Cognitive Neuroscience Cognitive Control

Cognitive control is executive function that allows us to “use our perceptions, knowledge, and goals to bias the selection of action and thought from a multitude of possibilities”.

Perception-action cycle The perception-action cycle is an essential aspect of cognitive control. Remember that it refers to the recurring sequence of perception and action that repeats as the organism interacts with the environment. Each action sets the stage for a new perception, which in turn enables a new action. The perception-action cycle depends on the functional linkage between posterior cortical areas for perception and frontal areas for action, each organized hierarchically.

Set and expectancy Set refers to any preparatory process carried out by the brain. It may be related to perception, action, or higher-level (preparatory) attention. The lateral prefrontal cortex is involved in tasks requiring the integration of contingencies over time. Temporal integration is necessary when the information required for behavioral performance is discontinuous in time, i.e. the items of information are available only at certain times, with gaps in between. The lateral prefrontal cortex is involved in mediating contingencies between elements of information that are separated in time. The mediation of cross-temporal contingencies is a crucial function for complex behavior that allows information that was only available at some time in the past to be used as if it were available at the present. Potentially, the availability of past information may have been distributed over many previous times, and all the information is available together at one time. This capability gives species that possess it an advantage in competition with other species that do not.

In the mediation of cross-temporal contingency, the lateral prefrontal cortex makes use of two complementary functions: working memory and preparatory set. (a) Working memory is attention directed to the internal representation of previously encountered sensory information. (b) Preparatory set is the priming (i.e. partial pre-activation) of perceptual and motor memory networks for expected events. Preparatory set may be considered attention focused on expected events and on the anticipated consequences of present events and actions. This ability to anticipate future consequences of present activity lies at the heart of the ability to formulate plans of action. Lesions of lateral prefrontal cortex disrupt this ability.

Neural correlates of set: 1. The Contingent Negative Variation (CNV) is a slow surface-negative potential that is maximal over frontal cortex during the period between S1 and S2, where the contingency of S2 on S1 has been learned. Because it is related to the expectancy of S2 brought about by the occurrence of S1, the CNV is also called an expectancy wave. The size of the CNV increases with time during the S1-S2 period. Therefore, it is larger when this period is longer. The CNV is related to task set. 2. The Lateralized Readiness Potential (LRP) is also a slow surface-negative potential like the CNV. The LRP occurs when a subject is about to make a planned motor response. It is often studied in a paradigm where the subject receives a warning stimulus (WS), then has the opportunity to prepare a motor response (RS) during the delay period, then makes the RS in response to an imperative stimulus (IS). The LRP occurs during the delay period of this paradigm. It is maximal over SMA, and is larger contralateral to the RS when the RS is unilateral. The LRP is related to motor set.

3. Neurons related to set have been found in lateral prefrontal cortex in close proximity to working memory cells. They increase their discharge as the time of an impending action approaches. Discharge accelerates as a function of the probability with which the animal can precisely predict the motor act to be performed. The relation between frontal EEG slow waves (e.g., CNV & LRP) and unit activity in set is explained by the excitability regulation theory of Rockstroh et al (1989):

a) The excitability of prefrontal cortex is increased by depolarization of the apical dendrites of pyramidal neurons – by synapses from input fibers from the nonspecific thalamic nuclei.

b) Depolarization of pyramidal cell populations in frontal cortex gives rise to: (a) increased surface negativity in the EEG; and (b) increased firing rates of the pyramidal cells.

Summary: preparatory set complements working memory in temporal integration. For temporal integration to take place, information must be transferred from neurons representing a signal in the recent past (working memory cells) to neurons that prepare the consequent response to that signal in the near future (set cells).

Executive control In the control of behavior, the selective allocation of neural resources requires executive control. The selective allocation of executive networks and effector networks is essential for purposive action. Since the prefrontal cortex is the highest and most integrative stage in the action hierarchy, it follows that it exerts the highest level of executive control. Remember that damage to lateral prefrontal cortex leads to a set of symptoms that have been characterized as the disexecutive syndrome (Baddeley 1986), the inability to formulate, initiate, and execute plans of action. This syndrome is attributed to a failure of activation of networks that store action schema, as well as those necessary for the initiation of action, and the maintenance of the course of action toward a goal. Executive control is needed to select from alternatives that exist about the course of action to follow at every moment. It arises from the interaction of executive networks and perceptual networks.

