Cognition

Abstract

Cognition is the set of mental processes by which the brain acquires, transforms, stores, and uses information, the machinery that turns sensory input into perception, knowledge, decision, and action. This article treats cognition as the organizing subject of cognitive psychology: it recounts the cognitive revolution that displaced behaviorism by admitting the internal processes between stimulus and response, develops the information-processing metaphor that models the mind as a system of representations and operations, and follows mental chronometry from simple reaction time to the timing of memory scanning and mental rotation. It surveys the major domains of cognition, sets out the three levels at which any cognitive system can be explained, and weighs the embodied-cognition challenge to the classical symbolic view. Three interactive demonstrations sample three of those domains in turn: a psychometric threshold in perception, spatial cueing in attention, and memory scanning.

Keywords: cognition, information processing, mental chronometry

Cognition is the collective name for the mental processes that stand between sensation and behavior: perceiving, attending, remembering, forming concepts, reasoning, using language, solving problems, and deciding. Ulric Neisser, whose 1967 textbook gave the field its name and its charter, defined cognition as all the processes by which sensory input is transformed, reduced, elaborated, stored, recovered, and used (Neisser, 1967). The definition is deliberately broad, because its point was to insist that everything a mind does with information is a proper object of experimental study, at a time when the dominant psychology held that only observable stimuli and responses were admissible (Miller, 2003). The sections below trace how that conviction became a science: the revolution that made internal processes respectable, the information-processing framework that gave them a language, the chronometric methods that made them measurable, the domains they divide into, the levels at which they are explained, and the challenge that the body and the world pose to the classical account.

What Cognition Is

Cognition is defined less by any single process than by a common function: the handling of information. A visual scene falls on the retina as a pattern of light, and cognition is what converts that pattern into a recognized object, a remembered name, a judgment of distance, and a reach of the hand. Between the input and the output lie representations, internal states that stand for something in the world, and processes that operate on them. This is the premise that unifies the field, and it is why perception, memory, and reasoning, which look like separate faculties, are studied as facets of one system. The scope is wide. It runs from the low-level registration of a contour to the high-level manipulation of abstract concepts, and it includes processes that never reach awareness alongside the deliberate reasoning that does. Much of cognition is fast, parallel, and automatic; a smaller part is slow, serial, and effortful, and the relation between the two organizes a great deal of modern theory, from attention to the study of executive function. What makes the whole an object of natural science, rather than of introspection alone, is that these hidden processes leave measurable traces in behavior, chiefly in what people get right and how long they take.

The Cognitive Revolution

For the first half of the twentieth century, mainstream American psychology was behaviorist: it held that a science of mind must confine itself to observable stimuli and responses and treat the intervening organism as a black box whose internal states were unknowable and unnecessary. The cognitive revolution overturned that stricture. George Miller, one of its architects, dated its symbolic birth to a 1956 symposium at the Massachusetts Institute of Technology and described it as the recovery of the mind as a legitimate scientific subject after decades of exile (Miller, 2003). Several currents converged. Miller's own demonstration that immediate memory is limited to about seven chunks showed that the mind imposes a measurable structure on information (Miller, 1956). Noam Chomsky's review of Skinner argued that the generativity of language, the capacity to produce and understand endlessly many novel sentences, could not be explained by conditioned responses and required internal rules (Chomsky, 1959). Allen Newell and Herbert Simon showed that problem solving could be modeled as the manipulation of symbols by a program, making the notion of internal computation concrete rather than metaphorical (Newell & Simon, 1972). Gerald Mandler, surveying the period, cautioned that the shift was less a single overnight coup than the resurfacing of a mentalistic tradition that behaviorism had suppressed but never extinguished (Mandler, 2002). Whichever framing one prefers, the outcome was the same: by the 1960s the internal processing of information had become the central subject of psychology, and cognition had a science.

