Perception, cognition and emotion in interaction

Every design rule about perception, cognition or emotion rests on a claim about a person. Targets should be large. Menus should be short. Colour should never carry meaning on its own. Alerts should be hard to miss. Each rule is a conclusion from evidence, and the evidence is uneven. Some rest on the physiology of the eye and hold for nearly everyone. Others come from one laboratory study that few have repeated. Read each rule as a claim about the people who were tested. Ask whether the sample was young, sighted, literate and online.

One dispute runs under the whole area: how fixed are these limits? One tradition treats the person as a machine with measurable channels, and working memory as a constant to design around; its evidence is the controlled experiment, which isolates a cause. Another treats a limit as a starting point that practice and tools can move, and studies work as it happens. They disagree over their evidence, not from carelessness. So test a limit with your own users, and treat a laboratory average as a hypothesis.

Why the eye is not a camera

Light carries wavelength, and colour is a construction the eye builds. The retina compares three cone types down opponent channels. A red and a green can look the same to a person whose red-green channel is weak. Contrast is computed separately from hue, and it decides whether text can be read (Ware, 2020). The web accessibility standard sets a floor here: 4.5 to 1 for ordinary text, 3 to 1 at large sizes (W3C, 2023).

Layout rests on weaker work. Ware brings the gestalt laws of grouping into the same argument: they describe how the eye binds marks into one object (Ware, 2020). Those laws predict direction, not size, and Ware offers them as heuristics, so a grouping judgement is a prediction.

Take a dashboard that reports health in colour alone: green for normal, red for fault. For a user with a red-green deficiency the two states collapse into one grey. The same screen thins for an older eye or a bright room, because contrast, not hue, carries the text.

Design rule: never encode meaning in colour alone. Pair hue with shape, position or a written label. But redundancy is not free: a second channel adds marks and can slow the search it was meant to help. The colour evidence is strong and the grouping evidence weak, so treat grouping as a direction to test.

Attention decides what you see

Perception is selective, and the user does not control that selection. Change blindness is the failure to notice a large change when something disrupts the instant it occurs. Inattentional blindness is the failure to notice an object when attention is elsewhere.

In the best known demonstration, people counting passes in a filmed game miss a person in a gorilla suit. Three high-powered experiments with 4,493 participants repeated it and varied the speed of the unexpected object (Wallisch et al., 2023).

Speed decided detection. Fast objects were noticed even while the counting task ran; slow ones were often missed. The fastest object was briefest on screen, so exposure was not the cause; similarity to the tracked objects mattered. What the effect licenses is contested: many studies find it with different scenes, but the detection rate moves with the scene, the task and the object's speed (Wallisch et al., 2023). Whether the miss is a failure of seeing or of holding what was seen is disputed. Design rule: give anything that must not be missed motion, on a channel the primary task does not occupy.

Interruption costs accuracy when tasks share a resource

That selection has a further cost: attention is limited, and the standard text treats it as several resources, not one. Wickens and colleagues give the field its working reference on two tasks at once (Wickens et al., 2022). Tasks sharing a perceptual or response resource interfere; tasks on different ones interfere less.

The multi-resource model is a working framework, not a settled finding: a single shared capacity accounts for much of the same data and makes similar predictions, so the dispute rarely blocks practice.

Interruption is the everyday form: a clinician entering prescriptions while colleagues ask questions carries a second demand on the same resources, and a dropped connection adds the same cost. Items are skipped, and the record shows an empty field, not a wrong one.

Design rule: protect the primary task, and make interrupted work resumable, showing where the user left off and what remains. Interference between tasks that share a resource is reliable (Wickens et al., 2022), but its size in a real setting is not fixed.

Working memory holds about four things

That second limit sits in working memory, the small store holding what you are thinking about. The field quoted seven items for decades, a figure that began as a rough estimate and hardened into a slogan. Cowan's review argues the seven was closer to a rhetorical device than a measured capacity, removes the conditions that inflate the count, and finds a single central limit averaging about four chunks; the limit falls on chunks, not raw items, so recoding raises how much each carries (Cowan, 2001). This is the field's clearest disagreement about a number.

Cowan's article drew much commentary; one line treats capacity as a resource divided among what is held, so the four marks the middle of a shifting limit. Capacity also varies between people, so the four is an average, not a property of every person (Cowan, 2001).

The design consequence runs through recognition over recall. Recognition means choosing from visible options; recall means producing from memory. A menu supports recognition, a command line demands recall, so the menu wins for a novice and the command line for an expert. The gap widens for a reader working in a second language, who must also recall the exact words the system expects. Design rule: keep simultaneous demands to a few items, chunk anything longer, and let people recognise rather than recall.

