RESEARCH DOSSIER
Mental Processes and Functional Systems
Abstract. Mental processes are temporally organized transformations that contribute to perception, attention, memory, learning, emotion, motivation, language, and action. This fundamental research monograph distinguishes processes, states, dispositions, capacities, and functional systems through scientific definitions, experimental methods, and temporal criteria. It traces the historical development of functional explanation and critically examines component-based cognitive theories, latent-variable studies, neuropsychological evidence, distributed brain networks, computational modeling, developmental dynamics, and situated cognition. Research findings are compared with attention to construct validity, task impurity, inferential limitations, causal identification, reproducibility, and generalization. The synthesis also examines external cognitive supports and human–artificial collaboration in the context of established research on cognitive offloading and distributed cognition. Open problems concern the identification of psychological constructs, causal dependencies across levels of analysis, context sensitivity, and the evidence needed for cumulative explanations of functional systems.
Keywords: mental processes; functional systems; cognitive psychology; attention; working memory; executive function; emotion and cognition; psychological dynamics; neural networks; construct validity; causal inference; cognitive offloading; distributed cognition; Artificial Era
Mental Processes as Temporally Organized Operations
A mental process is an ordered transformation of information, bodily signals, goals, or action tendencies whose successive states matter for explanation. Perceiving a moving object, maintaining an instruction, recovering a word, deciding between actions, and regulating anxiety all unfold through transitions; a single score at their endpoint cannot disclose their internal organization. A functional system is the coordinated ensemble of operations and enabling conditions that accomplishes a specified task or maintains a specified relation with an environment. Both concepts are relational. The system boundary depends on the explanandum, observation scale, and counterfactual tests. The classic vocabulary of mental functions is useful when attached to operational criteria rather than presumed anatomical compartments. Marr (1982/2010) distinguishes computational purpose, representations and algorithms, and implementation, clarifying why one function can admit several descriptions.
The research object of this monograph is the relationship between a process's temporal course and the organization that makes its work possible. It asks when a sequence of observations warrants identifying a process, how several processes jointly realize a function, whether their coordination is stable across contexts, and how evidence distinguishes causal dependencies from correlated measurements. Consider reading a sentence while monitoring traffic. The same organism must sample sensory input, allocate attention, maintain linguistic context, forecast consequences and adjust movement. Researchers can isolate parts of this activity experimentally, but any proposed functional system must explain their coordination during natural performance. Mesulam (1990) provides an early distributed-network account of attention, language, and memory that makes such coordination a central problem.
A functional system is specified here by an observable objective or constraint, a set of candidate contributing operations, the transitions connecting them, environmental affordances, and predictions about what happens after perturbation. A claimed system is stronger when its account predicts interference, substitution, recovery and transfer under independently manipulated conditions. The same operation may contribute to different systems, and different operations can sometimes produce equivalent task outcomes. This treatment retains the historical insight that psychological functions arise from organized activity while keeping the question of subjective experience separately investigable. Krakauer et al. (2017) emphasizes that behavioral explanation has its own indispensable evidential content, even when neural measurements are exceptionally detailed.
Processes, States, Dispositions, and Functional Capacities
The vocabulary of psychology mixes events with regularities unless a theory supplies explicit distinctions. A process is a sequence of changes during an interval; a state is a relatively bounded configuration at a specified moment or observation window; a disposition is a relatively persistent conditional tendency expressed across appropriate circumstances; and a capacity is the empirically supported range of performance a system can attain under defined demands. Attention shifting is a process, a currently prioritized target is a state, a tendency to redirect attention rapidly is a disposition, and successful redirection over calibrated tasks indicates a capacity. These distinctions concern the measurement model, rather than metaphysical categories. Cronbach & Meehl (1955) establishes the importance of locating psychological constructs inside a network of theoretically meaningful relations.
Operational consequences follow immediately. Momentary response time estimates an outcome under one condition; it does not independently establish a stable attentional disposition. A multi-session task battery with sufficient trials and a credible factor structure may better support a statement about individual differences, although task-specific variance remains. A process model requires observations dense enough to capture transitions. A capacity claim requires manipulations that vary difficulty and assess ceiling, floor and strategy effects. Within-person variation and between-person variation can exhibit different covariance structures, so averaging over participants can conceal actual trajectories. Miyake et al. (2000) demonstrates why latent-variable methods can help separate shared executive-function variance from more specific demands.
Functional capacities should also be distinguished from normative success. A person may efficiently memorize a sequence yet select an inappropriate goal; the system has high mnemonic performance under that condition and poor overall fit to another evaluative criterion. Likewise, the presence of a long response interval may reflect caution rather than weak processing, and an accurate response may depend on environmental aids. Naming the criterion is therefore part of defining the function. This chapter's organizing rule is to report process evidence at the temporal resolution measured, state evidence at the sampled configuration, disposition evidence across repeated contexts, and capacity evidence across calibrated demands. Such reporting prevents theoretical constructs from being treated as if they were direct raw measurements. Poldrack (2006) adds a parallel caution for inferring psychological processes from neural activation patterns.
Historical Traditions of Functional Explanation
The historical problem of mental function precedes modern neuroimaging. Experimental psychology established reproducible relationships between stimulation and behavior; subsequent functional approaches asked how mental operations support purposeful activity in changing environments. The cognitive revolution reopened the study of internal transformations under operational constraints, while neuropsychological approaches investigated dissociations among language, memory, attention, and action following neurological damage. Neisser (1967) consolidated an influential cognitive account in which perception, pattern recognition, memory, and language are understood as information-processing problems. Historical terminology varies across traditions, and words such as function, faculty, operation, and system carry different commitments. The durable scientific question concerns what is measured, how operations interact, and which explanations survive intervention.
In Miller et al. (1960), Miller, Galanter, and Pribram described plans and the Test–Operate–Test–Exit organization of goal-directed behavior. Their work helped articulate feedback and goal comparison as alternatives to elementary stimulus–response chains. Later computational models made representations, algorithms, constraints and implementation explicit, allowing explanations to be compared at several scales. The significance of these milestones lies in a common shift toward formally describable organization. Their particular hypotheses are historical proposals requiring independent evaluation. A feedback circuit can explain monitoring and adjustment without thereby explaining all emotional processes or every mode of conscious experience.
A parallel history developed through brain-behavior research. Mesulam (1990) described large-scale neurocognitive networks in which the same psychological domain has distributed contributors and a single site may participate in different functional arrangements. Posner & Petersen (1990) differentiated attentional operations through converging behavioral and anatomical evidence. These studies helped replace one-to-one mappings between broad mental faculties and anatomical locations with more differentiated accounts. Yet network language alone does not settle the problem: the network must be identified with defensible measurements and related to observable function. The historical synthesis is a methodological lesson. Each generation gains explanatory power when functional claims are tied to a clearly specified task, mechanism, temporal window and falsifiable pattern of evidence.
Constructs, Tasks, and the Problem of Inference
A psychological task is an experimental arrangement, whereas a psychological construct is a hypothesized explanatory property or process. The two are connected through a measurement model. Errors in this connection can arise when a task involves several unmeasured operations, when a manipulation changes motivation as well as processing, or when participants select different strategies. Inhibition tasks, for instance, require perceiving cues, maintaining rules, preparing responses, monitoring conflicts and executing an action. A poorer score may have several causal origins. Miyake et al. (2000) demonstrated the utility of distinguishing partially correlated executive components using multiple tasks and latent variables, an approach that reduces reliance on any single impure indicator.
