Measuring Intuitive Use: Theoretical Foundations

Abstract

Intuitive use is a notoriously vague concept. Several research groups have been developing a wealth of definitions and ways of measuring intuitive use that show only few commonalities. Here we review previous approaches combined with newer theoretical developments in psychology. From this review we propose that high effectiveness, low cognitive effort and a strong metacognitive feeling of fluency are the defining characteristics of intuitive use whereas many other measures are typical correlates of these defining characteristics. Distinguishing between defining characteristics and typical correlates allows researchers and practitioners to refer to a common measurement definition of intuitive use while maintaining their flexibility to adapt measures according to their needs.

Keywords:

• Intuitive use
• intuitive interaction
• processing fluency
• mental effort
• two-process theories
• fringe of consciousness

Intuitive use is often seen as a vague and poorly defined concept in human-computer interaction (Naumann et al., 2007). Yet intuitive use is in high demand, fuelled by a need for technology that is easy to operate. As interactive products become more ubiquitous their user base becomes more heterogenous regarding age, cultural background, prior technology experience, as well as sensorimotor and cognitive abilities. Users also find it difficult to recruit the time and resources to learn new functionalities and interaction styles. Further challenges are the growing complexity of products, their interconnectedness and an increasing inevitability of technology in leading ordinary lives (Hurtienne, 2011; Lawry, 2012; Lawry et al., 2019; Reinhardt et al., 2018).

In their attempts of defining what intuition is, researchers are often turning to other disciplines (e.g., Dane & Pratt, 2009; Hodgkinson et al., 2008). Intuition has a tradition of research in education and instructional psychology (e.g., Hogarth, 2001; Simonton, 1980), judgment and decision-making (e.g., Evans & Stanovich, 2013; Gore & Sadler-Smith, 2011; Kahneman & Frederick, 2002; Klein, 1993; Patterson, 2017), creative thinking (e.g., Bastick, 1982, 2003; Bowers et al., 1995; Dorfman et al., 1996; Mednick, 1962) and general cognitive science (e.g., Damasio, 1994; Lieberman, 2000). O’Brien et al. (2010), in their review, list 64 definitions of “intuitive” found across HCI, psychology, (management) decision making and neuroscience. They gathered 112  key attributes and categorised these into 17 groups spanning from “attentional focus” to “subjective ease”.

In HCI, intuition has been mainly discussed in terms of intuitive use or intuitive interaction (e.g., Blackler, 2019b; Blackler & Hurtienne, 2007; Blackler & Popovic, 2015; Hurtienne, 2011; O’Brien et al., 2010; Ullrich, 2014). There is a long list of characteristics ascribed to intuitive use. Intuitive products, services or systems are often described as requiring no prior training and feeling familiar to the user (Blackler, 2019a; Raskin, 1994, 2000). There is no need to think about interacting with the technology (Blackler, 2019a; Reinhardt et al., 2018). Users can rely on their gut feeling or instinctive thoughts (Mohs et al., 2006; Ullrich & Diefenbach, 2011; Wennberg et al., 2018). Intuitive use involves non-conscious information processing and the utilisation of prior knowledge (Blackler, 2006, 2019a; Blackler & Popovic, 2015; Hurtienne, 2011; O’Brien et al., 2010; Wennberg et al., 2018). Intuitive interactions are said to be fast (Blackler, 2006), easy to learn (Baerentsen, 2000; Spool, 2005), embodied (Desai et al., 2019; Turner, 2008) and imbued with magical experience (Ullrich, 2014). While many characteristics seem to describe cognitive processes, others describe design features from simplicity to whole interaction styles like natural user interfaces (Wennberg et al., 2018). There are also contradictions between single characteristics, for example, whether intuition is about applying prior knowledge (Blackler, 2006; Hurtienne, 2011), the ease of acquiring new knowledge (Baerentsen, 2000; Spool, 2005) or whether it includes trial and error and improvisation (O’Brien et al., 2010); whether intuitive use is generally fast and unconscious (Blackler, 2006) or not; whether there is a core set of necessary criteria (Hurtienne, 2011) or loosely coupled patterns of characteristics describing intuitive use (Ullrich, 2014).

The suggested methods for evaluating intuitive use can be as diverse as the stated characteristics of intuitive use. For the summative evaluation, mainly subjective methods (i.e., questionnaires) are used, sometimes complemented by objective methods including the coding of behavioural observations by experts and measures of effectiveness such as the proportion of correctly solved tasks (Blackler, Popovic, et al., 2019). Often, user tests with concurrent or retrospective think-aloud protocols function as the main setting for data gathering (Blackler, Popovic, et al., 2019; Reinhardt & Hurtienne, 2018). Sometimes intuitive use is assessed as mental workload with questionnaires or, more rarely, second-task performance measures (Reinhardt & Hurtienne, 2017).

As this cursory overview suggests, the varieties of definitions of intuitive use and characteristics to be measured are very large. Researchers and practitioners are left with the question of what is at the core of measuring intuitive use, how different characteristics of intuitive use relate to each other, and what evaluation methods they should use. Research to date has disagreed about which characteristics must be fulfilled to determine intuitive use and in which combination these characteristics must occur. Moreover, there is a lack of a common theoretical basis from which all postulated characteristics could be derived.

Previous analyses of how intuitive interaction and intuitive use are described by HCI researchers have tried to incorporate the breadth of the literature and have shown the diversity of the concepts (O’Brien et al., 2010; Turner, 2008; Wennberg et al., 2018). Recent reviews of the field of intuitive use have looked at its history, definitions and research methods in general (Blackler, 2019a; Blackler, Desai, et al., 2019; Blackler, Popovic, et al., 2019; Blackler & Popovic, 2015). Often the focus was on generating design knowledge, rather than bringing the theory and measurement of intuitive use together. A review of Still and Still (2019) comes close by focussing on the cognitive mechanisms of how prior knowledge comes into place and by discussing the usefulness of several questionnaires in determining intuitive use.

As long-standing eye-witnesses of the field we have been able to identify four research groups across the world who have been investigating intuitive use more deeply and have come up with theoretical and methodological suggestions that have been more integrated and more substantial than others. Nevertheless, as will be seen, also these approaches come to heterogeneous and sometimes contradictory conclusions. In this article we review this prior work, and, drawing on newer developments in psychological research, offer an integrative account of what intuitive use is and how it can be measured. We derive a new measurement definition of intuitive use with three central characteristics: high effectiveness of use, low cognitive effort and a strong metacognitive feeling of fluency. This allows us to differentiate between the defining characteristics and typical correlates of intuitive use suggested by prior research. It also paves the way to search for or develop alternative measures. This work should thus be seen as a proposal that can provide a common ground for further research and empirical validation. Note that although there would be much to say about the design for intuitive use, our focus is on the evaluation of intuitive use. The flavour of our approach will be cognitive-psychological, focussing on action-oriented information processing. Alternative approaches (e.g., phenomenological) would also be illuminating but are not pursued here.

We proceed as follows. First, we review the most elaborated and influential theories of intuitive use that have been proposed to date and discuss their implications for measuring intuitive use. Second, we review three psychological theories that allow to explicate and further our understanding of intuitive use: action regulation, processing fluency and default-interventionist theories of decision making. We develop a theory of intuitive use around the tripartite model of the mind (e.g., Stanovich et al., 2014) and propose our new definition for measuring intuitive use. Third, we discuss the implications of the new measurement definition in the light of previous work and show how the theory and new measurement definition can help in the selection and further development of measures of intuitive use. Finally, we point out limitations of our work and define areas of future research.

