Loading field map
Cognitive HCI began with a deceptively simple question: can we treat human-computer interaction as a form of information processing, and if so, what follows for design? Over four decades, five major frameworks have offered competing answers. The earliest treated cognition as a private, internal computation. Later frameworks pushed outward—first to the constraints of the work environment, then to the distributed socio-technical system, and finally to the lived, embodied experience of interaction. None of these frameworks has fully replaced the others. Instead, they have accumulated into a pluralistic landscape where researchers choose a framework based on what they want to explain: the speed of a skilled user, the safety of a nuclear plant, the coordination of a ship's navigation team, or the felt quality of a gesture-controlled interface.
In 1983, Stuart Card, Thomas Moran, and Allen Newell published The Psychology of Human-Computer Interaction, which laid out the Human Information Processing (HIP) framework. Their central move was to treat the human user as an information processor with measurable parameters—a perceptual processor, a cognitive processor, and a motor processor, each with characteristic cycle times. The Model Human Processor (MHP) distilled these parameters into a predictive tool. Alongside it, the GOMS model (Goals, Operators, Methods, Selection rules) allowed designers to estimate task execution times by decomposing a task into its cognitive and motor steps.
HIP was an engineering framework, not merely a descriptive one. Its ambition was to give HCI the same predictive power that physics gives to engineering. By modeling the user as a bounded, rational processor, designers could compare interface alternatives without running full user studies. The framework assumed that cognition happens inside the head, that it can be decomposed into discrete stages, and that the relevant unit of analysis is the individual user performing a well-defined task. This assumption—cognition as internal, sequential, and individual—became the target of every later framework in the subfield.
Just five years after HIP, Jens Rasmussen and Kim Vicente proposed Ecological Interface Design (EID), a framework that shifted attention from internal mental models to the external structure of the work domain. EID drew on Rasmussen's Skills, Rules, Knowledge (SRK) taxonomy, which distinguished three levels of human performance: skill-based (automatic, sensory-motor), rule-based (if-then procedures), and knowledge-based (analytical problem-solving). The key insight was that interfaces should reveal the constraints of the physical system being controlled—the thermodynamics of a power plant, the aerodynamics of an aircraft—so that operators could perceive what to do directly, without having to compute it internally.
EID did not reject HIP's engineering ambition, but it narrowed the scope. Where HIP aimed to predict performance for any task, EID focused on safety-critical, complex domains where operators must handle unanticipated events. Its method was not to model the user's internal processor but to analyze the work domain using an abstraction hierarchy and then map that structure onto the interface. The framework remains active today in process control, aviation, and medical informatics, where its core principle—make the constraints visible—has proven durable. EID coexists with later frameworks by occupying a specialized niche: it is less concerned with general theory of cognition than with domain-specific display design.
Computational Cognitive Modeling (CCM) extended HIP's predictive program by building runnable cognitive architectures. The most influential of these, ACT-R (Adaptive Control of Thought—Rational), developed by John Anderson and later adapted for HCI by researchers such as Wayne Gray and Bonnie John, simulates the cognitive processes of a user moment by moment. Unlike the static equations of the MHP, ACT-R models learning, memory decay, and error. A designer can run a simulation of a user performing a task and observe where the model slows down or makes mistakes.
CCM preserved HIP's core assumptions—cognition is internal, computational, and individual—but deepened them. Where HIP gave back-of-the-envelope estimates, CCM offered dynamic, fine-grained predictions. The framework also absorbed the SRK taxonomy from EID in a limited way: ACT-R models can switch between rule-based and knowledge-based processing, though the ecological emphasis on external constraints is largely absent. CCM remains a living tradition, used today for modeling visual attention in interfaces, predicting learning curves for new software, and evaluating assistive technologies. Its strength is precision; its limitation is that it scales poorly to collaborative or culturally situated tasks.
Edwin Hutchins's Cognition in the Wild (1995) introduced Distributed Cognition (DCog), a framework that fundamentally broke with the internalist tradition. Hutchins studied the navigation team on a U.S. Navy ship and argued that cognition is not confined to individual minds but is distributed across people, artifacts, and the coordination between them. The ship's chart, the bearing log, the verbal call-and-response among crew members—these are not merely aids to individual cognition; they are part of the cognitive system. The unit of analysis shifts from the individual to the socio-technical system.