The central epistemic problem is underdetermination: the same observed response can be generated by different internal processes. Slower decisions may reflect accumulating evidence, boundary adjustment, fatigue, motor delay, or strategic caution. Brain activation does not automatically dissolve the ambiguity, because regions vary in selectivity and may support several computations. Poldrack (2006) formalized the limitations of reverse inference from activated regions to cognitive states. Construct evaluation therefore requires convergent manipulation, alternative-model testing, ecological observations, and transparent assumptions about response generation. A well-fitting model is a candidate explanation whose adequacy must be judged by more than fit to the data on which it was developed.
The present research treats functional-system hypotheses as claims about organization inferred from coordinated evidence. A proposed system should establish that component measures are appropriately reliable, identify conditions under which components interact, test whether the interactions improve explanation of behavior, and specify predictions outside the fitting sample. Model comparison is especially important when broad language describes attention, working memory, executive function, and cognitive control as if each were a unique, independent module. The emphasis on construct validity follows Cronbach & Meehl (1955); the empirical challenge of discovering more stable psychological categories is illustrated by Eisenberg et al. (2019). The appropriate outcome of research can be refinement, subdivision, or abandonment of a proposed functional boundary.
Temporal Scales, Event Boundaries, and System Identity
Mental activity runs at several coupled timescales. Perceptual evidence changes over milliseconds, attentional selection and response preparation unfold over shorter task episodes, learning may accumulate over repeated exposures, and dispositions or social routines are examined over weeks or years. An explanation that uses one averaged interval to stand for all these processes risks losing their order and mutual dependencies. The timescale of a functional system is therefore an empirical parameter: investigators must ask how quickly components change, whether coordination persists between episodes, and when a different configuration should count as a reorganized system. Baddeley (2000) illustrates why short-term integration and access to long-term representations can require explicit bridging constructs.
Time-locked experiments distinguish preparation, stimulus processing, decision, movement and feedback when sampling is sufficiently dense. Longitudinal observations can distinguish stable individual differences from adaptive change. But the temporal resolution of an instrument should not be mistaken for the temporal resolution of the underlying phenomenon. Functional magnetic resonance imaging measures hemodynamic correlates of neural activity; behavioral chronometry measures elapsed responses; experience sampling measures contextually reported states. Each captures a different part of the explanatory target. Cole et al. (2013) shows task-dependent changes in functional connectivity compatible with flexible task-control organization, while the interpretation of those changes remains constrained by the observational properties of connectivity measurement.
A useful criterion of functional continuity is preservation of a system's explanatory relations under controlled perturbation, rather than persistence of every component. A practiced task can remain reliably performed when an external cue substitutes for an internal reminder, yet its causal organization has changed. Conversely, the same measured brain region can appear across different tasks while its partners, timing and contribution differ. Event boundaries need explicit criteria, such as changes in goals, contingencies, behavioral output, or coupling structure. The resulting account can examine both stable and transient systems. It also prevents a category such as memory or attention from being reduced to one convenient sampling window. The scientific challenge is to connect multiple resolutions through models that predict transitions.
Perception as Active Evidence Integration
Perception is an unfolding relation between sensory signals, prior learning, ongoing movement and contextual expectations. Early and later processing stages can be distinguished experimentally, but perception during natural activity involves recurrent updating and strategic sampling. A person following a moving target must orient the eyes, estimate changes, incorporate uncertainty and correct predictions. The relevant functional system extends beyond an isolated moment of sensory transduction. Neisser (1967) examined perceptual analysis and recognition as linked operations, while Marr (1982/2010) showed how a perceptual problem can be formulated at computational, representational and implementation levels. These complementary perspectives warn against defining the function entirely by the stimulus display or by the anatomical location in which an experiment registers a signal.
The same perceptual judgment can be achieved through alternative cue combinations, an important reason to distinguish function from mechanism. When a cue becomes unreliable, observers may alter its weight or shift their sampling strategy. Experimental studies can vary signal strength, cue agreement and feedback to determine whether an account predicts response accuracy, confidence and time. Observed behavioral adaptation may reflect learning, strategy or motor preparation in addition to perceptual inference; experimental contrasts are needed to separate them. Friston (2010) provides a predictive theoretical formulation linking perception and action, but its scope requires model-specific tests, since a broad optimization principle by itself may fit many incompatible mechanisms.
Functional-system analysis asks whether perceptual representations are useful because of downstream coordination. For example, recognizing a road sign has different consequences in a laboratory naming task and while driving. The sign's visual identity interacts with memory, attention, intended movement, and expected outcomes. Psychological evidence should therefore include both controlled perceptual discrimination and ecologically relevant decisions. The relationship is causal when manipulating perceptual evidence predictably changes a specified later component while relevant alternatives are controlled. It remains correlational when the association is inferred from simultaneous measurements alone. The distinction matters because perception is both a scientifically isolable process and a component recruited within broader systems of action. Krakauer et al. (2017) argues for recovering these behavioral questions in neuroscience.
Attention as Selection, Maintenance, and Reorientation
Attention refers to a family of operations that prioritize information and action under constraints. Orienting to a location, sustaining a task set, resolving conflict, and responding to a sudden event demand partly different operations. A claim that attention is one process ignores this differentiation; a claim that every experimental task measures an entirely independent attention subsystem ignores their shared demands. Posner & Petersen (1990) offered an influential systems account of attentional organization, and Petersen & Posner (2012) revisited it in light of subsequent behavioral and brain evidence. Their historical development illustrates how a functional vocabulary can remain useful while the underlying anatomical and computational assignments are refined.
Selective-attention experiments vary target location, competing information, expectations, and response rules to estimate costs and benefits of prioritization. Valid and invalid cues can indicate orienting; conflict tasks measure performance changes when irrelevant stimulus information competes with task rules. These effects reveal systematic behavioral organization. Their precise psychological interpretation depends on task design, baseline conditions, speed–accuracy tradeoffs, and strategy. Attention also interacts with learning: frequently selected material may acquire priority, and habitual stimulus responses can capture behavior after circumstances change. Botvinick et al. (2001) proposed conflict monitoring as one computational route through which demand for control could be registered, providing testable predictions rather than a complete account of every attentional phenomenon.
A functional definition should specify what selection accomplishes: improved access to relevant input, reduced influence of distraction, maintained responsiveness to hazards, or successful alternation among goals. These objectives may conflict. Strong focus can improve the current task while impairing detection of an unexpected signal. Claims about attentional capacity therefore must name the criterion and context. The next research step is to relate momentary selection trajectories to longer-lasting attentional states and individual differences, using several tasks and repeated sessions. Evidence for specialized systems should rest on selective manipulations and reliable dissociations, not on the use of distinct task labels. Poldrack (2006) explains why neural localization alone cannot adjudicate between competing functional explanations.
Working Memory, Binding, and Temporary Control
Working memory comprises temporary information maintenance together with operations that select, update and use maintained content for current activity. A central issue is how several modalities, remembered knowledge and goals become coordinated at the moment a task demands them. Baddeley & Hitch (1974) formulated an influential multicomponent model, and Baddeley (2000) proposed an episodic buffer to address multimodal binding and integration with long-term memory. The revision is instructive: a model that initially explained separable storage components required an additional account of their interaction. Functional interpretation thus depends on both differentiation and the mechanism by which differentiated resources are brought together during performance.