1. Current definitions, characteristics and measures of intuitive use

Previous reviews in HCI show that it is useful to look into cognitive and psychological theory for a theoretic underpinning of what intuitive use is (Blackler, 2019a; Blackler, Desai, et al., 2019; Blackler, Popovic, et al., 2019; Blackler & Popovic, 2015; Still & Still, 2019). Looking at such previous accounts, it becomes clear, that a systematic review of how the term “intuitive” is used in HCI research might not be useful for a theoretical article, because the term has become so ubiquitous that it has acquired a very broad meaning denoting general usability or just “good interaction” (cf. Wennberg et al., 2018). Thus, rather than presenting a systematic literature review, we focus here on a narrative review of the main research strands that have elaborated more clearly their understanding of what intuitive interaction is and how it can be measured. Our style of literature review can be described as hermeneutic, involving several cycles of literature search and acquisition, analysis and interpretation including critical assessment and argument development (cf. Boell & Cecez-Kecmanovic, 2014). Thereby, our understanding and insights have been evolving continuously.

Four main research groups have provided elaborated and influential theories and definitions of intuitive interaction and have suggested and developed means of measuring intuitive use: the QUT (Queensland University of Technology) research group from Australia, the IUUI (Intuitive Use of User Interfaces) and INTUI research groups from Germany, and a group from Georgia Tech (GT) in the US. There are many other researchers working on intuitive use who will be briefly reviewed in a further subsection. Despite some commonalities in their understanding what intuitive use is, these approaches differ in their viewpoints, theoretical details and implications for measuring intuitive use.

1.1. QUT: First conceptualisations of intuitive interaction

1.1.1. QUT: Definition and characteristics of intuitive use

The research group around Alethea Blackler at the Queensland University of Technology (QUT) in Australia was one of the first groups theorising intuitive use in HCI. The basis of their definition is a literature review in the fields of intuitive decision-making, creative thinking, philosophy, psychology of learning, neuroscience and cognitive science (Blackler, 2006, 2019b; Blackler et al., 2002, 2003, 2010; Blackler & Popovic, 2015). From this, and after several revisions, the research group derived the following definition (Blackler, 2006, p. 120): “Intuitive use of products involves utilising knowledge gained through other experience(s) (e.g. use of another product or something else). Intuitive interaction is fast and generally non-conscious, so that people would often be unable to explain how they made decisions during intuitive interaction.”

Three characteristics of intuitive use can be derived from the QUT definition: (1) the unconscious application of prior knowledge during use, (2) the temporally efficient (fast) use and (3) the lack of verbalisability. Although the QUT research group does not make it explicit in their definition, these characteristics are understood as variables; thus, the overall level of intuitive use can vary on a continuum (Blackler, 2006). The extent of intuitive use is reflected equally in all three characteristics, i.e., it is not specified whether one characteristic is more important than another to indicate intuitive use.

Regarding the first characteristic, the unconscious application of prior knowledge, Blackler (2006) suggests that intuition takes place at the rule-based level in the Skills-Rules-Knowledge Model by Rasmussen (1983). Here, familiar if-then rules are retrieved from memory and executed when triggered by the appropriate cues from the environment. On the skill-based level there is non-conscious automatic processing of highly over-learned tasks that would be regarded automatic, not necessarily intuitive (but would also fall under the QUT definition of intuitive use). In contrast, at the knowledge-based level conscious, non-intuitive processing would occur when dealing with unfamiliar situations.

As a second characteristic of intuitive use, the QUT research group describes the high temporal efficiency of intuitive use (e.g., Blackler, 2006, 2019a; Blackler et al., 2010; Blackler & Popovic, 2015). The predominantly unconscious application of prior knowledge should lead to a higher speed of information processing—in contrast to conscious information processing, which takes much longer (Baars, 1993; Bastick, 1982, 2003; Evans & Stanovich, 2013; Hammond, 1993; Hogarth, 2001; Price & Norman, 2008; Zajonc, 1980). Intuitive use therefore is visibly more elegant, faster, more accurate and economical than behaviour based on conscious cognitive information processing (Agor, 1986; Blackler, 2006, 2019; Blackler & Popovic, 2015). However, high temporal efficiency in cognitive information processing should not be confused with physical efficiency in executing motor actions (Blackler, 2006). Fast cognitive processing can be measured both objectively as a measure of task performance and subjectively in terms of a perceived low temporal demand in cognitive information processing (Blackler, 2019; Blackler, Popovic, et al., 2019).

The third characteristic of intuitive use is the users’ low ability to verbalise their decisions. When intuitive use is characterised by sub-conscious information processing, users should not be aware of the content of that processing and hence unable to verbalise it (Blackler, 2006, 2019a; Blackler & Popovic, 2015; Still & Still, 2019). Intuitive cognitive processes are rarely comprehensible, difficult to remember and can be hardly formalised (Agor, 1986; Bastick, 1982, 2003; Fischbein, 1987; Gigerenzer, 2007; Hammond, 1993; Klein, 2008; Price & Norman, 2008; Zander et al., 2016). The caveats are that verbalisation ability can vary greatly between people, and that people often fail to verbalise in the face of complete lack of knowledge, uninformed trial-and-error behaviour and when they are under high cognitive workload (Blackler et al., 2011).

1.1.2. QUT: Measuring intuitive use

The QUT research group has developed two methods for assessing intuitive use: the Technology Familiarity Questionnaire (TFQ) and a behaviour coding method based on a user test. In addition, in their research they also used traditional usability measures like error rates, click counts and success levels (Blackler, Popovic, et al., 2019).

The Technology Familiarity Questionnaire (TFQ) refers to the first characteristic of intuitive use, the unconscious application of prior knowledge. This questionnaire inquires about users’ prior experience of technology with regard to a specific product (e.g., a digital camera) in terms of the number of functions or features used and the frequency of use. These dimensions are also recorded for similar systems and added up to an overall technology familiarity score. The assumption is that frequently used products and features lead to a better knowledge and higher familiarity—and thus the unconscious application of prior knowledge is more likely when interacting with these products (Blackler, 2006; Blackler et al., 2010).

As there is no universal TFQ, a TFQ must be created for each specific product to be evaluated. As a consequence, the ranges of TFQ scores can vary greatly, and the TFQ scores of different studies are not comparable. However, it has been shown that the TFQ can sensitively predict differences in intuitive use and diagnostically identify different causes of (un)intuitive use, for example, the lack of prior experience with a particular system (e.g., Blackler, 2006; Blackler et al., 2010; Desai et al., 2015; McEwan et al., 2014). Further variants of the TFQ include a Games Technology Familiarity Questionnaire (GTF; McEwan et al., 2014) and an Airport Environment Familiarity (AEF) questionnaire (Cave et al., 2014).

All three characteristics of intuitive use can be assessed through a video coding procedure developed by the QUT group. A fourth characteristic (effectiveness through correctness of use) is introduced through the use of this method. Users’ interactions with a system are video-recorded in a usability test with a concurrent think-aloud protocol (Blackler, 2006). Experts then analyse each interaction assessing the correctness of use and the type of use (Figure 1). Correctness of use is assessed by assigning one of five codes (1) correct use, (2) correct use but inappropriate for the task, (3) incorrect use, (4) attempted but unsuccessful use, or (5) help received from the experimenter.

