8) long-term memory, which includes internal images of the world (knowledge, belief, and situation) and language (sensor data and affector data).

The model provides for specific interconnections between components that appear to occur in humans. Specifically, long term memory is affected only by short term memory but affects short term memory and buffer memories for sensors and affectors. The executive sends information to the local manager and acts in a controlling function over short term memory and the controller. The short term memory interacts with the long-term memory, receives information from sensor buffers, and interacts with the local manager. The local manager receives information from the global executive and interacts with the short term memory. The sensor observes reflections of the world and sends the resulting signals through incoming buffer memory to short and long term memory. Long term memory feeds information to output buffers that then pass the information on to affectors.

This depiction is reflected in a different structure which models the system processes of cognition.

Lambert's Model of Cognition

In this depiction, we see the movement of information from senses through a cognitive process that includes reflexes, conditioned behavior, intuition, and reasoning, and a movement back down to action. Many more details are provided, but this is the general structure of cognition with which Lambert worked. From a standpoint of understanding deception, the notion is that the reflections of the world that reach the senses of the cognition system are interpreted based on its present state. The deception objective is to control those reflections so as to produce the desired changes in the perception of the target so as to achieve compliance. This can be done by inhibiting or inducing cognitive activities within this structure.

The induction of signals at the sense level is relatively obvious, and the resulting reflexive responses are quite predictable in most cases. the problems start becoming considerable as higher levels of the victim's cognitive structure get involved. while the mechanism of deception may involve the perception of feature, any feedback from this can only be seen as a result of conditioned behaviors at the perceive form level or higher level cognitive affects reflected in the ultimate drives of the system. For this reason, while the model may be helpful in understanding internal states, affects at the perceive feature level are aliased as affects at higher levels. Following the earlier depiction of deceptions as consisting of inhibitions and inducements of sensor data we can think of internal effects of deception on cognition in terms of combinations of inhibitions and inducements of internal signals. The objective of a deception might then, for example, be the inhibition of sensed content from being perceived as a feature, perhaps accomplished by a combination of reducing the available signal and distracting focus of attention by inducing the perception of a different form and causing a simultaneous reflexive action to reduce the available signal. This is precisely what is done in the case of the disappearing elephant magic trick. The disappearing elephant trick is an excellent example of the exploitation of the cognitive system and can be readily explained through Lambert's model.

Example: This trick is set up by the creation of a rippling black silk curtain behind the elephant, which is gray. The audience is in a fairly close pack staring right at the elephant some distance away. Just before the elephant disappears, a scantily clad woman walks across the front of the crowd and the magician is describing something that is not very interesting with regard to the trick. Then, as eyes turn toward the side the girl is walking toward, a loud crash sound is created to that side of the crowd. They crowd's reflexive response to a crashing sound it to turn toward the sound, which they do. This takes about 1/3 to 1/2 second. As soon as they are looking that way, the magician causes another black silk rippling curtain to rise up in front of the elephant. This takes less than 1/4 second. Because of the low contrast between the elephant and the curtain and the rippling effect of the black back and front curtains, there is no edge line induced in the audience and thus attention is not pulled toward the curtains. By the time the crowd looks back, the elephant is gone and is then moved away while out of sight. The back curtain is lowered, and the front curtain is then raised to prove that only the wall remains behind the curtain.

For low-level one-step deceptions such as this one, Lambert's model is an excellent tool both for explanation and for planning. There are a set of known sensors, reflexes, and even well known or trainable conditioned responses that can be exploited almost at will. In some cases it will be necessary to force the cognitive system into a state where these prevail over higher level controlling processes, such as a member of the crowd who is focusing very carefully on what is going on. This can be done by boring them into relaxation, which the magician tries to do with his boring commentary and the more interesting scantily clad woman, but otherwise it is pretty straight forward. Unfortunately, this model provides inadequate structure for dealing with higher level or longer term cognitive deceptions. For these you need to move to another sort of model that, while still consistent with this model, provides added clarity regarding possible moves.