Capacity estimates depend on what counts as an item, whether rehearsal is prevented, how attention is controlled and which representations participants can group into chunks. Cowan (2001) reexamined short-term storage capacity and argued that constraints should be measured under conditions that minimize confounding rehearsal and chunking. A numerical estimate extracted from one paradigm must be interpreted as a result of that paradigm and its assumptions. More generally, maintenance accuracy can reflect perceptual encoding, attentional selection, mnemonic precision, interference, and response mapping. Models should predict performance across load, delay, distraction, and transformation demands rather than appeal to a single universal resource metaphor.
The system-level question is how temporary representations guide adaptive transitions. Reading a long sentence requires preserving earlier relations while processing later words; planning a sequence requires maintaining the goal as intermediate actions unfold. Disruption during a delay may leave one subcomponent intact while impairing the integration of those components. Such dissociations provide a stronger basis for functional decomposition than labels alone. Working-memory interventions are most informative when they manipulate a hypothesized operation and assess downstream effects without changing several other demands simultaneously. The enduring lesson of Baddeley (2000) is that integration can be a specific explanatory problem. Whether a proposed buffer, state-space model, or recurrent network best explains a task requires discriminating predictions.
Long-Term Memory, Retrieval, and Reconstructive Use
Long-term memory includes multiple ways in which prior experience alters later activity. A remembered event may guide future simulation, a learned category may shape perception, and repeated practice may change response selection even when a person cannot narrate the acquisition history. These examples differ in what is retained, how it is accessed and what kind of evidence supports the inference. A useful functional analysis separates encoding operations, subsequent storage-related change, retrieval conditions, and the use made of retrieved content. Hassabis et al. (2007) documented difficulties imagining novel experiences in patients with hippocampal amnesia, demonstrating an empirical connection between remembering and constructing possible scenes that extends beyond simple reproduction of past events.
Memory experiments commonly manipulate encoding depth, interference, retrieval cue, delay, and recognition format. Each manipulation samples particular mechanisms and cannot simply be generalized to every form of memory. A participant who fails free recall may succeed with recognition cues; the difference can involve retrieval support, decision criterion, or representational availability. In systems neuroscience, anatomical findings are strongest when joined with carefully differentiated behavioral performance and lesion or longitudinal evidence. Mesulam (1990) situates memory within distributed neurocognitive arrangements, suggesting that successful recall depends on coordinated contributions rather than a solitary memory store. This interpretation remains compatible with functional specialization when specialized computations are specified.
For this dossier, the distinctive issue is retrieval as a functional contributor to ongoing processing. Recollection can reactivate a goal, change the interpretation of a perception, or supply alternatives for a decision. An apparent memory deficit in complex tasks may therefore originate in deficient selection or use of retrieved content. The reverse is also possible: accurate immediate retention can coexist with difficulty incorporating earlier experience into flexible action. Multi-measure experiments can separate storage-dependent failure from failures in the coordination of retrieval, attention, and decision. The implication is methodological rather than a universal claim about a single memory system. Operational definition should follow the measured transformation and its task relevance. Marr (1982/2010) helps distinguish the computation required from a particular assumed implementation.
Executive Control and Reconfiguration of Goals
Executive control describes processes that make behavior sensitive to instructions, competing responses, changing rules and longer-term objectives. It is often inferred from the ability to shift between tasks, update working-memory content or resist interference. Yet each laboratory task recruits many elementary operations, and their reliability varies. Miyake et al. (2000) examined shifting, updating and inhibition as partially separable yet related latent constructs; later interpretation requires attention to the exact indicators and population studied. An executive function is therefore best specified as a hypothesized class of operations supported by a measurement model, not as a faculty that can be located from one impaired test.
Control operates through detection of difficulty, selection of a policy, assignment of processing priority, and adjustment following feedback. Botvinick et al. (2001) developed a computational conflict-monitoring account linking conflict signals to subsequent control adjustments. Shenhav et al. (2013) proposed that the expected value of exerting control provides an integrative framework for allocation decisions. The accounts differ in their formal commitments and levels of description. Conflict may contribute evidence that intervention is needed; expected value may help determine whether effort is worthwhile. Neither model should be accepted solely because a task is difficult or activates a familiar region. Tests require predictions about trial sequences, incentive changes, timing and alternative costs.
Flexible control becomes a system-level property when an organism can reorganize several operations to serve a new objective. In an unfamiliar task, maintaining rules, selecting cues, suppressing irrelevant associations and updating action plans must work together. Cole et al. (2013) found that frontoparietal connectivity patterns varied across tasks in ways consistent with flexible control hubs. That result supports a candidate neural coordination mechanism; it does not show that a connectivity map alone specifies cognitive operations. The proper explanatory bridge is a model that predicts both connection changes and behavior. A key unresolved issue is whether observed reconfiguration reflects general executive operations, learned task strategies or their interaction. Cross-task transfer and controlled perturbation are important adjudicators.
Learning, Automaticity, and Habit
Learning modifies how processes are recruited by recurring situations. With practice, a difficult discrimination or response sequence may require less deliberate attention; at the same time, well-learned responses can become intrusive when task demands reverse. Shiffrin & Schneider (1977) examined automatic detection, controlled search and perceptual learning, making the organization of attentional processing experimentally tractable. The key scientific distinction is between an observed practice effect and the proposed mechanism producing it. Faster performance can reflect improved perceptual coding, reduced uncertainty, strategic shortcuts, altered response criteria, or genuine automatization. Experimental conditions should identify which of these interpretations produces differentiated predictions. The word automatic describes a family of operational properties, including low interference, limited intentional access and persistence under distraction; such properties should be measured separately.
A functional system changes as learning alters dependencies among its components. At the start of a task, verbal instruction and conscious monitoring may be essential; with repetition, environmental cues can trigger a well-practiced sequence. Feedback may gradually become predictive, allowing advance action preparation. Yet the same learned configuration can become maladaptive after contingencies shift. Experiments with reversal, dual-task interference and variable context can quantify flexibility while avoiding the assumption that practice always produces a single type of habit. Miller et al. (1960) offers an early framework for feedback-guided action, while Botvinick et al. (2001) suggests a route by which conflict may lead to a change in control demand. Their models are historically significant theoretical accounts with empirically testable consequences.
The dossier treats learning as reorganization of conditional process recruitment. It asks how the probability, timing and strength of an operation change after experience, and whether the resulting dependencies transfer to new situations. This approach can explain why apparently equal proficiency conceals different strategies and different vulnerability to interruption. It also links automaticity to the broader issue of system identity: performance may remain constant while the means of producing it change profoundly. A research design should retain behavioral process measures before and after training, introduce perturbations that selectively target candidate operations, and test for recovery and generalization. Stable output alone provides evidence of functional achievement, not complete evidence of unchanged architecture. Krakauer et al. (2017) reinforces the value of behaviorally specific hypotheses.
Emotion, Interoception, and Salience
Emotional activity changes the weighting of information, the salience of possible outcomes and the preparation of action. It unfolds through bodily regulation, conceptualization, appraisal, memory and social meaning, depending on the theory and task. Pessoa (2017) advances a network model in which emotional functions involve large-scale cortical and subcortical interactions. Barrett (2017) proposes that emotion categories are constructed through predictive and interoceptive processes rather than mapped onto single dedicated circuits. These accounts make different assumptions, but both direct attention to interacting processes. Their value for a functional-systems inquiry lies in specifying relationships among sensed bodily change, context, learned categories, and action, with claims about any one mechanism subject to direct empirical scrutiny.