Execution and monitoring Executive control is intimately related to the actual execution of action. Both consist in the selective allocation of components of the motor system for a specific action. Systems for executive control highly overlap those for execution of action: the same cortical areas involved in executive control are also involved in controlling gaze, head position, and body postures – actions involved in orienting behavior – actions that promote executive control. Executive control systems are involved in visual spatial attention. 1) They control the oculomotor system to control the direction of gaze. 2) By controlling the direction of gaze, they restrict the content of visual perception. 3) In keeping with the perception-action cycle, the control of gaze by executive control areas is to a large degree based on perceptual analysis of the environment.

Although the frontal cortex is organized hierarchically, it does not follow that participation of frontal areas in execution of movements follows a rigid top-down sequential order. There is evidence (e.g., Kalaska et al 1998) that different frontal areas are active in parallel, controlling different movement parameters, during preparation for, and execution of, movements. Executive control is also highly distributed in space and time for planning and control of action. Automatic, well-rehearsed actions may be performed largely in parallel at lower levels of the motor hierarchy, and without executive control. Actions requiring high levels of executive control, however, require the temporal integrative function of prefrontal cortex, and consequently a greater reliance on serial processing.

The term monitoring refers to attention directed to sensory input specifically for the guidance of action. Monitoring is a “tool” of executive control, used by the prefrontal cortex to control actions, both as they occur and in preparation for prospective events. Monitoring reflects the role of prefrontal cortex in executive control:

1) The prefrontal cortex exerts both inclusionary and exclusionary control on perceptual and motor systems.

2) The prefrontal cortex receives perceptual signals that modulate its control of action.

3) Perceptual signals to prefrontal cortex may: a. result from processing of sensory inputs by perceptual cortical areas b. come from those areas without sensory inputs c. arise as a consequence of movement in the form of proprioceptive

feedback d. be generated in response to efferent copies of current motor action

received as corollary discharge from the motor system 4) The exclusionary control is needed to protect goal-directed actions from

interference caused by distracting stimuli and incompatible schema; it is thought to originate in orbital frontal cortex (OFC).

Development of high-level cognitive function First we look at the development of high-level cognitive functions. Humans have a number of high-level cognitive abilities of which all other species are incapable. These include:

1) language 2) abstract thought 3) sophisticated tool manufacture & use

High-level cognitive functions are supported by other cognitive functions, such as attention and memory, and thus depend on the development of these functions. The development of human high-level cognitive functions depends on the development of adaptive behavior. Adaptation to the world involves reasoning directed to the pursuit of goals.

Intelligence High-level functions are the hallmark of intelligence. Other species are capable of intelligent behavior. E.g. a monkey performing a delayed match to sample task is engaged in problem solving, a high-level cognitive function. However, even though humans are not the only intelligent living beings, human intelligence represents a “quantum leap” in the evolution of cognition. Intelligent high-level cognitive functions depend on the development of cortical cognitive networks, and the development of efficient information processing in them. Variation in the efficiency of cortical information processing may underlie variation in the degree of intelligence seen across individuals. Efficiency may be defined as “the ability to use available means, including prior knowledge, to attain a goal such as the solution of a problem”. Intelligence tests commonly emphasize some sort of efficiency of performance.

In early human development, the individual usually becomes progressively more efficient at information processing with increasing age. Conclusion: The development of intelligence closely correlates with cortical development, especially development of association cortex. The development of cognitive ability in the individual member of primate species in the “formative” years parallels maturation of the cortex, especially association cortex. However, there are no known correlates between cortical structure, either macroscopic or microscopic, and intelligence measures. If there is a basis for intelligence in brain structure, it is not accessible by any measure currently available. For example, the brain of Albert Einstein has been highly scrutinized with the goal of finding some structural marker that distinguishes it from the average human being. It has been reported that the brain has a higher than normal proportion of oligodendroglia, and also that the inferior parietal lobe is 15% wider than normal. In summary, the reported effects are still controversial, and Einstein’s is just one famous brain that has received a great deal of attention. In general, no significant correlation has been established between cortical structure and intelligence.