DCog replaced the laboratory experiment with cognitive ethnography: long-term observation of real-world work practices. Its methods include video analysis, artifact tracing, and interaction analysis. The framework does not reject the idea of internal processing, but it insists that explaining how a task gets done requires looking beyond the skull. This put DCog in direct tension with CCM. Both frameworks study cognition, but they define it differently. For CCM, cognition is what happens inside a single mind; for DCog, cognition is the property of a system that includes minds, tools, and their interactions. The two frameworks do not compete on the same ground—they ask different questions. CCM asks: how fast and accurately can an individual perform this task? DCog asks: how does this system of people and artifacts produce coordinated action?
DCog also challenged EID, though from a different angle. EID had already moved attention outward to the work domain, but it still assumed a single operator perceiving constraints. DCog showed that in many real settings, no single operator has the full picture; cognition is distributed across multiple actors and artifacts. The two frameworks can complement each other: EID provides principles for designing displays that reveal constraints, while DCog provides methods for understanding how those displays are actually used in collaborative practice.
Paul Dourish's Where the Action Is (2001) introduced Embodied Interaction, a framework that drew on phenomenology—particularly the work of Martin Heidegger and Maurice Merleau-Ponty—to argue that cognition is not just distributed but fundamentally embodied. We do not think about the world from a detached, computational standpoint; we act in the world through our bodies, and meaning arises from that embodied engagement. Embodied Interaction rejected the representationalism shared by HIP, CCM, and even DCog: the idea that cognition involves manipulating internal representations of the world.
Embodied Interaction emerged from the same critique of internalism that motivated DCog, but it pushed further. Where DCog still talked about information flow and coordination, Embodied Interaction talked about presence, skill, and the felt quality of action. Its practical implications were most visible in tangible user interfaces, gesture-based interaction, and ubiquitous computing—systems that leverage the body's natural capacities rather than requiring abstract symbolic reasoning.
In the broader HCI discipline, Embodied Interaction has become a recognized sibling subfield with its own frameworks (Movement-Based Design, Somaesthetic Design, Tangible User Interfaces). It is no longer contained within Cognitive HCI. Yet its historical roots lie in the same debate that defines this subfield: where is cognition, and how should we study it? Embodied Interaction answers: cognition is in the body's engagement with the world, and we should study it through phenomenological analysis and design practice, not through computational models or cognitive ethnography.
Today, four of the five frameworks remain active. Each occupies a distinct niche, and their assumptions conflict in ways that matter for research practice.
Computational Cognitive Modeling and Distributed Cognition represent the deepest divide. A researcher who adopts CCM commits to the individual as the unit of analysis, to formal models that can be simulated and validated against reaction times and error rates, and to the goal of predicting performance. A researcher who adopts DCog commits to the socio-technical system as the unit of analysis, to ethnographic methods that capture real-world practice, and to the goal of explaining how coordination happens. These are not complementary methods for the same question; they are different definitions of what cognition is. A CCM study of a pilot's interaction with a cockpit display asks: how fast can this pilot read this instrument? A DCog study of the same cockpit asks: how does the pilot, the co-pilot, the autopilot, and the instrument panel together produce a safe landing?
Ecological Interface Design occupies a middle ground. It shares with CCM a commitment to formal analysis and prediction, but it shares with DCog a focus on the external environment rather than internal processing. EID is less ambitious in scope than either: it does not claim to be a general theory of cognition, only a method for designing interfaces in complex, safety-critical domains. This specialization has kept it alive and productive, but it rarely engages directly with the CCM-DCog debate.
Embodied Interaction, as noted, has branched into its own subfield. Its influence on Cognitive HCI is indirect: it reminds researchers that not all interaction can be captured by information-processing models or ethnographic descriptions of task coordination. The felt, bodily dimension of using a computer matters, and frameworks that ignore it are incomplete.
The leading frameworks today—CCM, DCog, and EID—agree on one fundamental point: HIP's original model of the user as a simple information processor is insufficient. Cognition is not just a sequence of internal stages. They disagree, however, on what to put in its place. CCM keeps the individual mind as the locus of cognition but makes the model more dynamic and realistic. DCog dissolves the boundary between mind and environment, treating cognition as a system-level phenomenon. EID focuses on the structure of the environment itself, arguing that good design makes cognition unnecessary by making action direct. There is no consensus, and there may never be. The choice of framework is a choice about what counts as cognition and what methods can reveal it.
The history of Cognitive HCI is not a story of linear progress from a primitive framework to a sophisticated one. It is a story of accumulation and specialization. Each framework emerged to address a limitation in its predecessors, but none has rendered the others obsolete. A student entering the field today inherits a toolkit of frameworks, each suited to different scales of analysis and design challenges. The question is not which framework is correct, but which framework is useful for the problem at hand.