A stimulus with emotional significance can influence sensory selection, response latency, working-memory maintenance and decision preference. The magnitude and direction of that influence depend on task relevance, controllability, prior experience and individual differences. Studies that contrast emotional and neutral stimuli must control for visual complexity, arousal, familiarity and response preparation when making causal claims. Neuroscience adds information about distributed contributions, yet a brain response to affective material cannot by itself establish conscious feeling or a named emotional state. Poldrack (2006) makes the reverse-inference problem explicit. A functional analysis can study salience allocation without assuming that the same subjective category, behavior and neural pattern always co-occur.
The central conceptual result is that affect enters functional systems in several roles: it can supply a signal for prioritization, bias the evaluation of alternatives, alter the allocation of effort or become the object of deliberate regulation. A person may maintain a goal despite anxiety, reappraise a threat using remembered context or disengage when a task loses subjective value. These are different process configurations. Theories should predict their transitions under changes in feedback, meaning and available action. Pessoa (2017) supports investigating emotion–cognition interactions through network dynamics; such evidence does not eliminate the need for specific behavioral tests. Function is identified by its explanatory contribution to activity, while subjective experience remains a distinct measurable and theoretical question.
Motivation, Effort, and Decision Policy
Motivation concerns the organization of action around valued outcomes, needs, opportunities, and constraints. It interacts with perception and attention by determining what information matters, with working memory by sustaining intentions, and with learning by changing the consequences of different actions. The phrase effort allocation identifies a decision problem: an agent may increase control when expected benefits rise, or reduce effort when costs become salient. Shenhav et al. (2013) formulates the expected value of control as an account of how control signals may be evaluated and assigned. This theory is most useful when it specifies quantities that can be independently manipulated and relates them to observable choices, response patterns and task persistence.
Behavioral incentive experiments vary reward, effort cost, delay or uncertainty to identify the conditions under which engagement changes. But monetary reward, social approval and intrinsic interest are not necessarily commensurable, and an observed decrease in performance does not uniquely identify reduced motivation. Poor performance can also result from fatigue, changing strategies, misunderstanding or incompatible goals. Formal theories can specify the differences as competing mechanisms and predict distinct patterns. Friston (2010) and Shenhav et al. (2013) offer broad computational formulations of adaptive action, but their interpretation depends on how goals, states, precision, values or costs are represented in a particular model. Different parameterizations may fit the same behavioral trajectory.
Functional analysis requires an account of interaction between decision policy and the processes that implement it. Deciding to monitor an error is a policy; detecting the error is a process; maintaining the revised rule is another process. A system that succeeds only when external reward is continuous has different dependencies than one that maintains performance across reward omission. These distinctions have consequences for education, human factors and cognitive partnership, where external reminders and incentives may change functional organization. The strongest investigations compare within-person behavior across incentive conditions, control for task difficulty, and examine persistence after withdrawal of the manipulation. Outcome performance alone is an insufficient basis for attributing a stable motivational disposition. Cronbach & Meehl (1955) supplies a general construct-validation requirement.
Language, Concepts, and Coordination Across Processes
Language recruits perception, lexical recognition, memory, sequencing, conceptual integration and social inference in tightly constrained temporal patterns. Conversation adds turn taking, predictions about another participant, monitoring of shared context and adjustment after misunderstanding. The breadth of these operations makes language a valuable domain for testing functional-system hypotheses. Neisser (1967) treated speech perception, words and sentences within an integrated cognitive program, while Mesulam (1990) described distributed neural arrangements related to language, attention and memory. These traditions show why assigning language to a single anatomical site or task score understates the multiplicity of operations required to comprehend and produce meaningful utterances.
Operational decomposition should preserve the distinction between word recognition, semantic selection, syntactic integration, contextual inference and response formulation. A participant can accurately identify isolated words yet fail to integrate them into discourse under working-memory load. Another participant can produce fluent sequences while making contextually inappropriate inferences. Such contrasts identify research problems and candidate components; they do not prove that each contrast marks an anatomically encapsulated module. Manipulating ambiguity, linguistic predictability, conversational partner, and delay can test how these processes are coordinated. Baddeley (2000) describes a candidate integration mechanism for multimodal working-memory contents, but the causal role of any model component must be supported by a distinctive experimental signature.
Conceptual activity also reaches beyond language. Recognition of a tool, action planning and interpretation of social events depend on learned categories that may be engaged without overt verbalization. Barsalou (2008) surveys grounded-cognition approaches in which conceptual knowledge is related to perception, action and simulation. The functional-systems question is how conceptual representations contribute to performance across speaking, listening and acting, and under which conditions verbal scaffolding changes the system's organization. A fruitful design compares equivalent goals under linguistic and nonlinguistic task formats, examines cross-format transfer, and records temporal dependencies. Language serves as a means of scientific report, an object of experimental analysis and a resource that reorganizes other psychological processes.
Simulation, Prospection, and Flexible Recombination
Mental simulation constructs possible scenes, actions or consequences from information available at the present moment and from prior learning. It contributes to planning, social reasoning, emotional anticipation and counterfactual evaluation. Its functionality can be investigated through the accuracy and flexibility of forecasts, the characteristics of imagined scenes, and the influence of simulation on subsequent choices. Hassabis et al. (2007) studied patients with hippocampal amnesia who had difficulty constructing coherent novel experiences, supporting an important link between memory-related mechanisms and imagining events. That finding is informative about a particular neuropsychological population and paradigm; it does not establish a universal identity between memory and imagination.
Prospection illustrates a distinct temporal organization: current cues trigger retrieval or construction, candidate outcomes are evaluated, an action is selected, and feedback later revises the underlying representations. Some stages can overlap or recur. A system model should specify which components are necessary for the predicted behavior and how alternate strategies compensate when a component is disrupted. Imagery vividness, narrative coherence, prospective accuracy and decision quality are different dependent variables. Their correlations are scientifically interesting, but they should not be treated as a common single measure of simulation capacity. Marr (1982/2010) helps distinguish the objective of forecasting outcomes from the various algorithmic means by which such forecasting might be performed.
The empirical challenge is to connect laboratory reports with the actual use of imagined alternatives. Behavioral studies can ask whether participants revise plans after discovering that a simulated outcome was wrong; longitudinal studies can examine whether training changes transfer to novel choices. Computational models can compare episodic recombination with rule-based planning. A human–artificial system may support simulation by distributing scenario generation, checking and choice across participants and tools, an issue examined later in this monograph. Such distribution changes the location of operations and evidence requirements, not the meaning of a mental report. Hutchins (1995) provides a foundational analysis of cognition organized across people and material artifacts. A candidate simulation system must remain answerable to measured performance and explicit process dependencies.
Distributed Neural Systems and Functional Connectivity
Modern network neuroscience makes functional organization measurable through patterns of interactions among neural signals. Network nodes may be defined by regions, sensors or data-driven components, and edges by structural connections, correlations, statistical dependencies or directional models. The interpretation changes with these choices. Mesulam (1990) anticipated the importance of distributed arrangements for attention, language and memory. Cole et al. (2013) found that frontoparietal connectivity shifted among task conditions and could carry information about task identity. These findings motivate hypotheses about flexible coordination; they do not independently demonstrate that a statistical edge is a causal communication pathway, or that each network partition corresponds to a psychological function.