Piaget’s developmental studies Jean Piaget pioneered the study of intellectual development in the child. His studies consisted mainly of observational field work. He did not use rigorous quantification or statistical testing. Piaget’s theory of child intellect development consists of 4 distinct stages, each within a well-defined age range.

1. sensory-motor stage (birth – 2 years)

a. the child learns to integrate complex sensations and movements, extending the basic (simpler) reflexes present at birth.

b. the child begins to develop schemata of sensory-motor integration (characterized by stereotypical pantomimes).

2. representational stage (2 – 7 years)

a. the child extends the use of symbolism to the verbal domain to represent the world.

b. the child develops the ability to manipulate objects, which becomes progressively more regulated by feedback from the environment, and refined by trial and error.

c. as language comprehension develops, this feedback includes progressively more language from other people.

3. concrete operations stage (7 – 11 years) a. the child becomes less stimulus-bound, and more independent in being

able to organize behavior (including use of language) to achieve goals. b. there is a great increase in improvisation and creativity.

4. formal operations stage (11 – 15 years)

a. the child begins to use hypothetical reasoning and problem-solving abilities.

b. inductive and deductive logical skills develop. c. the capability for purposive behavior develops along with the ability to

construct temporal gestalts of logical thought and action to achieve distant goals.

d. language is well-integrated in goal-directed behavior as the formulation of propositions in the construction of goal-directed gestalts.

Piaget’s approach is valuable, & his methodology is useful, but it is “insufficient cognitively and neurobiologically”. Later work has provided necessary qualifications. Some criticisms of Piaget’s theory:

1. the stages are too rigid in their boundaries: different children transition between stages at different times.

2. it lacks the exclusionary aspect of attention in behavior: the child must develop the ability to suppress distracting sensory inputs, alternate constructs, conceptually competing categorizations, etc.

3. children can reason with numbers at an earlier age than he proposed. Some important implications of Piaget’s theory:

1. Piaget’s scenario of stages properly emphasizes the increasingly higher levels of integration (formation of high-level associations) that occur in childhood intelligence development.

2. in transitioning from one stage of development to another, cognitive networks at progressively higher levels take over integrative functions that were supported by lower-level networks in earlier stages; in this process, lower levels may be subordinated rather than suppressed.

3. the transitioning between successive stages of intelligence development may require progressively higher levels of integration, and the onset of participation of a hierarchical level at a particular stage of development may depend on the structural maturation of the areas at that level.

Example 1: the use of verbal symbolism in the representational stage may depend on maturation of association areas supporting symbolic memory networks. Example 2: the high-level cognitive functions of the formal operations stage may depend on maturation of the prefrontal cortex, which is latest cortical region to mature. Functions that depend on interactions of the prefrontal cortex depend on prefrontal maturation for creation of: 1) intricate behavioral sequences 2) logical constructs 3) elaborate sentences Sternberg classification of intelligence

1) analytical intelligence: based on reasoning 2) practical intelligence: based on problem-solving abilities acquired mostly by

ordinary life experience 3) creative intelligence: based on conceiving, imagination, intuition

Individuals vary as to their use and command of these 3 types of intelligence. Therefore, it is considered useful to have a measure that quantifies general intelligence. This goal has led to the development of intelligence tests. General intelligence tests In 1905, Binet developed a test for intelligence to be given to French school-children. Many other similar tests have been developed since then. Most rate intelligence in reference to standardized scales of a person’s mental age. Mental age is combined with chronological age to derive an Intelligence Quotient (IQ). The most common tests used in the US are:

1) Stanford-Binet 2) Wechsler-Bellvue

a. Wechsler Intelligence Scale for Children (WISC) b. Wechsler Adult Intelligence Scale (WAIS)

Most measures of intellectual ability tend to correlate with one another. This has led to attempts to derive a measure of general intelligence. The Spearman g-factor was derived for this purpose.

Raven’s Progressive Matrices (RPM) test is a test of “fluid” intelligence that has been used in many studies of analytic intelligence; it can be used to derive the Spearman g-factor. Fluid intelligence refers to the performance of tasks that require manipulation of novel information; it is typically non-verbal. Crystallized intelligence refers to tasks performed by retrieval of existing knowledge from long-term memory; it is commonly verbal. Since attention ensures the selective allocation of cognitive networks to the processing of information in new situations, as in the solution of new problems, it is highly correlated with intelligence.