The Information-Processing Approach

The framework that gave the new science its working language was the information-processing approach, which models the mind as a system that receives inputs, encodes them into representations, stores and transforms those representations through a series of stages, and produces outputs. The digital computer supplied the guiding analogy, not because the brain is believed to be a computer in any literal sense but because the computer showed how a physical system could carry out symbol manipulation, and so made internal computation a respectable thing to posit. Figure 1 lays out the canonical flow. Information passes from a brief sensory register through a bottleneck of attention into a limited-capacity working memory, where it is actively maintained and operated on, and from there into a vast long-term memory from which it can later be retrieved. Alan Baddeley's multicomponent model refined the workspace at the center of this diagram, adding an episodic buffer that binds information from different codes into a unified representation (Baddeley, 2000). The most influential early quantification of a stage came from Miller, whose magical number seven captured the narrow capacity of immediate memory and, crucially, showed that the limit is measured in chunks rather than raw items, so that recoding several items into one chunk multiplies how much can be held (Miller, 1956). The framework was never only about storage. Craik and Lockhart recast encoding itself as a matter of the depth at which information is processed, meaning-based analysis leaving a more durable trace than surface analysis, which reframed memory as a byproduct of the operations cognition performs rather than a passive deposit (Craik & Lockhart, 1972). At the far end of this tradition, John Anderson assembled the stages into a unified computational architecture, ACT, in which a single set of mechanisms for representing knowledge and applying production rules aims to account for the full range of complex cognition (Anderson, 1996). Because that limit is counted in chunks rather than raw items, recoding several items into a single chunk multiplies how much the narrow store at the heart of the diagram can effectively hold, which is why an expert who groups material into larger units remembers far more of it (Miller, 1956).

Mental Chronometry

If cognition is a sequence of internal processes, the most direct evidence of their existence is the time they take. Mental chronometry, the use of reaction time to infer the structure of unobservable processing, is the signature method of cognitive psychology, and its logic is that an added mental step adds measurable time. The idea has a long lineage, reaching back to the subtractive method of the nineteenth-century physiologist Franciscus Donders, who reasoned that inserting an extra decision into a task should lengthen the response by exactly the duration of that decision, so that subtracting the times of two tasks isolates the added step. Saul Sternberg gave the method its modern form with a memory-scanning task: participants held a short set of digits in mind and judged whether a probe digit was among them, and their response time rose linearly with the size of the set, by a constant increment for each additional item (Sternberg, 1966). The slope of that line, a few tens of milliseconds per item, estimates the duration of a single comparison, and the fact that the slope is the same for present and absent probes implies that the scan is exhaustive, checking every item even after a match is found. A second landmark extended chronometry to internal imagery. Roger Shepard and Jacqueline Metzler showed observers pairs of three-dimensional shapes and asked whether they were the same object at different orientations; the time to decide rose linearly with the angular difference between them, as though the observer were mentally rotating one shape into alignment with the other at a roughly constant rate (Shepard & Metzler, 1971). The linearity is the crucial result, because it shows that a mental operation can have the metric structure of the physical transformation it represents. Chronometry also underwrote the cognitive analysis of attention: Posner and Petersen used the timing of responses to cued and uncued locations to decompose attention into distinct networks for alerting, orienting, and executive control, each with its own signature and neural substrate (Posner & Petersen, 1990). Across all three cases the logic is the same: a linear function whose slope measures a hidden operation, the increment per item in scanning, the added time per degree in rotation, or the cost of reorienting attention, turns the invisible structure of processing into a number.

The Domains of Cognition

Cognition is conventionally divided into domains, each a research tradition with its own paradigms, though the divisions are ones of emphasis rather than sharp boundaries, and every real act of thought recruits several at once. Perception builds structured representations of the world from sensory signals. Attention selects among competing inputs and allocates limited processing capacity (Posner & Petersen, 1990). Memory encodes, retains, and retrieves information across every timescale, from the seconds of working memory to the years of the long-term store (Baddeley, 2000). Language comprehends and produces structured symbolic communication. Higher cognition, the domain of reasoning, judgment, problem solving, and decision making, manipulates representations to reach conclusions and guide action. Table 1 sets out the domains with their central questions and characteristic methods. The point of the taxonomy is not to carve the mind at fixed joints but to organize inquiry, and the deepest questions in the field concern how the domains interact: how attention gates what enters memory, how language draws on and reshapes conceptual knowledge, and how the fast automatic processes of perception feed the slow deliberate processes of reasoning.

Note. The domains are traditions of inquiry rather than separate organs; a single act of reading recruits perception, attention, memory, and language together. The attention and memory rows follow the chronometric and multicomponent traditions (Posner & Petersen, 1990; Baddeley, 2000).

The three demonstrations that follow sample the first three rows of the table in turn, one method per domain, so that the interactives span perception, attention, and memory rather than crowd into any one. The first takes up perception, studied by psychophysics, which relates the intensity of a stimulus to the probability of detecting it. The resulting psychometric function is an S-shaped curve whose midpoint defines the absolute threshold, the intensity an observer detects half the time; the demonstration lets the reader trace that curve and read the threshold off it (Wichmann & Hill, 2001).