Figure 1. QUT video coding scheme after Blackler (2006), adapted from Reinhardt et al. (2018).

The type of use is assessed in terms of five criteria: (1) low latency: the user initiates an action within 5 s; (2) decision certainty: actions are executed with confidence (no trial & error); (3) lack of verbalisation of cognitive information processing during interaction: there are no signs of conscious processing and no reference to interaction details; (4) the user states clear expectations based on action-relevant prior knowledge; (5) users refer to past experience with the product or features, thereby indicating prior knowledge. In order to be considered intuitive, an interaction must be correct and fulfil at least two type-of-use criteria.

This initial coding scheme has evolved to also include a third category of partially intuitive codings that describe a combination of intuitive and non-intuitive cognitive processes (Cave et al., 2014; Desai et al., 2019). In later research by the QUT research group, also larger chunks of video have been coded, comprising of small subtasks rather than single features (Blackler, Popovic, et al., 2019). In other studies, the following additional features of intuitive use were employed in video coding:

• Grouping: users describe actions as coherent units, not in detail (Lawry, 2012);

• Procedure: users’ actions reveal chains of individual operations that occur continuously and without pause (Lawry, 2012);

• Anticipation: users prepare their next action step before completing the previous one, e.g., positioning the mouse pointer on the “Next” button before reading a page (Lawry, 2012);

• Aspects of embodiment, i.e., actions determined by physical and perceived affordances, emergence of actions as dynamic systems or knowledge evolves, environmental scaffolding, and cooperative activity (Desai et al., 2019).

While allowing for a great detail of analysis at the interaction level, these video coding procedures have several drawbacks: they require a high degree of expert knowledge of the coders; they are time-consuming to implement; and their results can be context-dependent (e.g., users’ willingness and ability to verbalise).

1.2. IUUI: Intuitive use as a sub-concept of usability

1.2.1. IUUI: Definition and characteristics of intuitive use

The IUUI research group was formed in 2005 at the Technical University of Berlin (Germany). Based on a review of usability criteria, an analysis of manufacturers’ statements about “intuitive use”, user interviews and workshops with usability experts, a definition of intuitive use was derived and subsequently refined (Hußlein et al., 2007; Israel et al., 2009; Mohs et al., 2006; Naumann et al., 2008). The latest version reads: “Intuitive use is defined as the extent to which a product can be used by subconsciously applying prior knowledge, resulting in an effective and satisfying interaction using a minimum of cognitive resources” (Hurtienne, 2011, p. 29).

Four characteristics of intuitive use can be derived from the IUUI definition: (1) the unconscious application of prior knowledge during use, (2) effectiveness, (3) mentally efficient use, and (4) satisfactory use. The first characteristic is seen as the prerequisite for intuitive use. The other three characteristics refer to consequences of intuitive use (Hurtienne, 2011). All characteristics can vary on a continuum: intuitive use is higher the more subconscious, effective, mentally efficient and satisfactory the interaction is (Hurtienne, 2011). All characteristics must be fulfilled. There is no intuitive use without, for example, sufficient effectiveness or satisfaction.

The IUUI group explicitly understands intuitive use as a sub-concept of usability (Hurtienne, 2011; Naumann et al., 2007). According to ISO standard 9241-11 usability is “the extent to which a system, product or service can be used by specified users to achieve specified goals with effectiveness, efficiency and satisfaction in a specified context of use” (ISO, 2018, p. 6). It follows that intuitive use, like usability, is not a property of the product, but describes the fit between goals, user characteristics, the product and the environment. As in the definition of usability, intuitive use is about effectiveness, efficiency and satisfaction:

• Effectiveness is “the accuracy and completeness with which users achieve specified goals” (ISO, 2018, p. 9). Effectiveness is thus seen as a performance characteristic. If users are not able to achieve their goals with adequate precision and completeness, the use is neither usable nor intuitive (Hurtienne, 2011).

• Efficiency is defined as “the resources used in relation to the results achieved.” (ISO, 2018, p. 10). This is about maximising benefits at the lowest possible cost. However, according to IUUI, efficiency is limited to mental efficiency due to the subconscious application of prior knowledge. Intuitive use does not include physical or temporal efficiency (Hurtienne, 2011). This is to acknowledge that also a physically demanding or slow sequence of movements can serve to enhance mental efficiency. The example is an installation wizard that requires a user to make many mouse clicks for the benefit of being guided through a new and complex task with high mental efficiency (Hurtienne, 2011; Naumann et al., 2007). As task completion time is always a mix of mental and motor components, time measures can only be used as a criterion for assessing intuitive use when it is ensured that they represent mental demands only (e.g., latency in the CHAI method, see below).

• Satisfaction is defined as “the extent to which the user’s physical, cognitive and emotional responses that result from use of a system, product or service meet user’s needs and expectations.” (ISO, 2018, p. 11). The subjective assessment of the outcome of use is equally important for intuitive use and usability, although indicators related to information processing are predominantly suitable for assessing intuitive use. Various studies show that conscious information processing is not only less effective and less mentally efficient, but also leads to lower satisfaction than subconscious information processing (cf. Dijksterhuis & van Olden, 2006; Ullrich, 2014).

The major differences between the QUT and IUUI approaches are that IUUI sees general fastness not as a criterion of intuitive interaction if it is not related to mental efficiency, that IUUI is critical about the value of thinking-aloud analyses and that it explicitly mentions user satisfaction in the definition of intuitive use.

1.2.2. IUUI: Measuring intuitive use

Several methods for measuring intuitive use have emerged from the IUUI research group, including a questionnaire (QUESI), a coding procedure for behavioural observation via video analysis (CHAI), a secondary task measure (IntuiBeat) and, more speculative, measures of entropy (ECM).

QUESI (Questionnaire for the subjective consequences of intuitive use) was designed for measuring the satisfaction component of intuitive use. From the definition of intuitive use, IUUI derives five indicators that form the five subscales of the questionnaire (Hurtienne, 2011; Naumann & Hurtienne, 2010):

• Low subjective mental workload (three items, e.g., “I could use the system without thinking about it”)—derives directly from the characteristic of mental efficiency.

• High perceived achievement of goals (three items, e.g., “I achieved what I wanted to achieve with the system”)—derives directly from the characteristic of effective use.

• Low perceived error rate (two items, e.g., “No problems occurred when I used the system”)—derives directly from the characteristic of effective use.

• High familiarity (three items, e.g., “I could interact with the system in a way that seemed familiar to me”)—derives indirectly from the unconscious application of prior knowledge, as this should lead to a sense of familiarity among users.

• Low perceived effort of learning (three items, e.g., “The system was easy to use from the start”)—derives indirectly from the (subconscious) application of prior knowledge. Due to the available knowledge, the user’s learning effort remains comparatively low or learning occurs implicitly and incidentally (Baerentsen, 2000; Spool, 2005; Still & Still, 2019).

Users rate their agreement with each item on a five-point Likert scale. For all subscales, a higher value indicates a higher probability of intuitive use (Naumann & Hurtienne, 2010). The QUESI subscales as well as QUESI overall show a high internal consistency and good convergent and divergent validity (Hurtienne et al., 2009). With QUESI, the unit of evaluation is the whole interaction with the system, while the following methods allow for more fine-grained measures on the task level or the level of single interaction steps.