Electrocortical studies of intelligence Many studies have tried to find an electrocortical correlate of intellectual performance. It has been shown that a relation exists between IQ and the EEG: 1) a relation has been reported between IQ and EEG frequencies in high alpha/low beta range. 2) it has been reported that coherence in the theta range between frontal and posterior cortices is a reliable correlate of intelligence. This finding suggests that intelligence depends on reentrant processing between these areas.

Neuroimaging studies of intelligence There is abundant evidence for prefrontal network involvement in performance of high-level cognitive function. Many tasks show activation of the anterior cingulate cortex (ACC) as well as the lateral prefrontal cortex. Duncan (1996) proposed that general intelligence, as measured by the Spearman g-factor, depends on the prefrontal cortex. The Spearman g-factor is lowered by prefrontal lesion and is elevated in normal subjects by tasks that activate prefrontal cortex. A 2000 study by Duncan recorded PET in subjects performing problem-solving tasks high in Spearman g-factor. The results show that those tasks activate lateral prefrontal cortex. The tasks may also have involved a variety of perceptual memory networks in widely distributed posterior cortical areas.

Reasoning Reasoning is logical thought, including both deductive and inductive logic. It includes mathematical as well as linguistic operations. It may be unconscious as well as conscious. From a phenomenological perspective, reasoning may be defined as the formation of new knowledge from prior knowledge. From the network perspective, reasoning is the formation of new cortical cognitive networks from existing ones. A new cortical cognitive network that is formed from existing ones may be considered to be an inference. An inference may be based on:

1) pre-existing knowledge 2) new sensory information 3) recent sensory information

2 main methodologies are used in cognitive science to explain reasoning:

1. Linguistic (SSP): predicated on rule-based symbolic formalization of knowledge. Linguistic reasoning is propositional.

2. Connectionist (PDP): predicated on fast, expert, spontaneous, parallel information processing in cognitive networks. Connectionist reasoning is nonpropositional.

The human brain uses both linguistic and connectionist forms of reasoning, and both probably rely on interaction of lexical and cognitive networks. However, the brain mechanisms of reasoning are unknown. 1. Symbolic reasoning models typically are based on an executive processing unit (agent). The agent plans the successive rule-based processing stages of words and propositions. The representations in these models are generally intelligible at every stage. However, these models are often neurobiologically implausible:

a) they lack biophysical detail. b) they are based on assumptions formed from questionable interpretation of

neural data.

2. Connectionist reasoning models are not based on rules and do not have a central executive. Knowledge is distributed throughout the network. Reasoning occurs in parallel operations of network units. Knowledge exists in the connections between units that are formed by learning. Representations are generally unintelligible at all stages. Deductive reasoning Johnson-Laird (1995) proposed that deductive reasoning (“top-down” logic linking premises and conclusions) consists of the construction of nonverbal, nonpropositional “mental models” of reality. Multiple alternative mental models, and their potential consequences, are tested against reality. Functional neuroimaging studies of deductive reasoning An implication of the Johnson-Laird view is that model construction and testing takes place in nonverbal processing areas of cortex. This view is supported by evidence that right-hemisphere lesions impair nonverbal aspects of reasoning.

However, some functional imaging studies do suggest that deductive reasoning depends on language areas of the left hemisphere. The results of some lesion studies also agree that the left hemisphere supports deductive reasoning. A PET study by Goel et al. (1998) had subjects perform 3 different kinds of deductive reasoning (syllogism, spatial relational inference, nonspatial relational inference). Control tasks required judgments about the semantic content of the sentences without judgment based on logic. The differences between task and control PET images were left lateralized. Left-hemisphere activation included inferior & middle frontal gyri, as well as temporal and anterior cingulate cortex. The Goel et al. study suggests that:

1) left-hemisphere language areas are used for deductive reasoning 2) left-hemisphere language areas are activated in spatial reasoning 3) spatial & nonspatial deductive reasoning both use the same left-

hemisphere cortical areas 4) strong activation of dorsolateral PFC likely indicates that this region is

involved in the integrative processing required by deductive reasoning. It is likely that non-linguistic cortical areas are also involved in deductive reasoning. These other areas may provide weaker and/or more time-limited contributions that do not show up in neuroimaging.