A Cognitive Model for Higher Level Deceptions

The depiction below attempt to provide additional structure for higher level cognitive deceptions. This model starts to look at how humans interact to create deceptions and how those deceptions can, at a broad level, cause interpretation and behavior in the target that is compliant with the deceiver. It also shows the recursive nature of deception because of the regress induced by both time and symmetry.

Model of Human Cognition for Deceptions

The depiction shows interaction between two human or group cognitive systems. The interaction all takes place through the world using human senses (small, taste, hearing, touching, seeing, pheromones, and allergic reactions). Deception is modeled by the induction or suppression of target observables by the deceiver.

Cognitive processes responding directly to inputs include sensory data which, after sensor bias and the filter of a set of observables, becomes observable. Sensory data, after bias, can trigger reflexive responses which also induce observable internal changes. Other actions can also be generated and expectations actively control everything in this list. Focus of attention can also be affected at this level because of detection mechanisms and their triggering of higher level processes. This paragraph summarizes what we will tentatively call the 'low level' cognitive system.

Cognitive processes in, what we tentatively call, the middle level of cognition include conditioned and other automatic but non reflexive responses, measurement mechanisms and automatic or trained evaluation and decision methods, learned and nearly automated capabilities including skills, tools, and methods that are based on pattern matching, training, instinctual responses, the actions they trigger, and the feedback mechanisms involved in controlling those actions. This level also involves learned patterns of focus of attention.

The remaining cognitive processes are called high level. This includes reason-based assessments and capabilities, expectations, which include biases, fidelity of interest, level of effort, consistency with observables, and high-level focus of attention, and intent, which includes objectives, qualitative evaluation, schedule and budgetary requirements. The link between expectations and the rest of the cognitive structure is particularly important because expectations alter focus of attention sequences, cognitive biases, assessment, intent, and the evaluation of expectations, while changing of expectation can keep them stable, move them at a limited rate, or cause dissonance.

Deceptions of Low-level Cognition

In this model, we have collapsed the lower levels (up to conditioned response) of Lambert's model into the bottom two boxes (Observables and Actions) and created a somewhat more specific higher level structure. Details of these deceptions are provided in the sections 6 and 7 of Lambert's data collection. Low-level visual deceptions are demonstrated by Seckel [15] and described by Hoffman [36]. Audio deceptions are demonstrated on an audio CD-ROM by Deutsch [47].

Deceptions of Mid-level Cognition

The notion is that there are pattern matching and reason-based assessments and capabilities that interact to induce more thoughtful decisions than conditioned response. While pattern matching cognition mechanisms are more thoughtful than conditioned response, they are essentially the programmed behaviors identified by Cialdini [34] and some of the negotiation tactics of Karrass [33]. These include, but are not limited to, reciprocation, authority, contrast, commitment and consistency, automaticity, social proof, liking, and scarcity, and as Karrass formulates it, credibility, message content and appeal, situation setting and rewards, and media choice are all methods.

The potential for decisions to be moved to more logical reasoning exists, but this is limited by the effects identified by Gilovich [14]. Specifically, the notions that people (erroneously) believe that effects should resemble their causes, they misperceive random events, they misinterpret incomplete or unrepresentative data, they form biased evaluations of ambiguous and inconsistent data, they have motivational determinants of belief, they bias second hand information, and they have exaggerated impressions of social support. More content is provided in the sections numbered 1, 2, and some portions of 4 and 8 of Lambert's data collection.

Deceptions of High-level Cognition

Karrass [33] also provides techniques for affecting influence in high-level thoughtful situations. He explains that change comes from learning and acceptance. Learning comes from hearing and understanding, while acceptance comes from comfort with the message, relevance, and good feelings toward the underlying idea. These are both affected by audience motives and values, the information and language used for presentation, audience attitudes and emotions, and the audience's perception and role in the negotiation. Karrrass [33] provides a three dimensional depiction of goals, needs, and perceptions and asserts that people are predictable. He also provides a set of tactics including timing, inspection, authority, association, amount, brotherhood, and detour that can be applied in a deception context. Handy also provides a set of influence tactics that tend to be most useful at higher levels of reasoning, including physicality, resources, position (which yields information, access, and right to organize), expertise, personal charisma, and emotion. More content is also provided in the sections 4 and 8 of Lambert's data collection.