CHAI—Coding Heuristics for Assessing Intuitive Interaction—is a revision of the behavioural coding scheme of the QUT research group (Horn, 2008; Reinhardt et al., 2018). In contrast to the QUT procedure, CHAI does not use a concurrent think-aloud protocol to avoid its associated problems: think aloud acts as an additional task for the user; it can increase cognitive load (and thus influence intuitive use); its interpretation is highly dependent on contextual factors (e.g., users’ willingness and ability to verbalise) and on the evaluators’ expertise. CHAI uses only three dichotomous features: (1) latency: action initiation within 3 s; (2) effectiveness: correct and goal-directed interaction without assistance; and (3) as a measure of mental efficiency the estimated certainty of action execution, i.e., action execution without interruptions, prolonged searching or major deviations from the direct path of movement. All three characteristics must apply, i.e., as soon as one characteristic is not fulfilled, a specific interaction is not considered intuitive (Figure 2). The procedure provides a summary measure of intuitive use: the number of intuitive interactions (“intuitive clicks”) divided by the total number of interactions (i.e., total clicks). By just using three characteristics and termination conditions coding effort is greatly reduced in comparison to the QUT procedure.

Figure 2. Decision tree to differentiate between intuitive and non-intuitive interactions using the CHAI method (Reinhardt et al., 2018).

IntuiBeat is a secondary-task measure of mental efficiency operationalised as cognitive effort in use (Park & Brünken, 2015; Reinhardt, 2020; Reinhardt et al., 2018). While interacting with a system, users are asked to tap a given rhythm with their foot. This technique takes advantage of mental inhibition processes during action execution and thus can measure cognitive effort independently of modality. The extent of rhythm deviations correlates with the extent of intuitive use—the fewer deviations, the more intuitive (for psychometric validations of the method see Reinhardt, 2020).

ECM (Entropy of Controller Movements) dispenses with administering a secondary task and attempts to estimate mental efficiency from the irregularities of hand movements when interacting with a system. The method is based on findings that mental processes can influence motor processes, i.e., that cognition influences ongoing (inter)action (e.g., Freeman, 2018; Stillman et al., 2018). Initial validations with cursor data from mouse interactions (Reinhardt & Hurtienne, 2018) and movements of VR controllers (Reinhardt et al., 2019, 2020) show that this is a viable approach, as it allows to capture intuitive use without disturbing ongoing use and, most of the time, without the need for additional hardware.

1.3. INTUI: The experiential perspective

1.3.1. INTUI: Definition and characteristics of intuitive use

While the QUT and IUUI definitions of intuitive use predominantly consider objective performance indicators such as effectiveness, mental and temporal efficiency, the INTUI research group, located in Munich (Germany), focuses on the user experience in the form of subjective perceptions of intuitive use (e.g., Diefenbach & Ullrich, 2015; Tretter et al., 2019; Ullrich, 2014; Ullrich & Diefenbach, 2010a, 2010b). According to INTUI, these are reflected in four components:

• Low verbalisation—as adopted from the definition of the QUT research group (Diefenbach & Ullrich, 2015; Ullrich, 2014; Ullrich & Diefenbach, 2010a, 2010b). Here it is measured as the subjective judgment instead of an objectively observable behaviour.

• Effortlessness—describing the generally high level of effortlessness that is subjectively perceived during intuitive use (Diefenbach & Ullrich, 2015; Ullrich, 2014; Ullrich & Diefenbach, 2010a, 2010b).

• Gut feeling—capturing how information processing feels guided by feelings in a holistic manner (Diefenbach & Ullrich, 2015; Tretter et al., 2019; Ullrich, 2014; Ullrich & Diefenbach, 2010a, 2010b). Following Hammond (1993), intuitive use is understood as information processing that provides an action tendency (i.e., solution, answer, insight) without the user being aware of how the concrete solution is found. Even if a user cannot explain where their decision comes from, they are highly certain that it is the right decision (Ullrich, 2014).

• Magical experience. Despite this subjective sense of certainty, intuitive use is experienced as rather vague, unaccountable, uncoordinated and gut-feeling (Price & Norman, 2008). Because the user has difficulty explaining and verbalising the source of their increased perceived certainty (Hammond, 1993), this positive feeling can feel almost ‘magical’ to the user (Diefenbach & Ullrich, 2015; Ullrich, 2013, 2014).

These four components do not need to occur simultaneously and to the same extent, i.e., not all characteristics have to be present for a product, service or technology to be perceived as intuitive (Ullrich, 2014). In a specific context of use, for example, verbalisation can manifest strongly while effortlessness hardly manifests at all. The extent of intuitive use is thus reflected in so called INTUI patterns (Diefenbach & Ullrich, 2015; Tretter et al., 2019; Ullrich, 2014). Game consoles and household appliances, for example, are clearly different in their INTUI patterns: the former are characterised more by gut feeling and magical experience, the latter have an emphasis on effortlessness and verbalisability (Diefenbach & Ullrich, 2015; Ullrich, 2014; Ullrich & Diefenbach, 2010b).

1.3.2. INTUI: Measuring intuitive use

To measure the subjective consequences of intuitive use, the INTUI group has developed the INTUI questionnaire. The questionnaire is used to assess the entire interaction with a system, not just a single task (Ullrich, 2014). It captures the four INTUI components with 16 items, which are presented as a seven-point semantic differential:

• Low perceived verbalisation ability (three items, e.g., “In retrospect, I have no problem describing the individual operating steps/it is hard for me to describe the individual operating steps“)

• High perceived effortlessness (five items, e.g., “Using the product came naturally/was hard“)

• High perceived gut feeling (four items, e.g., “While using the product, I was guided by feelings/I was guided by reason“)

• Strong magical experience (four items, e.g., “Using the product was fascinating/was dull”)

An additional item asks for intuitive use directly: “Using the product was very intuitive/wasn’t intuitive at all.” The scores of the four components are calculated by averaging the corresponding items and are interpreted in their relative values to each other (i.e., the INTUI pattern). The questionnaire does not provide an overall score of intuitive use. The subscales of the INTUI questionnaire show a high internal consistency and a confirmatory factor analysis attested the construct validity of the four components (Ullrich, 2014).

1.4. Georgia Tech: Extending the theoretical framework

1.4.1. GT: Definition and characteristics of intuitive use

O’Brien et al. (2008, 2010) from Georgia Tech in the US provide a “Framework for Intuitive Human-Computer Interaction”, based on a review of research on intuition in general psychology, educational psychology, management psychology, decision-making, cognitive engineering, and neuroscience combined with a review of HCI research on novice users of technology and current design guidelines.

Their framework has five key assumptions: (1) User behaviour is goal-oriented. Goals may be concrete and functional like completing specific tasks; goals can also be more abstract needs like striving for beauty, pleasure, or truth. (2) Users select cognitively efficient, well-learned activities if they are immediately accessible and unconsciously judged to be appropriate for the current environment and context. (3) Users determine what to do next by relying on few and redundant cues from the environment and “fill in” the rest based on prior experience. They do so until a constraint is met that requires the additional evaluation of available actions in the current context to proceed with the activity. (4) Metacognition monitors behaviour and guides the selection of what to do next by relying on a ‘feeling of knowing’ based on prior knowledge. (5) Knowledge in the world should be designed to map to knowledge in the head. Knowledge in the world encompasses system features and environmental information available for particular interactions. Knowledge in the head encompasses prior knowledge that has been implicitly or explicitly learned by the user. Three activities are necessary for intuitive use: seeking general goals, performing well-learned activities, and determining what to do next.