Another PET study by Houdé et al. (2000) investigated the cortical basis of reasoning (about colored geometrical forms). Subjects initially made numerous errors due to incorrect perceptual bias from misinterpreting the rule. However, through training they were able to develop a logical bias that overcame the perceptual bias and improved performance. The PET results showed a cortical displacement of posterior activation to frontal activation with training. This shift was interpreted as due to the suppression of the perceptual bias from posterior areas as a result of training. The exercise of logical reasoning in the Houdé et al. study thus appears to have come from the PFC taking executive control of the task, which overcame perceptual biasing influences from the posterior cortex.

Conclusions about deductive reasoning: 1) deductive reasoning is a high-level (at the top of the PAC) integrative

process. 2) inferences are reached by sequential matching of alternative new

cortical cognitive networks against sensory input and against existing established networks.

3) left-hemisphere linguistic processing plays a role in the processing. 4) deductive reasoning requires lateral PFC involvement for the integration

of complex information over time and the suppression of alternative conclusions.

5) suppression of inferences that are “logically nearly true” provides a strong load on the information processing capacity.

Problem solving Problem solving is a cognitive process directed at achieving a goal when a solution method is not obvious to the problem solver. Inductive reasoning is used more than deductive reasoning for solving problems in everyday life. Both forms of reasoning appear to share the same cortical substrate. Deductive reasoning seeks to draw and verify logically valid inferences from premises. Validity is the logical consistency between inference and premises. Inductive reasoning seeks to draw plausible inferences from current observations and preexisting knowledge, and to estimate the probability of those inferences. At best, induction attains high probability, not truth. Scientific discovery depends heavily on inductive reasoning, “as natural science is a process of successive probabilistic approximations of personal knowledge to the reality of the physical world”. Ordinary problem solving uses practical intelligence to make inductive inferences leading to the goal of a problem solution.

When the inferences derive from similarities and lead to conclusions by similarity in a cognitive process, the logic is called analogical reasoning. In analogical reasoning, similar stimuli are treated as equal and the problem-solving strategy is re-used on a new similar problem. Analogical reasoning is applied to relationships between stimuli, objects, and events, as well as to the stimuli, objects, and events themselves. The process is similar to that proposed by Gestalt psychology for the making and recognition of percepts. Analogical reasoning involves creation and use of analogical mappings, that are essentially abstract cortical cognitive networks used as gestalts. Problem solving uses neural interactions among cortical networks of established knowledge, as well as of current and recent sensory information, to reach the solution of a problem. Problem solving also calls on attention, perception, memory, and the integration of conditional contingencies to reach the solution goal. The integration is temporal when multiple complex contingencies are involved.

Problem-solving tests In general, there is no widely accepted definition of problem solving that would allow its reliable and valid measurement. The CRESST (Center for Research on Evaluation, Standards, and Student Testing) model of problem solving includes four elements:

1) content understanding 2) problem-solving strategies 3) metacognition (planning, self-monitoring) 4) motivation (self-efficacy, effort)

In humans, the Raven’s Progressive Matrices (RPM) test is used to test analogical reasoning. In nonhuman primates, problem solving is tested using cross-temporal contingency tasks, e.g., delay tasks. These tasks test temporal integration functions, such as working memory and prospective set, in problem solving.

Functional neuroimaging studies of problem solving These studies provide understanding of the cortical topography of the cortical cognitive networks involved in problem solving. Functional neuroimaging during problem solving reveals the activation of large regions of cortex, mostly in the left hemisphere but, also in the right in some cases. Two groups of areas are consistently activated: 1) the first group is in posterior cortex: the location and extent of activation depends on the type of information used to problem solve. Problem solving that involves: a) spatial reasoning (e.g., imagination of 3-D rotational transformation) activates the TPJ region (the junction of the occipital, temporal, and parietal lobes). b) analogical reasoning (e.g., the RPM task) activates the posterior parietal cortex. c) text processing activates Wernicke’s area. 2) the second group is in frontal cortex; areas involved are anterior cingulate cortex, Broca’s area, lateral PFC. The relative timing of the activations of posterior & frontal areas is not revealed by functional neuroimaging: they may be concurrent or sequential.