Moving from High-Level to Mid-level Cognition

Karrass also augments Cialdini's notions [34] of rush, stress, uncertainty, indifference, distraction, and fatigue leading to less thoughtful and more automatic responses and brings out Maslow's needs hierarchy (basic survival, safety, love, self worth, and self-actualization). By forcing earlier sets of these issues, reasoning can be driven away and replaced by increased automaticity. Tactics of timing can also be used to drive people toward increased automaticity. Thus we can either drive the target toward less thought or use Karrass's methods of negotiation to cause desired change.

Moving from Mid-Level to High-level Cognition

Cognition moves to higher levels only when there are intent-based forcing factors that lead to deeper analysis, (e.g., when objectives are oriented toward more in-depth thought, quality requirements drive more detailed consideration, schedule availability provides free time to do deeper consideration, or extra budget is available for this purpose) or when expectations are not met (i.e., the fidelity of the deception is inadequate, biases trigger more detailed examination, inconsistencies or errors are above some threshold, or the difference between expectations and observations is so great or changing at so great a rate as to cause dissonance). In these cases, higher levels of reasoning are applied, complete with all of their potential logical fallacies and their special skills, tools, and methods. Higher level reasoning is desired when we wish to change intent or make radical changes in expectations, while we try to drive decisions to lower cognitive levels when we can induce less thoughtful responses in our favor.

An Example

To get a sense of how the model might be applied to deceptions, we have included a sample analysis of a simple human deception. The deception is an attack with a guard at a gate as the target. It happens many times each day and is commonly called tailgating.

The target of this deception is the guard and our method will be to try to exploit a natural overload that takes place during the return from lunch hour on a Thursday. We choose the end of the lunch hour on Thursday because the guard will be as busy as they ever get and because they will be looking forward to the weekend and will probably have a somewhat reduced alertness level. Thus we are intentionally trying to keep processing at a pattern matching level by increased rush, stress, indifference, distraction, and fatigue.

We stand casually out of the guard's sight before the crowd comes along, join the crowd as it approaches the entry, hold a notepad where a badge appears on other peoples' attire, and stay away from the guard's side of the group. Our clothing and appearance is such that it avoids dissonance with the guard's expectations and does not affect the guard's intent in any obvious way.

We tag along in the third row back near someone that looks generally like us and, when the guard is checking one of the other people, we ease our way over to the other side of the guard, appearing to be in the already checked group. Here we are using automaticity and social proof against the guard and liking by similarity against the group we are tailgaiting with. We are also using similarity to avoid triggering sensory detection and indifference, distraction and fatigue to avoid triggering higher level cognition.

As the group proceeds, so do we. After getting beyond the guard's sight, we move to the back of the group and drop out as they round a corner. Here we are using automaticity, liking, and social proof against the group to go along with them, followed by moving slowly out of their notice which exploits slow movement of expectations followed by concealment from observation.

Team members have used variations on this entry technique in red teaming exercises against facilities from time to time and have been almost universally successful in its use. It is widely published and well known to be effective. It is clearly a deception because if the guard knew you were trying to get past without a badge or authorization they would not permit the entry. While the people who use it don't typically go through this analytical process at a conscious level, they do some part of it at some level and we postulate that this is why they succeed at it so frequently.

As an aside, there should always be a backup plan for such deceptions. The typical tailgaiter, if detected, will act lost and ask the guard how to get to some building or office, perhaps finding out that this is the wrong address in the process. This again exploits elements of the deception framework designed to move the guard away from high level cognition and toward automaticity that would favor letting the attacker go and not reporting the incident.

In the control system isomorphism, we can consider this same structure as attempting to maintain internal consistency and allow change only at a limited rate. The high level control system is essentially oblivious to anything unless change happens at too high a rate or deviations of high level signals from expectations are too high. Similarly, the middle levels operate using Cialdini's rules of thumb unless a disturbance at a lower level prompts obvious dissonance and low-level control decisions (e.g., remain balanced) don't get above the reflexive and conditioned response levels unless their is a control system failure.