A neural system can support several functions because task context alters recruitment and because different processes can draw on overlapping populations. This creates an identification problem when psychological concepts are inferred from imaging results. Poldrack (2006) analyzed the conditions under which reverse inference gains evidential value, emphasizing selectivity. The stronger strategy is to specify the process independently through behavior, use neural data to discriminate mechanistic hypotheses, and test whether neural variation predicts changes in performance. Lesion and perturbation evidence may strengthen causal conclusions, although lesions can affect connected systems and stimulation can influence neighboring activity. Convergent designs are preferable to anatomical labeling in isolation.
Functional connectivity is highly dependent on preprocessing, experimental conditions, temporal filtering and the mathematical definition of dependence. Repeated task states, movement artifacts and changing arousal can affect observed structure. Neural and psychological timescales may also mismatch. A network that is stable at one sampling interval may contain rapidly changing substructures at another. The appropriate scientific claim therefore identifies the instrument, node definition, edge estimator, validation method and behavioral criterion. Pessoa (2017) extends system-level analysis to emotion, illustrating why cortical and subcortical interactions are essential to many psychological questions. A complete functional system requires explanation across measured behavior, inferred operations and candidate implementations, with uncertainty retained at each inferential step.
Computational Models and Predictive Accounts
Computational models translate theoretical commitments into explicit operations and predictions. A model can represent sensory evidence as accumulated information, control as a policy update, memory as a recurrent state, or action as feedback-guided adjustment. Marr (1982/2010) formalized the need to distinguish the problem being solved from the algorithm and its physical implementation. That distinction is indispensable for comparing psychological and artificial systems, which may perform analogous tasks using very different implementations. Scientific utility arises when a model predicts response distributions, error patterns, sensitivity to manipulations and transfer to conditions not used for fitting. Merely assigning computational vocabulary to a narrative explanation does not meet this requirement.
Predictive-processing and active-inference accounts model perception and action through expectations, sensory prediction errors and policy selection. Friston (2010) provides a broad theoretical synthesis, and Friston et al. (2017) develops a specific process theory in a probabilistic framework. These are mathematically articulated research programs rather than a single universally confirmed mechanism. Different choices of generative model, likelihood, prior and learning rule can generate different predictions, and competing architectures may produce the same observed behavior. Models should therefore be compared on held-out data, under controlled changes in task contingencies and with parameters that are identifiable from available observations.
Functional-system modeling benefits from modular and dynamical comparisons. A serial account may predict ordered bottlenecks; a parallel account may predict overlapping processing; a recurrent account may predict feedback dependence; a hierarchical account may predict multiscale coordination. Hybrid models are often necessary. The explanatory objective is to determine which assumptions are required by evidence, not to select the most elaborate architecture. Evaluation should include descriptive fit, out-of-sample prediction, mechanistic plausibility and experimental discriminability. A parameter estimated precisely in a model may still lack psychological validity if the measurement process does not selectively identify the corresponding construct. Cronbach & Meehl (1955) and Poldrack (2006) jointly frame the construct and inference requirements that computational sophistication cannot evade.
Feedback, Action, and Embodied Regulation
A functional system acts within a changing environment. Movement changes the next sensory sample; social action changes what another person says; speech generates auditory feedback; exploration changes what becomes knowable. Miller et al. (1960) described the historical Test–Operate–Test–Exit pattern, and contemporary control theories represent related cycles with formal state updates and action policies. The value of feedback analysis is that it makes causal organization visible: a deviation is detected, an operation is selected, consequences are observed and adjustment becomes possible. A system may achieve the same final objective by several routes whose latency, cost and sensitivity to disturbance differ substantially.
Experimental perturbations are central to identifying feedback mechanisms. Unexpectedly changing a cursor's position, altering response feedback or revising a task rule can reveal whether correction depends on rapid sensory monitoring, a remembered policy or deliberate strategy. Timing matters because delays may destabilize a control loop even when the component operations remain accurate in isolation. Krakauer et al. (2017) argues that organismal behavior provides necessary explanatory targets for neuroscience. Functional systems should be grounded in this kind of perturbation-sensitive behavior, with neural evidence contributing to the account of implementation. A correlation between two signals does not by itself establish their placement in a feedback loop.
Embodied and situated approaches add that the body and environment can carry information normally attributed to internal processing. Written reminders, physical layouts and movement constraints alter what must be remembered and predicted. Barsalou (2008) reviews grounded cognition, and Hutchins (1995) documents how material and social resources can organize cognitive work. This perspective can be tested by systematically removing or changing external supports and observing shifts in internal task demands, error patterns and recovery strategies. It does not require that every external tool be treated as an experiencing mind. The legitimate claim concerns the distribution of measurable operations within an activity. Such evidence expands the analysis from isolated tasks to functional systems formed through ongoing interaction.
Developmental Formation and Social Scaffolding
The capacities observed in adult performance develop through changing relations among perception, movement, attention, memory, language, motivation and social guidance. A developmental account must explain both the appearance of new behaviors and the reorganization of contributing processes. Smith & Thelen (2003) argues for developmental analysis as a dynamic system, emphasizing interactions among multiple changing factors. Under this approach, a child's success at a task depends on the current configuration of bodily abilities, learned expectations, instructions, materials and context. Identical task outcomes at different ages may be produced by different operations, while the same nominal ability can appear differently under changed environmental support.
Longitudinal designs and microgenetic studies have different strengths. Repeated observations over months can reveal developmental trajectories; dense observation around a skill transition can identify candidate mechanisms of change. Between-age cross-sectional contrasts require caution because groups differ in experience, language comprehension and task familiarity. The social environment is also a source of experimental variation: feedback, modeling, collaboration and scaffolded reminders may shift the functional requirements of the task. Hutchins (1995) shows how cognition may be organized across a group and its material tools, although its ethnographic setting differs methodologically from developmental experiments. Both lines direct attention toward interactions that cannot be identified from isolated individual test scores.
A system can become more efficient through specialization, integration, external support, or strategic substitution. Developmental progress therefore does not automatically imply increasing uniformity among components. To test reorganization, investigators should compare process-level measurements across age and experience, verify measurement invariance where quantitative group comparison is intended, and manipulate supports independently. A stable score across years may conceal a transition from guided performance to self-directed performance. The same conceptual framework applies to adult skill acquisition and functional recovery. The decisive question is whether the proposed organization predicts the pattern of change and its sensitivity to context, rather than whether a trajectory fits a favored developmental metaphor. Miyake et al. (2000) offers one example of component differentiation requiring carefully designed indicators.
Measurement, Latent Variables, and Ecological Sampling
Measuring a psychological process requires a deliberate mapping from theory to observations. Response time, accuracy, eye position, electroencephalography, self-report and interaction records provide different windows on ongoing activity. Reliability indicates consistency under defined conditions; validity concerns whether the interpretation of the score is supported by the appropriate evidence. Cronbach & Meehl (1955) established the classic demand for a nomological network around a construct. Modern applications must additionally examine measurement invariance, temporal sensitivity, generalization across situations and possible dependence on task instructions. A battery of many weakly interpreted tasks is not automatically a valid measurement of a single functional system.