Conclusions about problem solving Problem solving activates at least two regions:

1) one region of posterior cortex that specializes in the representation and processing of the cognitive content used to arrive at the correct solution.

2) prefrontal cortex (PFC) The problem task is likely to activate both perceptual memory networks, representing sensory and/or semantic content, and executive networks, that carry out execution of the task in time. In the 2000 meta-analysis of Duncan and Owen, 3 PFC regions were activated in many tasks:

1) anterior cingulate cortex (ACC) 2) lateral PFC 3) orbitofrontal cortex (OFC)

These 3 regions appear to contribute:

1) heightened attentive effort 2) temporal integration 3) exclusionary attention: suppression of posterior memory networks that

interfere with the correct solution of the problem

“The integrative role of the prefrontal cortex in problem solving is graded and adjusted to need.” 3 factors determine the degree of involvement of the PFC in problem solving:

1) integration time 2) relational complexity 3) novelty

PFC involvement corresponds to the degree of “mental effort” required. Thus, tasks that have long integration time, are highly complex, and/or are highly novel require the highest degree of PFC involvement, and hence, the most effort. In Kroger et al. (2002), the level of PFC activation correlated with the number of relations to be integrated in a RPM analogical reasoning task. In Waltz et al. (1999), impairment of performance on an analogical task correlated with the degree of the task’s relational complexity in PFC-lesioned patients. In Osherson et al. (1998), more-demanding inductive reasoning problems activated left PFC to a greater degree than less-demanding problems.

An economy of resources in PFC (presumably related to the economy of attentive effort) has been reported: problems elicit less PFC activation as they become automatic and effortless (Jahanshahi et al. 2000; Reichle et al. 2000). Dehaene et al. (1998) speculated that the PFC regulates the contribution of its neurons to a “global workspace” used for problem solving.

Decision making To make a decision to act at a certain time in a certain way seems to imply that a conscious human being must use rational reasoning. However: 1) other species besides humans can make decisions 2) not all decisions are rational 3) some decisions are unconscious 4) many (or most) decisions result from earlier experiences of which we are not aware (i.e., they depend on implicit memory) Nonetheless, a decision to act by any organism is tied to the executive functions of that organism. For this reason, we can assume that decision making by primates is a frontal lobe function. A role for prefrontal cortex in decision making must be the result of “upward expansion” of the role of hierarchically lower executive brain structures in action selection. Decisions have antecedents upon which they are based, just as antecedents provide the premises of an inference in reasoning. The antecedents to a decision by prefrontal cortex may be conveyed to it over its input pathways.

Thus, to understand the neural basis of decision making requires understanding: 1) executive prefrontal function 2) neural inputs to prefrontal cortex Often, the decision to act depends on a prior decision to process a certain kind of sensory information. Therefore, a behavioral decision may be considered to be a product of perception. In fact, there is electrophysiological evidence from both humans and monkeys that neuronal populations in posterior cortical areas are active in response to identical, uncertain, or ambiguous stimuli according to the behavioral decision made. This evidence suggest that behavioral decisions are made (at least partly) during the perceptual processing preceding an action. In other words, it appears that sensory processing categorizes stimuli in accord with the behavioral decisions determined by the stimuli, not just their physical properties. Thus the roots of behavioral decision making are found in posterior cortex before signals from perception reach the executive cortex.

Studies supporting this conclusion: 1) Evoked potentials (recorded at the vertex) reflect the choice of responding

to an identical visual stimulus by one motor act or another (Begleiter & Porjesz 1975).

2) In monkey area MT, cells that selectively respond to moving visual stimuli fire in correlation with the monkeys perceived and reported direction of dot movement, even when the common direction of movement of some dots on a screen is weak (compared to an overall random movement of all dots) and indistinguishable from random by human observers.

Reasoning may be another source of decisions, in addition to perception, and may actually be more important for human reasoning. Decision making may also be influenced by a range of emotional and social factors. In humans, social, esthetic, and ethical values (acquired by education) play an important role. Decision making is influenced by a host of factors carried by inputs to the prefrontal cortex, and its selection of executive memory networks from a host of available ones.