The second turns to attention, probed by spatial cueing, in which a cue draws attention to a location before a target appears. Relative to a neutral baseline, a valid cue speeds the response and an invalid one slows it, and the size of that benefit and cost is the chronometric measure of orienting (Posner, 1980). The demonstration models these costs and benefits; the mechanism is treated at length in the companion article on the orienting of attention.

The third returns to memory and the chronometric task introduced above. Holding a set of digits in mind and judging whether a probe belongs to it, the response time rises by a fixed increment for each added item, the signature of a serial scan whose equal slope for present and absent probes marks it as exhaustive (Sternberg, 1966). The demonstration lets the reader vary the set size and watch the predicted time.

Higher cognition also extends beyond reaching a conclusion to what happens once a choice is made, where thought proves to be motivated as well as inferential. In the founding demonstration, Leon Festinger and J. Merrill Carlsmith showed that people who acted against their beliefs for insufficient reward revised the beliefs to fit the behavior, evidence that reasoning serves the reduction of internal inconsistency and not only the pursuit of accuracy (Festinger & Carlsmith, 1959). The mechanism is treated in the companion article on induced compliance and cognitive dissonance.

Levels of Analysis

A recurring source of confusion in cognitive science is that a mental process can be described in several ways at once, and the descriptions are not rivals but answers to different questions. David Marr made the point canonical by distinguishing three levels at which any information-processing system must be understood. The computational level asks what problem the system solves and why, specifying the mapping from inputs to outputs as an abstract task. The algorithmic level asks how the system solves it, which representations it uses and which procedures transform them. The implementational level asks how those representations and procedures are physically realized, in neurons or in silicon (Marr, 1982). The three are logically independent: one computation can be carried out by many algorithms, and one algorithm by many physical substrates. Marr's warning was that explaining cognition at the wrong level, or conflating levels, produces confusion, and that an account of the neural hardware alone can never explain why a process computes what it does. Unified architectures such as Anderson's ACT can be read as attempts to specify the algorithmic level with enough precision that it connects downward to the brain and upward to the tasks a person performs (Anderson, 1996). The framework remains the standard grammar for saying what kind of explanation a given theory offers.

The Embodied Challenge

The classical view that emerged from the cognitive revolution treats cognition as the rule-governed manipulation of amodal symbols, representations whose form is arbitrary with respect to what they stand for, much as the bit pattern in a computer bears no resemblance to the number it encodes. That view has been challenged from two directions. The first is a debate about architecture. Connectionist models compute with distributed patterns of activation across simple units rather than discrete symbols, and Jerry Fodor and Zenon Pylyshyn mounted the classic defense of the symbolic view, arguing that thought is systematic and productive in ways that require a combinatorial syntax of the kind symbols provide and that networks, on their own, do not deliver (Fodor & Pylyshyn, 1988). The second challenge is deeper. Embodied and grounded approaches hold that concepts are not amodal symbols at all but are represented in the same perceptual, motor, and affective systems used to interact with the world, so that understanding the word grasp partially reactivates the neural machinery of grasping. Lawrence Barsalou has developed the leading account of grounded cognition, marshaling evidence that conceptual processing routinely engages modality-specific brain systems and is shaped by the body's states and actions (Barsalou, 2008). Margaret Wilson sorted the many claims traveling under the embodied banner into six distinct theses, some well supported and some contentious, and cautioned that the strongest versions, which deny any role for internal representation, outrun the evidence (Wilson, 2002). The debate is unresolved, and its stakes are high, because it concerns whether the representations at the heart of cognition are abstract and self-contained or are anchored in the sensing, acting body.

Worked Example

The logic of mental chronometry can be made quantitative with the memory-scanning task in the demonstration above. Sternberg found that the time to judge whether a probe belongs to a memorized set is well described by a straight line, response time equals an intercept plus a slope multiplied by the set size, where the intercept collects the fixed costs of encoding the probe and executing the response and the slope estimates the duration of one comparison (Sternberg, 1966). Take the illustrative fit used in the demonstration, an intercept of 397 milliseconds and a slope of 38 milliseconds per item. For a set of two items the model predicts 397 plus 38 multiplied by 2, which is 397 plus 76, or 473 milliseconds. For four items it predicts 397 plus 38 multiplied by 4, which is 397 plus 152, or 549 milliseconds. For six items it predicts 397 plus 228, or 625 milliseconds. Each additional item adds the same 38 milliseconds, and that constancy is the whole point: a flat increment per item is the signature of a serial scan that examines the items one at a time. The exhaustive character of the scan is read from a second feature of the data, that the slope is the same whether the answer is yes or no. If the scan stopped as soon as it found a match, positive responses would need to check only half the set on average and their slope would be half that of negative responses; because the two slopes are equal, the scan must run to the end of the set every time. The numbers here are illustrative, but the shape, a straight line whose slope measures a hidden process and whose form reveals that process to be serial and exhaustive, is exactly what the reaction-time data show.