Important for intuitive use to develop, especially with regard to determining what to do next, is a “lenient learning environment.” In such an environment, small errors are expected and can be managed easily, and frequent salient feedback allows learning and progress toward the goal. Guesses do not have to be correct but only sufficient for rapid hypothesis testing and online correction. Taken together, the authors define intuitive use as the “interaction between humans and high technology in lenient learning environments that allow the human to use a combination of prior experience and feedforward methods to achieve their functional and abstract goals” (O’Brien et al., 2010, p. 5). Note that this definition is descriptive rather than providing concrete prescriptions for measuring intuitive use.

O’Brien et al. (2008, 2010) point out promising concepts that are not emphasised by the other frameworks. Goal-directedness, for example, as a central means of regulating activity, may be fuelled by concrete and abstract goals. O’Brien et al.’s framework is also unique in its reliance on metacognition for selecting between modes of processing. There are three modes. Analytic processing largely follows Norman’s (1990) seven stages of action: goal setting, intention forming, action selection, action execution, perception of feedback, interpretation, and evaluation. Intuitive processing is split into the two components: performing well-learned activities and ‘true’ intuitive processing. ‘True’ intuitive processing involves low-effort hypothesis testing to determine the next action. It relies on the environment for the fast evaluation of action possibilities and whether the (mentally simulated) goal has been reached. Meta-cognitive feelings determine which type of processing is applied and how “true” intuitive processing with its guess-action cycles will proceed. First, prior research has established that feeling of knowing can predict the accuracy of problem solving. If feeling of knowing exceeds an activation threshold, an action will be automatically executed. Second, a newly proposed feeling of directness results from performing well-learned activities and can be seen as a form of the certainty with which an action is selected. Third, feeling of progress towards the goal is derived from the general goal seeking and can regulate the guess-action cycle so that it successfully proceeds to goal achievement.

Note that this framework allows for more flexible action control that goes beyond performing well-learned activities that seem to be in the focus of, for example, the IUUI framework. For this to be successful, actions depend on lenient learning environments. O’Brien et al.’s framework thus develops intuitive interaction further and directly binds it to the design of intuitive technology. However, the full framework only appeared as a working paper and so far, it has not been applied and validated empirically.

1.4.2. GT: Measuring intuitive use

O’Brien et al. (2008, 2010) suggest several measures to determine intuitive use. The first is the temporal analysis of computer mouse patterns, where bursts of activity with low latency and low variability between clicks are expected when executing intuitive and well-learned behaviour. Behavioural chunks corresponding to the achievement of subgoals are expected to be segregated by pauses between mouse clicks. Thus, within-chunk behaviour should be different from between-chunk behaviour. Pauses after errors should be short for intuitive but substantial for analytical processing. More research is needed, however, to specify the expected patterns of mouse clicks and pauses for matching with intuitive vs. analytic behaviour patterns. Second, users can be directly asked to rate system intuitiveness on a Likert scale at designated points during interaction. Third, metacognitive judgments can be captured by having the users record or say, for example, how confident they are in their current position to reach the goal. Fourth, using think aloud, user’s commentary can be recorded and subsequently coded for the use of analytic or intuitive processes (cf. Hamm, 1985). However, all these methods remain speculative as they have not been applied so far.

1.5. Other groups and measures

1.5.1. Other groups: Definition and characteristics of intuitive use

Most researchers writing about intuitive use are referring to one or more of the above cited groups (QUT, IUUI, INTUI, GT) for definitions and theoretical background. There is, for example, agreement that intuitive use includes the subconscious use of prior knowledge, but there are differences as to what the consequences of intuitive use are and how these should be measured.

Macaranas (2013, p. 10), for example, defines intuitive interaction as “interaction with an unfamiliar system where the user knows how to act quickly and automatically and with unconscious effort and attention.” Interesting here is the focus on unfamiliar systems while earlier definitions like Raskin’s (1994, 2000) highlight that “intuitive equals familiar” and the design of intuitive, i.e., familiar, interfaces will eventually hamper progress because new interaction styles will not be invented that way. One attempt at reconciling this seemingly contradiction is to look not only at familiar knowledge about technology, but also from other sources, so that although a piece of technology is unfamiliar to users they still can apply knowledge from other domains (Blackler, 2006; Hurtienne et al., 2015). Turner (2008), after a review of the HCI literature, concludes that the intuitive has at least two meanings: familiarity and embodiment, the latter mainly understood as action-perception coupling.

Baerentsen (2000, p. 32) highlights that an intuitive interface “is immediately understandable to all users without the need for special knowledge by the user nor for the initiation of special educational measures” (emphasis ours, for a similar take see Bullinger et al., 2002). Learning, however, might still take place, if it is “a spontaneous product of the activity of use.” A similar argument has been made from a practitioner’s perspective by Spool (2005), who sees intuitive use as a match between the knowledge required to operate a system and the knowledge a user already has. If the system and user knowledge do not match, intuitive use is still possible, as far as this knowledge gap can be closed with low cognitive effort. This can be likened to the “determining what to do next” type of intuitive use in the framework of O’Brien et al. (2008, 2010).

From an instructional design perspective, Fischer et al. (2015b) include in their definition aspects of prior knowledge matching and cognitive effort: “a device is ‘intuitive’ when it may be used as intended, without apparent effort, training or outside help” (p. 256).

Looking at the subjective perception of intuitive use, McAran (2019) extends the Technology Acceptance Model (TAM) aimed at predicting the degree of use of any technology, with a new factor intuitive interaction. This factor consists of the three constructs perceived intuitiveness (“the degree to which the technology product is perceived by the user as capable of being used without conscious awareness of rational thinking”), compatibility (“the degree to which the use of technology is perceived by the user to be consistent with his or her practice style or preferences”) and perceived ease of use (“the degree to which an individual believes that using a particular system will be free of physical and mental effort”).

1.5.2. Other groups: Measuring intuitive use

Approaches differ in what measures of the consequences of intuitive use they propose and employ. A first cluster of approaches uses sophisticated methods to quantify intuitive use based on an assessment of prior knowledge. A second cluster specifically employs mental workload measures, and a third cluster measures intuitive use subjectively. While many more approaches may be possible, we focus here on these that have been used in an explicit connection to intuitive use. Many studies, however, combine several methods. Macaranas (2013), for example, combines questionnaire items, interviews, and expert ratings of user interactions. Ratings include conformity to expectations, verbalisation ability and locus of attention (e.g., “Attentional focus on operating the system or on completing the task?”). Intuitive use is assumed when users act correctly but fail to verbalise what they have done (Antle et al., 2009; Bakker et al., 2009; Macaranas, 2013).

In the first cluster, there have been very original approaches to quantitatively assess the prior knowledge of users. In their schema-matching account, Fischer et al. (2015a, 2015b), for example, assume that intuitive use occurs when the knowledge schemas available to the user match with those required by the system. When a system is intuitive to use, new knowledge does not need to be constructed. Thus, giving users time to study the user interface prior to using the system will not increase their performance. If performance gets better after studying the interface, it is not intuitive to use. More specifically, Fischer et al. (2015a, 2015b) use different priming techniques (e.g., keyword or function matching) and compare these to the unprimed use of the software to determine the performance differences.