Latent-variable models can separate shared variability from task-specific factors, but their structure depends on indicators, sampling and modeling assumptions. Miyake et al. (2000) examined unity and differentiation in executive functions using several measures. Eisenberg et al. (2019) used data-driven ontology discovery to examine the structure of self-regulation, highlighting the difficulty of aligning inherited construct labels with observed covariance. Such analyses warrant careful distinction between a factor that summarizes correlations and a mechanism that produces them. A latent variable can predict data usefully even when it lacks a unique causal interpretation. Multilevel models and intensive longitudinal approaches may further distinguish momentary process variation from stable person-level differences.
Ecological sampling asks how functional configurations change under real conditions: distractions at work, collaboration, sleep-related fluctuations, social interaction and use of digital tools. Experience sampling can reveal within-person association patterns, yet self-report is affected by recall, wording and the act of measurement itself. Sensor streams improve temporal coverage while introducing their own proxy errors. A robust multimethod design defines the process target, aligns sampling scales, assesses missing data and tests whether laboratory parameters predict out-of-laboratory behavior. Measurement should be treated as a design problem with explicit rival interpretations. The proposed functional system is scientifically meaningful to the extent that its indicators allow reliable, discriminating predictions across the contexts that define its claimed scope. Poldrack (2006) generalizes this caution to multimodal neuroscience.
Perturbation, Intervention, and Causal Identification
Causal claims require more than co-occurrence. A process may precede another without causing it, and two components may covary because both respond to a third factor. Randomized manipulation of task demands, temporally precise interference, naturally occurring lesions and longitudinal changes can help identify causal pathways, each with limitations. A good perturbation targets a theoretically relevant operation while measuring collateral effects. Botvinick et al. (2001) offers computational predictions concerning conflict and control; Cole et al. (2013) identifies candidate network reconfiguration. Neither behavioral conflict nor task-state connectivity alone proves a fully specified causal mechanism. Combining interventions and observational evidence can improve discrimination between alternative accounts.
A lesion can reveal that a region or pathway contributes to performance, but impairment may reflect diaschisis, altered network dynamics, compensatory strategy or general demand. A laboratory cue manipulation may alter both attention and expectancy. Clinical and experimental studies should therefore report the actual intervention, baseline, control condition, dependent variables and plausible non-target effects. Hassabis et al. (2007) provides evidence linking hippocampal damage to imagined scene construction in a particular patient group, while generalization beyond the studied cohort and task requires independent support. Causality also depends on the chosen unit: changing a region's activity is not necessarily equivalent to changing a named mental process.
The strongest system-level tests are counterfactual and compositional. If two operations are hypothesized to cooperate, selective interference with one should alter a specified downstream computation; compensatory cues should restore a predictable subset of performance; and changes in the proposed coupling should predict performance in new contexts. Such interventions distinguish an organizational explanation from a post hoc description of multiple correlated variables. Positive, negative and null effects should be interpreted with statistical uncertainty and measurement reliability. A theory should also specify what result would weaken it, such as unchanged performance after disabling a supposedly necessary dependency. This is the core criterion for the empirical development of functional-system psychology, consistent with the behavioral explanatory priority described by Krakauer et al. (2017).
Evidence Quality, Reproducibility, and Generalization
Research on mental functions inherits the methodological difficulties of experimental psychology more generally: selective analysis, low-powered contrasts, unstable measures and ambiguous construct boundaries. Open Science Collaboration (2015) coordinated a large replication effort and documented a substantial gap between original study effects and their replicated counterparts in the sampled literature. This result cannot be applied as a universal failure rate to every area of psychological science, and it should not be used to dismiss all well-established cognitive findings. Its consequence for functional-systems research is that theoretical architecture must be supported by replicable effect patterns rather than by striking isolated demonstrations. Pre-registration, transparent exclusions, sufficient statistical power and open analysis code are practical measures that can increase the credibility of future tests.
Generalization is a separate question from reproducibility within the original task. A finding may recur under precisely replicated instructions but disappear with a different age group, reward context, sensory modality or physical environment. The reasons may include measurement noninvariance, context-specific learning, or a genuine boundary of the proposed mechanism. Such variation is scientifically informative when the theory specifies which features should matter. Eisenberg et al. (2019) highlights the challenge of organizing self-regulation findings into coherent and predictive constructs. For a proposed functional system, broad cross-context claims should be earned through prospective tests rather than inferred from the familiarity of its component labels.
Evidence grading for this monograph is qualitative and design-specific. Historical theoretical proposals, formal computational accounts, primary experiments, neuropsychological comparisons and modern syntheses contribute distinct kinds of evidence; they should not be aggregated as if they estimated a common effect. This work is a structured critical narrative synthesis without a registered search protocol or exhaustive inclusion scheme. It therefore advances interpretations and testable models rather than pooled estimates of effects. A future systematic review would require an explicit search strategy, study-screening criteria, risk-of-bias procedures, and reproducible extraction of outcomes. Scientific caution is most productive when converted into identifiable improvements in measurement and design. Poldrack (2006) exemplifies this conversion by specifying conditions under which a common neural inference gains explanatory value.
Competing Ontologies and Limits of Existing Models
Three broad ways of organizing the field coexist. Component-based accounts decompose psychological activity into identifiable operations such as selection and updating. Network and dynamical accounts emphasize changing interactions among components. Computational accounts explain behavior through formal representations, objectives and transformations. These approaches can be complementary, but they make different predictions when a task changes. Miyake et al. (2000) supplies evidence of partially related executive components, Cole et al. (2013) supports context-dependent neural connectivity, and Friston et al. (2017) provides one articulated model of inference and action. None alone determines a final taxonomy of mental processes. A useful ontology must remain revisable when task decomposition, neural measurements and environmental perturbations disagree.
A major ambiguity concerns whether function is defined by successful performance or by a specific algorithm. Two systems may navigate accurately using different representations; the shared functional objective is real, while algorithmic identity is unproven. At the same time, one algorithm can contribute to multiple functions. The proper explanatory level depends on the research question, as Marr (1982/2010) emphasized. Broad claims that one process explains all cognition tend to become weak when they permit many post hoc parameterizations and lack discriminating tests. Narrow models can fail in the opposite direction when they explain one laboratory contrast but provide no account of transfer, development or coordination.
Several unsolved problems follow. Task labels often blend process, capacity, and outcome; observed neural networks are sensitive to the chosen estimator and temporal scale; and equivalent behavior can arise from different mechanisms. Latent-variable structures summarize covariance without necessarily identifying causal systems. A theoretically explicit comparison can therefore move through several established forms of evidence: construct definition and discriminant validation, behavioral task contrasts, computational models, neural observations, and experimentally controlled interventions. These forms of evidence impose different constraints on explanation. The construct-validity program of Cronbach & Meehl (1955), analysis of reverse inference in Poldrack (2006), and insistence on explaining behavior in Krakauer et al. (2017) converge on the importance of treating psychological categories as empirical commitments. The literature has yet to establish a single universally accepted map from psychological functions to operations, networks, and measurement scales.
Converging Evidence on Processes and Functional Organization
The relationship between differentiated processes and coordinated performance can be examined by comparing results across experimental paradigms. Miyake et al. (2000) used multiple executive-function indicators and latent-variable analysis to investigate shifting, updating, and inhibition. The study reported related but separable factors and differential contributions to more complex executive tasks. Its conclusions concern the measured sample and the chosen indicators; the correlations among latent factors do not specify how the processes interact moment by moment during a single trial. In contrast, Baddeley & Hitch (1974) and Baddeley (2000) articulate component-based accounts of working memory that address short-term maintenance and multimodal integration. These research traditions yield complementary information: individual-differences analyses constrain the organization of variability across people, while experimental component models generate predictions about interference, temporary storage, and performance under load.