Discussion

Cognition matters first because it is the subject that organizes an entire science: perception, attention, memory, language, and reasoning are not a loose federation of topics but facets of a single system that acquires and uses information, and treating them as such is the founding insight of the field (Neisser, 1967). It matters second because the methods that make hidden processes visible, above all the chronometric inference of structure from timing, turned introspective questions into experimental ones and remain among the most productive tools in psychology (Sternberg, 1966). It matters third because the framework has proven extensible: the information-processing metaphor that began with boxes and arrows has grown into precise computational architectures on one side (Anderson, 1996) and, on the other, into a live debate about whether representations are abstract symbols or are grounded in the body and the world (Barsalou, 2008). The open questions are correspondingly deep. How the levels of Marr's analysis connect, so that a computational account of a task meshes with the algorithms that perform it and the neurons that implement them, is unsettled. Whether the fast automatic processes and the slow controlled ones are two systems or a continuum is contested across attention, memory, and reasoning. What is settled is the revolution's central claim, now so thoroughly assimilated that it is easy to forget it was ever radical: that the processes between stimulus and response are real, structured, and open to measurement (Miller, 2003).

Common Misconceptions

Cognition is the same thing as conscious, deliberate thinking.

Most cognition is fast, parallel, and automatic, running below awareness. The attention networks that orient the mind and the encoding operations that build memories proceed without deliberate control, and are studied precisely because their effects on behavior can be measured even though they are not introspectively available (Posner & Petersen, 1990). Conscious reasoning is a small, late part of what cognition does.

The information-processing metaphor claims the brain literally is a digital computer.

The computer is an analogy that made internal computation respectable, not an identity claim about neural hardware. Embodied and grounded accounts argue that many representations are anchored in perceptual and motor systems rather than realized as amodal symbols, and even sympathetic reviews warn that the strongest literal versions of the metaphor outrun the evidence (Wilson, 2002). The metaphor is a level of description, not a wiring diagram.

Cognition is walled off from the body and the emotions.

A growing body of evidence indicates that conceptual processing routinely reactivates the sensory, motor, and affective systems used to interact with the world, so that understanding an action word engages the neural machinery of acting (Barsalou, 2008). On this view the body is not a peripheral input device but part of the substrate of thought itself.

Glossary

Attention.

The set of processes that select among competing inputs and allocate limited processing capacity, decomposable into alerting, orienting, and executive control networks.

Chunk.

A unit of meaningful organization in memory; recoding several items into one chunk increases how much information the narrow immediate store can hold.

Cognition.

The collective mental processes by which sensory input is transformed, stored, retrieved, and used to produce knowledge and action.

Cognitive revolution.

The mid-twentieth-century shift that displaced behaviorism and made the internal processing of information the central subject of psychology.

Computational level.

In Marr's scheme, the description of what problem a cognitive system solves and why, independent of the algorithm or the physical hardware.

Embodied cognition.

The view that cognitive representations are grounded in the perceptual, motor, and affective systems used to interact with the world rather than in amodal symbols.

Information processing.

The framework that models the mind as a system encoding, storing, transforming, and retrieving representations through a series of stages.

Levels of processing.

The principle that memory strength depends on the depth of analysis at encoding, meaning-based processing leaving a more durable trace than surface processing.

Long-term memory.

The durable store that retains knowledge, events, and skills over intervals from minutes to a lifetime, with no measured ceiling on capacity.

Mental chronometry.

The use of reaction time to infer the number, duration, and arrangement of the unobservable processes that intervene between stimulus and response.

Mental rotation.

The imagined turning of a shape, whose duration rises linearly with the angle, showing that an internal operation can mirror the structure of a physical transformation.

Metacognition.

Cognition about cognition, the monitoring and control of one's own mental processes, such as judging what one knows or how well one is learning.

Physical symbol system.