Another interesting approach uses image schemas to describe both, the existing knowledge of the user and the knowledge required to operate the system (Asikhia, 2015). Image schemas are abstract mental representations of basic sensorimotor experiences (Johnson, 1987). Example image schemas represent fundamental spatial relations such as up-down or centre-periphery, attributes such as bright-dark or heavy-light, and force relations such as resistance or blockage. Because image schemas are frequently used, they have become automated and can be processed subconsciously with little cognitive effort. Image schemas can be instantiated in language, behaviour and user interfaces, and due to their subconscious processing can help to improve intuitive use when designing systems (Hurtienne, 2017; Löffler et al., 2013). In Asikhia’s approach, the image schemas the users employ are analysed through observations and think-aloud protocols during system use. The image schemas that the system requires the users to know for effective operation, are analysed in the language of the user manual and in a list of necessary steps to achieve the user goals (Asikhia, 2015). The amount of overlap between the image schemas the users employ and the image schemas that the system requires the users to know is represented by the so-called Q-value. The higher the Q-value is the more intuitive use is likely because the required and the actual user knowledge overlap. Asikhia (2015) could show that the Q-value correlates highly with cognitive effort, effectiveness and time efficiency and thus represents a valid measure for summative evaluations of intuitive use.

The second cluster of approaches to measuring intuitive use employs mental workload measures (e.g., Chattopadhyay & Bolchini, 2014; Hurtienne & Blessing, 2007; O’Brien, 2019; Okimoto et al., 2012; Reinhardt et al., 2018; Tscharn et al., 2017; Winkler et al., 2016). This includes many questionnaires adopted from other research areas in Human Factors and HCI. One example is the Subjective Mental Effort Questionnaire (SMEQ)—also known as the Rating Scale Mental Effort (RSME) (Rubio et al., 2004; Zijlstra, 1993; Zijlstra & Van Doorn, 1985). The questionnaire consists of a single vertical scale with verbal and numerical anchors. Users can mark the scale after each task to indicate the mental effort expended. Another frequently used mental workload measure is the NASA Task Load Index (NASA TLX; Hart & Staveland, 1988). It uses six scales to assess effort: mental, physical and temporal demand, as well as subjective performance, effort and frustration (Galy et al., 2018; Hart, 2006; Hart & Staveland, 1988). Although the questionnaire originally included a weighting for these dimensions, for the subjective evaluation of intuitive use it is commonly used in its unweighted, “raw” version (i.e., NASA-RTLX). Most of the time, the six scales are collapsed into a single measure of effort (e.g., Chattopadhyay & Bolchini, 2014; O’Brien, 2019; cf. Grier, 2015; Hart, 2006).

In the third cluster of approaches, and as a direct subjective measure of intuitive use, McAran (2019), for example, in his extension of the TAM, proposes three items to measure perceived intuitiveness concerned with minimal training, being clear what to do to use the product and adaptation to user goals. Finally, in studies of video gaming, the PENS questionnaire (Player Experience of Need Satisfaction; Ryan et al., 2006) is frequently deployed (e.g., by McEwan et al., 2014). It contains the subscale Intuitive Controls with three items asking about ease of learning, ease of recall, and direct intuitive use.

1.6. Interim summary: Large variation of measures

By reviewing the main positions in the study of intuitive interaction we have seen that most approaches agree on the preconditions of intuitive use—that it is based on prior knowledge and that the application of prior knowledge is subconscious or at least in some way connected to using a minimum of cognitive resources. There is however, a fundamental difference in whether the required knowledge to use a system should be already available (IUUI) or can be actively constructed on-the-fly without too much cognitive effort (GT; Spool, 2005). Differences also exist in what kind of prior knowledge is focussed on, e.g., prior knowledge regarding the use of technology only (Raskin, 1994, 2000), a wider range of knowledge from different domains (IUUI, QUT), or specific knowledge like image schemas inspired by sensorimotor experiences (Asikhia, 2015). Clearly, the empirical operationalisation provides boundaries of what knowledge is operationalised in a specific study and how easily it can be extracted. Prior knowledge has been operationalised by using questionnaires like the Technology Familiarity Questionnaire (QUT), by video-coding user comments or by inferring chunks of procedural knowledge from the smooth execution of actions (QUT, GT).

The assumed consequences of intuitive use, and thus the measures applied to recognise them differ widely. Often these differences stem from the different paradigmatical perspectives the researchers take (e.g., user experience, usability, technology acceptance), the application domains considered (e.g., legal software, natural user interfaces), and the scope of application (e.g., all systems, only unfamiliar ones, lenient learning environments). Of interest for this article is what consequences of intuitive use the different research groups postulate and how these are operationalised in empirical studies (Table 1). Methodologically, we can differentiate between

Table 1. Characteristics and measures of intuitive use as proposed by previous research (see text for abbreviations)

Characteristic of intuitive use	Proposed by	Questionnaire	Behaviour coding	Verbal protocol analysis	Logfile analysis	Other
Subconscious application of prior knowledge	QUT, IUUI, GT, Others	TFQ, QUESI, TAM, PENS	QUT: Certainty, Anticipation, Embodiment	QUT: clear expectations, reference to past experience; GT: analytic and intuitive processing; Macaranas (2013)		Q-score Asikhia (2015); Schema-matching (Fischer et al., 2015a,b)
High temporal efficiency	QUT, Others		QUT: low latency, procedure; CHAI: low latency		GT: temporal analysis of mouse patterns
Low verbalisation	QUT, INTUI, Others	INTUI		QUT: Lack of verbalisation, grouping; Macaranas (2013)
Effectiveness	QUT, IUUI, GT	QUESI, INTUI	QUT/CHAI: correctness
Mental efficiency	IUUI, INTUI, Others	PENS, NASA TLX, QUESI, SMEQ/RSME, INTUI, TAM			IUUI: ECM; IntuiBeat
Satisfaction	IUUI	QUESI
Gut feeling	INTUI	INTUI
Magical Experience	INTUI	INTUI
Intuitiveness	INTUI, GT	INTUI, PENS, TAM
Metacognitive Feelings	GT	Feeling of Knowing, of Directness, of Progress
Locus of attention: system/task	Others		Macaranas (2013)

• Subjective evaluations by the user obtained via questionnaires;

• Performance measures obtained through expert coding of user behaviour, e.g., from video data;

• Expert coding of user utterances;

• Log file analysis; and

• Others (requiring specialised setups or employing mixed approaches).

Research groups differ in the importance granted to the different constructs and measures. For example, fast interaction is a central component in the QUT definition of intuitive use whereas IUUI actively excludes time efficiency—if it is not understood as a direct expression of mental efficiency, however. Sometimes theory and operationalisation are of different scope. The QUT group, for example, discusses a wide range of prior knowledge from different domains, but the operationalisation with the TFQ is done on prior technology experience alone. Several groups employ measures that are unique to them and that can be seen as spearheads in exploring new territory in measuring intuitive use e.g., the construct of magical experience by INTUI, the entropy of controller movements by IUUI or the Q-value based on analysing image schemas by Asikhia (2015).