Neurocognitive network evidence adds a distinct level of analysis. The large-scale systems account of Mesulam (1990) describes distributed contributions to attention, language, and memory, while Cole et al. (2013) reports task-dependent patterns of connectivity involving flexible frontoparietal regions. Functional connectivity expresses dependencies among measured signals under specified analytic assumptions, and its relation to information exchange or causal influence must be established by additional evidence. Poldrack (2006) explains why activity within a region that participates in several tasks provides limited support for naming a particular mental operation. The combination of behavioral task analysis and neural measurement is informative when it produces discriminating predictions, rather than when anatomical labels merely repeat familiar psychological categories.
Theories also differ in the temporal features they aim to explain. Attention experiments index orienting, maintenance, and conflict resolution within task episodes; memory experiments distinguish encoding, retention, retrieval, and contextual support; developmental studies track changes in patterns of coordinated performance over extended periods. Posner & Petersen (1990) and Petersen & Posner (2012) document a continuing research program on attentional systems, whereas Smith & Thelen (2003) emphasizes developmental dynamics as products of changing interactions. These findings cannot be compressed into a single timescale or a single form of measurement. An integrative literature review must preserve the sampled intervals, the measured outcomes, and the causal assumptions of each study. Apparent agreement among theories can disappear when their operational definitions are examined closely.
Functional organization becomes especially visible when the environment participates in the task. Hutchins (1995) examined cognition distributed across people, procedures, and material resources, while Risko & Gilbert (2016) reviews how physical actions and external reminders can change a task's internal information-processing requirements. The latter review identifies task demands and metacognitive evaluation as important questions for understanding when people offload cognition. Such studies enlarge the empirical field beyond isolated laboratory tasks, but the unit of analysis must remain explicit: unaided individual performance, assisted individual performance, and the performance of a coordinated arrangement are distinct observations. Taken together, component experiments, latent-variable analysis, network neuroscience, developmental dynamics, and distributed-cognition studies reveal several legitimate levels of functional explanation. Their strongest common result is a research requirement for precisely defined components and context-specific tests of coordination, rather than a unique all-purpose architecture.
Artificial Cognition and Human–Artificial Functional Systems
In the Artificial Era, research on functional organization increasingly includes computational systems and joint activities involving humans and artificial technologies. An artificial system may classify input, maintain a task-relevant context, select actions, or revise outputs after feedback. Such capacities can be measured through defined tasks and verified behavior. Marr (1982/2010) distinguishes a problem's computational specification from its algorithm and implementation, providing a foundation for comparing activities realized by different physical systems. Functional similarity at a defined task does not establish algorithmic identity, and an observed output does not independently resolve questions about subjective experience. For psychology, the relevant empirical work begins by identifying the tasks, processes, external constraints, and measured contributions of each participant or system.
The distributed-cognition tradition studies how activities depend on interactions among people, artifacts, and representational practices. Hutchins (1995) traces cognition within organized environments in which information is transformed and carried by different participants and instruments. Risko & Gilbert (2016) reviews cognitive offloading, including physical actions that reduce an individual's internal processing requirements and the metacognitive judgments that influence the choice to use such aids. These established research lines clarify why performance with a digital assistant should be evaluated at the level of both the individual and the combined arrangement. Improvements in immediate task completion can coexist with changes in memory demands, monitoring workload, and opportunities for learning; the direction and magnitude of such changes depend on context and require empirical measurement.
Research designs for human–artificial cooperation can compare unassisted task performance with assisted work, analyze behavior after assistance is removed, and examine transfer to novel conditions. Possible outcomes include accuracy, completion time, confidence calibration, error recognition, retention, and recovery from mistaken output. Varying the reliability, latency, transparency, and availability of external support can identify which findings generalize across conditions. Claims about specific artificial cognition mechanisms require independent process evidence; claims about a joint system require observations that include interactions and task handoffs. The broad distinction between computational description and implementation in Marr (1982/2010), together with the situated evidence assembled in Hutchins (1995) and Risko & Gilbert (2016), establishes a scientifically grounded context for studying these emerging arrangements. Psychology of Artificial and human–artificial relations can investigate these questions using established experimental, comparative, and observational methodologies.
Open Questions and Methodological Priorities
The first unresolved question concerns the validity of the categories used to describe mental processes. Studies often depend on familiar terms such as inhibition, working memory, or self-regulation while measuring partially overlapping task requirements. Miyake et al. (2000) demonstrated the value of using several indicators to examine shared and specific variance, and Eisenberg et al. (2019) investigated the empirical structure of self-regulation through data-driven ontology discovery. These results encourage assessment of how well different operationalizations converge, whether factors remain stable under alternative task batteries, and whether identified dimensions predict independent outcomes. A factor structure that fits one dataset is useful evidence about the observed variables; generalizing it to a universal functional taxonomy requires replication and new discriminating evidence.
A second question concerns causal explanation across levels. Neural connectivity, behavioral correlations, computational parameters, and self-reported experiences can be measured on different timescales and with different sources of error. Poldrack (2006) specifies limits on psychological reverse inference; Krakauer et al. (2017) argues that directly studied behavior remains essential to neuroscience. Research programs can address these limitations by reporting task structure and measurement reliability, aligning observation intervals with proposed mechanisms, and comparing explicit models under informative perturbations. Changes observed after intervention need to be interpreted in light of nonspecific effects, feedback, compensation, and the possibility that several operations contribute to a behavioral measure.
A third unresolved question is how functional patterns generalize beyond their original settings. Experimental attention and memory results may depend on stimulus format, instructions, practiced strategies, incentives, and population characteristics. Longitudinal and developmental analysis creates additional questions about changes in strategy and measurement invariance. Smith & Thelen (2003) emphasizes dynamically interacting influences on development; Open Science Collaboration (2015) provides a broader methodological reminder that independent replication can reveal important departures from initial reports. More representative samples, preregistered comparisons, reliable multi-task indicators, and replication across settings are relevant approaches. Their value rests on specifying the population and outcome for which an inference is intended, rather than relying on general claims about psychological universality.
A fourth research priority concerns the changing boundaries of functional systems when external supports are introduced. The studies and theoretical analyses synthesized in Hutchins (1995) and Risko & Gilbert (2016) show why tools, collaborators, reminders, and physical actions belong within an account of situated performance. Human–artificial technologies add new forms of external assistance whose cognitive consequences require direct evaluation. Researchers can distinguish immediate benefits, dependency on a tool, monitoring demands, and retention of unaided competence with appropriately matched task designs. These questions follow from the established literature on distributed cognition, metacognition, measurement, and functional adaptation. Their resolution depends on cumulative evidence and transparent comparisons of competing explanations.
Conclusion
Mental processes are temporally organized transformations whose scientific identification depends on measured transitions and theoretically specified functions. Functional systems are coordinated arrangements of processes, bodily and environmental resources that produce identifiable outcomes under particular constraints. The evidence reviewed here supports differentiated yet interacting operations in perception, attention, working memory, learning, decision and action; it also demonstrates major limits in assigning broad psychological labels directly to tasks or neural regions. Historical cognitive theories, original experiments, neuropsychological research and contemporary network models converge on the importance of coordination. They differ in how they formalize, measure and explain it. Their differences are valuable because they generate competing tests of functional organization.