Newell and Simon's hypothesis that intelligent action consists in the manipulation of symbol structures by a physical system operating under rules.

Reaction time.

The interval between a stimulus and a response, the primary dependent measure of cognitive psychology and the basis of chronometric inference.

Schema.

An organized structure of prior knowledge that guides how new information is perceived, interpreted, and remembered.

Working memory.

The limited-capacity system that holds and manipulates information in the service of ongoing cognition, the active workspace of the information-processing model.

Key Researchers

Lawrence W. Barsalou. Professor at the University of Glasgow and the leading proponent of grounded cognition; he has assembled the evidence that conceptual processing reengages the modality-specific systems of perception and action.
University of Glasgow - ORCID - Google Scholar - Wikipedia

George A. Miller (1920-2012). Psychologist at Harvard, Rockefeller, and Princeton and a principal architect of the cognitive revolution; his demonstration that immediate memory is limited to about seven chunks and his later histories of the field helped establish the information-processing study of mind.
Wikipedia

Ulric Neisser (1928-2012). Psychologist at Cornell and Emory whose 1967 textbook named cognitive psychology and defined its scope, later pressing the field to study cognition as it operates in the everyday world.
Wikipedia

Michael I. Posner. Professor emeritus at the University of Oregon; his chronometric and imaging studies decomposed human attention into distinct alerting, orienting, and executive networks, linking cognitive functions to specific brain systems.
University of Oregon - Wikipedia

Herbert A. Simon (1916-2001). Polymath at Carnegie Mellon and a founder of artificial intelligence and information-processing psychology; with Allen Newell he modeled problem solving as symbol manipulation and advanced the physical-symbol-system account of intelligence.
Wikipedia

Frequently Asked Questions

What is cognition?
Cognition is the collective name for the mental processes that acquire, transform, store, and use information, including perception, attention, memory, language, reasoning, and decision making. Neisser defined it as all the processes by which sensory input is transformed, reduced, elaborated, stored, recovered, and used, a definition broad enough to make everything the mind does with information a subject of experimental study (Neisser, 1967).

What was the cognitive revolution?
The cognitive revolution was the mid-twentieth-century shift that displaced behaviorism, which had confined psychology to observable stimuli and responses, and made the internal processing of information the central subject of the field. George Miller dated its symbolic birth to a 1956 meeting at MIT, though historians note it also revived an older mentalistic tradition that behaviorism had suppressed (Miller, 2003; Mandler, 2002).

Is the mind really a computer?
No. The computer is an analogy that showed how a physical system could carry out symbol manipulation, which made it respectable to posit internal computation, but it is a level of description rather than a claim about neural hardware. Grounded and embodied accounts argue that many representations are anchored in perceptual and motor systems rather than realized as abstract symbols (Wilson, 2002).

What is mental chronometry?
Mental chronometry is the use of reaction time to infer the structure of the hidden processes between stimulus and response, on the principle that an added mental step adds measurable time. Sternberg used it to show that memory scanning is serial and exhaustive, and Shepard and Metzler used it to show that mental rotation takes time in proportion to the angle turned (Sternberg, 1966; Shepard & Metzler, 1971).

What are the main domains of cognition?
The conventional domains are perception, attention, memory, language, and higher cognition, the last covering reasoning, judgment, problem solving, and decision making. The divisions are traditions of inquiry rather than separate organs, and any real act of thought recruits several at once, as attention gates what enters memory and language draws on conceptual knowledge (Posner & Petersen, 1990).

What are the three levels of analysis?
David Marr distinguished the computational level, which specifies what problem a system solves and why, the algorithmic level, which specifies the representations and procedures that solve it, and the implementational level, which specifies how these are physically realized. The levels are independent, since one computation can be run by many algorithms and one algorithm by many substrates (Marr, 1982).

What is embodied cognition?
Embodied or grounded cognition is the view that concepts are represented in the same perceptual, motor, and affective systems used to interact with the world, rather than as amodal symbols detached from the body. Barsalou has developed the leading version, while reviewers caution that the strongest claims, which reject internal representation entirely, go beyond the evidence (Barsalou, 2008; Wilson, 2002).

How do psychologists study cognition?
The core method is to infer unobservable processes from measurable behavior, chiefly accuracy and reaction time, supplemented by neuroimaging that links processes to brain systems. A linear rise in response time with set size, for instance, reveals a serial scan, and cueing paradigms expose the separate networks of attention (Sternberg, 1966; Posner & Petersen, 1990).

References