The groups also propose different relations between the components of intuitive use. According to the IUUI and QUT research groups, for example, all named characteristics of intuitive use should co-vary to some extent. In their coding procedures, CHAI and QUT video coding, however, the two groups differ on whether all proposed characteristics of intuitive use need to be given simultaneously to diagnose intuitive interaction (IUUI) or whether a subset is sufficient (QUT). The INTUI approach is still different: their four components of intuitive use do not need to co-vary: different contexts of use will lead to different INTUI patterns.

Together previous researchers have opened up a large space for defining and measuring intuitive interaction. But while there is some overlap between the approaches, no approach is large enough to encompass the others and some proposed characteristics even contradict each other. Everyone outside the field but eager to include measures of intuitive use in their research might be puzzled by the diversity of proposed characteristics and measures. They would need to decide (1) which approach to follow, (2) how to combine features/measures of different approaches or (3) to come up with their own theories and measures of intuitive use. The first approach risks incompleteness as no theory seems to cover all aspects of intuitive use. The second approach risks incompleteness, too, as it will be impractical to use all operationalisations the research groups have suggested. The third approach might require a large effort and may add even more confusion to the current state of measuring intuitive use.

To avoid that the decision on how to measure the consequences of intuitive use is guided by practical issues alone (e.g., access to a particular logging technology, time available for video coding) we need to have an idea of what constructs are at the core of intuitive interaction and which can be seen as derivations from or elaborations of the core measures. Temporal efficiency might be an important example. While the subconscious application of prior knowledge itself is hard to measure, it can possibly lead to swift interactions. Plus, response times are easily measured. While it is true that the subconscious application of prior knowledge leads to faster mental reactions, fastness in general can also be due to an ease of motor execution or more powerful commands that automate several manual steps. Thus, time efficiency is probably not the defining feature of intuitive use, but an important correlate under specific conditions (i.e., when it can be made sure that only mental components are reflected in the time measure).

To better clarify which of the proposed characteristics and measures ae defining intuitive use and which are its correlates we take a deeper look into current psychological theories. Most of the research groups presented here take their starting point in information processing theories in judgment and decision making. At the same time, there is an emphasis on action regulation when interacting with technology (IUUI, GT). Thus, in the following, we review developments in psychological theories on action regulation, metacognition and default-interventionist processes of action control. By this we draw on several psychological subfields: work, social and cognitive psychology. The result is an updated theoretical framework for measuring the consequences of intuitive use and a new measurement definition of intuitive use geared at operationalisation. We suggest which set of measures is mandatory and how previous (and possibly new) measures relate to this set.

2. Anchoring intuitive use in overarching theoretical frameworks

To determine which characteristics of intuitive use are defining the concept and which are correlates, we review three areas of psychological theory. First, using action regulation theory we set the broader basis of how intuitive use can be understood as goal-oriented action. Second, the theory of processing fluency combines objective and subjective components, which act as correlates of many measures of intuitive use. Third, two-process theories of cognition and their further development into the tripartite model of the mind (Evans & Stanovich, 2013; Stanovich et al., 2014) give us an account of how intuitive information processing advances. Altogether, these theories allow us to identify a key subjective (i.e., the metacognitive feeling of fluency), a key objective (i.e., cognitive load) and a pragmatic feature (i.e., effectiveness) of intuitive use as core elements of a new measurement definition.

2.1. Action regulation and information processing

There is agreement between researchers that intuitive use involves action regulation based on prior knowledge. QUT, for example, discusses this at the basis of Rasmussen’s (1983) Skills-Rules-Knowledge model (SRK) and IUUI refers to Rasmussen’s (1986) model of the human information processor to explain intuitive use as subconscious information processing. Common to these perspectives (but often not stated explicitly) is the assumption that action regulation can be seen as goal-related behaviour. O’Brien et al. (2008, 2010) make the goal-relatedness of action regulation a central part of their theoretical framework.

In the traditional account of action regulation, a goal-oriented exchange of information takes place in the form of users interacting with the system until the discrepancy between a target state, defined by the user goal, and the actual state (i.e., action result received from the environment) is overcome. The action cycle includes a feedback loop, and has been described as a TOTE unit: Test-Operate-Test-Exit (Miller et al., 1960). If the users have achieved their goal, the action regulation is finished, otherwise the action cycle is iterated until a satisfactory congruence between the actual and anticipated target state is achieved (Frese & Zapf, 1994; Hacker, 1986; Hacker & Sachse, 2013; Zacher, 2017; Zempel, 2003). More sophisticated models view action as progressing in several phases, e.g., Norman’s seven stages of action (1990) or as six phases that can be described as (1) goal development and selection between competing goals, (2) orientation including prognosis of future events, (3) plan generation, (4) plan selection, (5) execution and monitoring and (6) feedback processing (Frese & Zapf, 1994; Zacher, 2017).

2.1.1. Action regulation as information processing

In the process of action regulation, users actively collect information from the environment regarding their action goals. They store the information, combine it with their prior knowledge to form a mental model and then apply this model to solve problems when using a system (see Diefenbach & Hassenzahl, 2017; Saifoulline & Hemberger, 2011). The more reality-appropriate (i.e., the more accurate and differentiated) the model is during action regulation, the more successful an activity can be carried out (Ulich, 1994). Thus, intuitive use does not just require the application of prior knowledge; prior knowledge also needs to be reality-appropriate to be effective.

Long-term memory contains the user’s prior knowledge and handles the long-term storage of information. It has virtually unlimited capacity, but access is relatively slow (Atkinson & Shiffrin, 1968). Knowledge in long-term memory can be explicit or implicit. Explicit knowledge can be formally verbalised and can be consciously activated. Implicit knowledge is activated unconsciously and cannot, or only incompletely, be verbalised. Procedural knowledge (i.e., knowledge about action and skills) is predominantly implicit and declarative knowledge (i.e., knowledge about facts, meaning of symbols, concepts and principles of a particular domain) is predominantly explicit.

Prior knowledge is dynamic, since it is composed of different sources of knowledge and, through intensive and frequent use, can be made unconsciously accessible. Most of prior knowledge is learned through basic sensorimotor experiences, culture or areas of expertise such as one’s profession or hobbies (Bargh & Chartrand, 1999; Hurtienne, 2011).

Mental models are maintained in working memory (Hacker, 1986). Working memory is responsible for the short-term provision of information; its contents are accessible to conscious processing but is temporally and capacitively limited (Baddeley, 2000, 2001, 2012). It consists of four subsystems (Baddeley, 2000, 2012; Baddeley & Hitch, 1974; Lee et al., 2017). The first subsystem, the phonological loop, handles spoken and written information in phonetic form. The second subsystem, the visuo-spatial scratchpad, is specialised in storing and managing visual and spatial information. The third subsystem, the episodic buffer, allows information from the visuo-spatial scratchpad, the phonological loop and long-term memory to be integrated and stored as coherent episodes (i.e., chunks of information, Gooding et al., 2005).

A fourth subsystem is the central executive. It coordinates the other three subsystems: the phonological loop and visuo-spatial scratchpad as sensory inputs and the episodic buffer as the link between working memory and long-term memory (Baddeley, 2012). Other functions of the central executive include goal setting and prioritising, action planning, action implementation and the control of motor implementation (Jurado & Rosselli, 2007). For the purposes of this paper, we focus on three basic executive functions: inhibition, updating and shifting that are also involved in higher cognitive functions like goal setting and planning (Miyake et al., 2000; Nee et al., 2012):

• Inhibition describes the ability to control or inhibit impulsive unconscious action tendencies in order to make decisions through logical thinking and associated conscious control. This ability to inhibit allows us to perform less familiar but purposeful actions in unfamiliar situations. Inhibitory control implies filtering information and prioritising conscious control.