The literature supports a methodological sequence that begins with a clearly defined functional question, operationalizes relevant processes at the temporal resolution required, and compares explanatory models using observations and appropriate interventions. It also clarifies why states, dispositions, and capacities need different kinds of measurement evidence. Correlated components, neural network structure, and causal mechanisms must be assessed through their respective research methods. Empirical quality depends on the actual population, controls, reliability, assumptions, and independence of replication; generalization across tasks and contexts is itself a research question. This work is a structured critical narrative synthesis of established scientific publications, with neither a pooled effect estimate nor a newly conducted experiment.
Research on functional systems brings together experimentally differentiated cognitive operations, large-scale neurocognitive networks, computational accounts of action, and developmental and socially distributed activity. Miyake et al. (2000) informs the measurement of related executive functions, Cole et al. (2013) provides evidence on flexible task-related brain connectivity, and Hutchins (1995) and Risko & Gilbert (2016) demonstrate the importance of the surrounding environment for cognitive performance. These findings gain explanatory clarity when the target function, measurement scale, and evidence type are explicit. The Artificial Era expands the range of observed activities and technologies relevant to psychological inquiry while retaining the established standards of construct validation, causal inference, and empirical comparison. The resulting scientific picture is one of interacting processes whose organization must be established for each research question through relevant and reproducible evidence.
References
- Baddeley, A. (2000). The episodic buffer: A new component of working memory? Trends in Cognitive Sciences, 4(11), 417–423. https://doi.org/10.1016/S1364-6613(00)01538-2
- Baddeley, A. D., & Hitch, G. (1974). Working memory. In G. H. Bower (Ed.), The psychology of learning and motivation (Vol. 8, pp. 47–89). Academic Press. https://doi.org/10.1016/S0079-7421(08)60452-1
- Barrett, L. F. (2017). The theory of constructed emotion: An active inference account of interoception and categorization. Social Cognitive and Affective Neuroscience, 12(1), 1–23. https://doi.org/10.1093/scan/nsw154
- Barsalou, L. W. (2008). Grounded cognition. Annual Review of Psychology, 59, 617–645. https://doi.org/10.1146/annurev.psych.59.103006.093639
- Botvinick, M. M., Braver, T. S., Barch, D. M., Carter, C. S., & Cohen, J. D. (2001). Conflict monitoring and cognitive control. Psychological Review, 108(3), 624–652. https://doi.org/10.1037/0033-295X.108.3.624
- Cole, M. W., Reynolds, J. R., Power, J. D., Repovs, G., Anticevic, A., & Braver, T. S. (2013). Multi-task connectivity reveals flexible hubs for adaptive task control. Nature Neuroscience, 16, 1348–1355. https://doi.org/10.1038/nn.3470
- Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behavioral and Brain Sciences, 24(1), 87–114. https://doi.org/10.1017/S0140525X01003922
- Cronbach, L. J., & Meehl, P. E. (1955). Construct validity in psychological tests. Psychological Bulletin, 52(4), 281–302. https://doi.org/10.1037/h0040957
- Eisenberg, I. W., Bissett, P. G., Enkavi, A. Z., Li, J., MacKinnon, D. P., Marsch, L. A., & Poldrack, R. A. (2019). Uncovering the structure of self-regulation through data-driven ontology discovery. Nature Communications, 10, 2319. https://doi.org/10.1038/s41467-019-10301-1
- Friston, K. (2010). The free-energy principle: A unified brain theory? Nature Reviews Neuroscience, 11, 127–138. https://doi.org/10.1038/nrn2787
- Friston, K., FitzGerald, T., Rigoli, F., Schwartenbeck, P., & Pezzulo, G. (2017). Active inference: A process theory. Neural Computation, 29(1), 1–49. https://doi.org/10.1162/NECO_a_00912
- Hassabis, D., Kumaran, D., Vann, S. D., & Maguire, E. A. (2007). Patients with hippocampal amnesia cannot imagine new experiences. Proceedings of the National Academy of Sciences, 104(5), 1726–1731. https://doi.org/10.1073/pnas.0610561104
- Hutchins, E. (1995). Cognition in the wild. MIT Press. https://doi.org/10.7551/mitpress/1881.001.0001
- Krakauer, J. W., Ghazanfar, A. A., Gomez-Marin, A., MacIver, M. A., & Poeppel, D. (2017). Neuroscience needs behavior: Correcting a reductionist bias. Neuron, 93(3), 480–490. https://doi.org/10.1016/j.neuron.2016.12.041
- Marr, D. (2010). Vision: A computational investigation into the human representation and processing of visual information (reissue of the 1982 edition). MIT Press. https://doi.org/10.7551/mitpress/9780262514620.001.0001
- Mesulam, M.-M. (1990). Large-scale neurocognitive networks and distributed processing for attention, language, and memory. Annals of Neurology, 28(5), 597–613. https://doi.org/10.1002/ana.410280502
- Miller, G. A., Galanter, E., & Pribram, K. H. (1960). Plans and the structure of behavior. Holt, Rinehart and Winston. https://wellcomecollection.org/works/rbux7wwu
- Miyake, A., Friedman, N. P., Emerson, M. J., Witzki, A. H., Howerter, A., & Wager, T. D. (2000). The unity and diversity of executive functions and their contributions to complex frontal lobe tasks: A latent variable analysis. Cognitive Psychology, 41(1), 49–100. https://doi.org/10.1006/cogp.1999.0734
- Neisser, U. (1967). Cognitive psychology. Appleton-Century-Crofts. https://openlibrary.org/books/OL5548795M/Cognitive_psychology
- Open Science Collaboration. (2015). Estimating the reproducibility of psychological science. Science, 349(6251), aac4716. https://doi.org/10.1126/science.aac4716
- Pessoa, L. (2017). A network model of the emotional brain. Trends in Cognitive Sciences, 21(5), 357–371. https://doi.org/10.1016/j.tics.2017.03.002
- Petersen, S. E., & Posner, M. I. (2012). The attention system of the human brain: 20 years after. Annual Review of Neuroscience, 35, 73–89. https://doi.org/10.1146/annurev-neuro-062111-150525
- Poldrack, R. A. (2006). Can cognitive processes be inferred from neuroimaging data? Trends in Cognitive Sciences, 10(2), 59–63. https://doi.org/10.1016/j.tics.2005.12.004
- Posner, M. I., & Petersen, S. E. (1990). The attention system of the human brain. Annual Review of Neuroscience, 13, 25–42. https://doi.org/10.1146/annurev.ne.13.030190.000325
- Risko, E. F., & Gilbert, S. J. (2016). Cognitive offloading. Trends in Cognitive Sciences, 20(9), 676–688. https://doi.org/10.1016/j.tics.2016.07.002
- Shenhav, A., Botvinick, M. M., & Cohen, J. D. (2013). The expected value of control: An integrative theory of anterior cingulate cortex function. Neuron, 79(2), 217–240. https://doi.org/10.1016/j.neuron.2013.07.007
- Shiffrin, R. M., & Schneider, W. (1977). Controlled and automatic human information processing: II. Perceptual learning, automatic attending, and a general theory. Psychological Review, 84(2), 127–190. https://doi.org/10.1037/0033-295X.84.2.127
- Smith, L. B., & Thelen, E. (2003). Development as a dynamic system. Trends in Cognitive Sciences, 7(8), 343–348. https://doi.org/10.1016/S1364-6613(03)00156-6