Expertise rewrites the rules

Practice moves that store, because skill changes what a task demands. Performance improves quickly at first and then more slowly, a pattern the field calls the power law of practice (Wickens et al., 2022). Chase and Simon (1973) showed what practice does to memory. A master and a beginner studied a real position for five seconds, then rebuilt it: the master replaced most of the board, the beginner a few pieces. Three players took part, so the result concerns one game, not general ability.

The gap nearly vanished on a random scatter, so the advantage was chess knowledge rather than memory. An expert is made by hours inside one system, not by age, so a young user meeting a new screen is a novice. Two disputes follow: whether the power law holds across tasks, and whether expertise transfers between systems, which the chess study suggests it does not do within one game.

Design rule: design the path for the first hour and for the thousandth hour, and make the move between them visible. The description of practice effects is robust; the explanation of the curve is open.

A default is a decision to defend

Judgement under uncertainty runs on shortcuts, and they survive training. Tversky and Kahneman described three in 1974. Representativeness judges probability by resemblance. Availability judges frequency by how easily examples come to mind. Adjustment from an anchor moves too little from a given value. Their own summary is careful: the heuristics are usually effective and produce predictable errors.

Interfaces supply anchors and defaults constantly: a preselected option, a prefilled field or a list order all shift a choice. A form that pre-ticks the more expensive plan looks neutral, since the user can change it, and the anchor persuades.

Two disputes surround this. One frame treats a heuristic as a shortcut trading accuracy for effort; a competing programme treats it as an adaptive tool that often beats fuller analysis. The other dispute concerns size: laboratory effects are robust, field effects smaller and context-bound. Design rule: choose defaults deliberately and present proportionate information. The mechanism is credible and the field effect uncertain, so a default is a decision to defend.

Slips and mistakes need different remedies

Errors are ordinary cognition producing the wrong output. Reason draws the central line between two kinds: in a slip or a lapse, the intention was right and the execution failed; in a mistake, the intention itself was wrong. Reason follows those errors into systems, where a small error meets a design that lets it through.

Norman's 1981 paper analyses slips through an action system built from schemas. The first covers slips in forming the intention, such as mode and description errors; the second, faulty activation, such as losing an intention or misordering steps; the third, faulty triggering, such as blends and intrusions of habit.

A nurse selects the wrong patient record from a long list because the target matches a familiar surname. That is a capture slip, not a decision. The same nurse chooses the wrong dose because the guidance was misread, a mistake that more careful clicking would not prevent. The machinery is ordinary, so the remedy belongs in the design, not in a warning to the person.

Safety science supplies a strong counter. If the design permits the error, the account of failure should stop there; naming the operator's error ends the enquiry too early. Design rule: for slips, constrain and forgive with forcing functions, distinct modes and undo. For mistakes, supply better information, feedback and a way to check, and remember that the categories overlap.

Aiming time follows distance and target size

Motor control obeys a stable rule: Fitts's law. Fitts reported three experiments showing that movement time is governed by the information the task demands, and that capacity is roughly constant across distances and widths (Fitts, 1954). Movement time grows with distance and shrinks as the target grows. The web accessibility standard sets a floor of 24 by 24 CSS pixels for a pointer target, with stated exceptions (W3C, 2023).

Reaction time is separate: it depends on the number of alternatives and on how naturally a display matches the response (Wickens et al., 2022). Fitts's own account rests on visual and proprioceptive feedback, so a control that responds late breaks the loop a skilled user relies on.

Take a tightly packed toolbar in a screen corner against one large button near the thumb. The large button is faster to hit, and that follows from the law, not taste. An older hand or one with a tremor wants the same thing: a wider target and less travel. The law's reach is disputed. Its constants come from a stylus on a tablet, not a finger on glass, so a precise time read into a touch target goes beyond the evidence; the law is robust for simple aiming and silent on what belongs on the screen. Design rule: make frequent targets large, put them where the hand already is, and keep the response inside the time the user expects.

Feeling is part of the measurement, and evidence follows

That law describes action rather than feeling. Emotion has two working dimensions: arousal, from calm to excited, and valence, from unpleasant to pleasant, which place a state well enough to design for (Russell, 1980).

That dimensional account is itself disputed. One school treats emotion as a few basic kinds; another as a construction from a bodily state and a setting. The pair is a useful summary, not a settled theory: treat it as a measurement choice, not a fact about the mind. The instrument assumes a reader: Russell scaled English adjectives by self-report, so the measure asks for someone sharing that language.

McCarthy and Wright go further. In Technology as Experience they hold that the felt experience of technology is emotional, intellectual and sensual at once, and that people live with technology rather than use it, so a felt state is part of sense-making, not a decoration. Their cases include an ambulance control room. The account is interpretive, not measured, and its standards of good work are contested.

Experience measures are less standardised than task time, so this is the weakest evidence. Two checkouts can take the same time while one leaves the user anxious. Design rule: measure affect and meaning alongside task success, report a range rather than a single score, and say who took part. Turning these judgements into evidence you have gathered and defended is the work of the module that follows.