• Updating describes the ability to monitor and encode new incoming information. The extent to which the information is relevant to the current action is monitored and, depending on the result, a decision is made as to whether relevant information is maintained in working memory or whether irrelevant information is replaced by new information. Updating goes beyond passive storage and maintenance, as the processes involved also enable the active manipulation of information.

• Shifting allows people to dynamically switch back and forth between actions performed in parallel, between different operations and between conscious and unconscious information processing.

Current research assumes that although these processes can be distinguished theoretically, they always co-occur (Miyake et al., 2000). Accordingly, executive control can also be regarded as a unified system from which the user’s mental workload can be read, since its functions influence and condition each other (Jurado & Rosselli, 2007).

2.1.2. Levels of action regulation

Looking at action regulation as a sequence of several phases does not allow to determine whether it is an intuitive or a non-intuitive action based on the presence of unconscious and conscious information processing. Instead, a hierarchical perspective has been introduced (Figure 3). Most hierarchical models have a three-level structure. Here we distinguish the intellectual, perceptual-conceptual and sensorimotor levels of regulation (Frese & Zapf, 1994; Hacker, 1986; Hacker & Sachse, 2013; Zapf et al., 1992). HCI mainly uses the three-level model in a formulation by Rasmussen (1983) who labelled the levels “knowledge-based,” “rule-based” and “skill-based,” respectively. As these labels give the misleading impression that prior knowledge only plays a role at the first level, in the following, we use the more precise terms from Hacker’s (1986) action regulation theory.

Figure 3. Relationship between levels of regulation and functions based on Hacker (1986).

At the lowest level, the sensorimotor level (or skill-based in Rasmussen, 1983), the automatic implementation of movement sequences takes place with the help of kinaesthetic and proprioceptive feedback (Figure 3). Here, also non-physical automatic cognitive routines are regulated. Regulation processes at this level are triggered or interrupted by higher levels, although it is difficult to modify automatised action programs (Frese & Zapf, 1994; Hacker & Sachse, 2013; Zacher & Frese, 2018). Information processing is effortless and seemingly unlimited at this level; thus, the cognitive effort is very low (Frese & Zapf, 1994).

Next is the perceptual-conceptual level, often referred to as the level of flexible action patterns (Frese & Zapf, 1994; Hacker & Sachse, 2013; Zacher, 2017; Zacher & Frese, 2018) or rule-based level (Rasmussen, 1983). Action patterns can be conceptualised as schemata. Action schemata consist of ready-made action sequences that are learned through extensive experience or practice. The preparation for action at this level of regulation is therefore highly rule-governed and schematised, usually triggered by a few signals from the environment. Nevertheless, the schemata can be used flexibly in that they have control parameters that are specified according to the situation. Actions are prepared and controlled by processes that are capable of consciousness, even if they do not necessarily require consciousness. The cognitive effort at this level is intermediate.

At the top level, the intellectual level (knowledge-based in Rasmussen, 1983), the design, modification, selection and control of action plans for novel and complex actions, as well as the setting of a new action goal, takes place (Frese & Zapf, 1994; Hacker & Sachse, 2013; Zacher, 2017; Zacher & Frese, 2018; Zempel, 2003). The goals and action plans are implemented at the subordinate levels and, after a phase of practice, can be called up at lower levels without recourse to the intellectual level. Action regulation at the intellectual level is conscious and cognitive effort is high (Frese & Zapf, 1994; Zempel, 2003).

According to current research, however, it can also happen that no explicit, conscious goal formulation takes place. As goals are present in long-term memory and in the action-guiding mental models, goals can also be called implicitly including the boundary conditions of actions and action plans (Bargh & Gollwitzer, 1994; Cooper & Shallice, 2006; Dijksterhuis & Aarts, 2010; Hacker, 1986). Thus, people can also unconsciously pursue goals that are already stored in the mental model, if the goal representations are unconsciously triggered by behavioural or contextual information (Dijksterhuis & Aarts, 2010).

According to Hacker and Sachse (2013), the three levels of information processing are hierarchically or heterarchically nested: subordinate levels pass on their results in the form of feedback to the superordinate levels so that these are able to intervene or correct, if necessary. Higher levels, however, do not completely determine regulation at lower levels, because feedback from lower levels can also lead to changes at higher levels, e.g., the user changing the goal when an action cannot be executed (Frese & Zapf, 1994; Hacker & Sachse, 2013; Zacher, 2017; Zacher & Frese, 2018).

2.1.3. Implications for the understanding of intuitive use

There are several take-aways from action regulation for the theoretical conceptualisation of intuitive use. First, action regulation theory makes clear that action regulation is goal-centred. Reaching the goal as accurately and completely as possible is important to users and during their actions they closely monitor progress to adjust action plans and their executions according to the current situational circumstances. Thus, an essential requirement for measuring intuitive use is its effectiveness in reaching user goals. Although, some researchers explicitly leave effectiveness out of their definition of intuitive use (e.g., INTUI: Ullrich, 2014), most include effectiveness in their definitions (e.g, IUUI, GT) or in their proposed measures of intuitive use (QUT video coding). What is important to consider from the discussion above is that user goals may be intrinsic or extrinsic goals, and that these can arise consciously or subconsciously.

Second, revisiting the definitions of intuitive use, many require the “subconscious” (IUUI) or “generally unconscious” (QUT) application of prior knowledge. IUUI refer to the sensorimotor and perceptual-conceptual levels of action regulation. This includes the automatic unconscious regulation of action as well as regulation outside of conscious awareness. Similarly, GT in O’Brien et al. (2008, 2010) refer to “well-learned activities” and “determining what to do next” that we would map to regulation at the sensorimotor and perceptual-conceptual levels, respectively. The QUT group follows Wickens et al. (1998) and only calls regulation at the perceptual-conceptual level intuitive (i.e., Rasmussen’s rule-based level; Blackler, 2006), although their definition of intuitive use would also allow for regulation at the sensorimotor level.

In summary, we pose that only action regulation at the sensorimotor and perceptual-conceptual levels (i.e., Rasmussen’s skill-based and rule-based levels) can be called intuitive. Conscious processing at the intellectual level (i.e., Rasmussen’s knowledge-based level) cannot be called intuitive. Actions, however, always include all three levels to different degrees (Hacker, 1986, 2009; Hacker & Sachse, 2013). Furthermore, action regulation theory suggests that the lower levels are contained in the higher levels of regulation. If, for example, the initial goal formulation takes place consciously at the intellectual level, it triggers regulation at the lower levels (Hacker, 1986). To classify such an action sequence as non-intuitive, because of its conscious component might not be appropriate. Even in the case of completely conscious “unintuitive” information processing, at least the motor implementation may have unconscious components (Hacker, 2009). Thus, any reference to subconscious or unconscious processing appears to be insufficiently specified because the impression could arise that the upper level is explicitly excluded from intuitive use, which is not possible from an action regulation perspective. In practice, this inseparability of levels would not result in pure conscious versus pure unconscious processing, but in a summative “more or less”. Intuitive use is thus better understood on a continuum of values from low to high. This applies to whether one regards single interactions with particular interface features or whole action sequences initiated to complete a whole task.

Third, the hierarchical three-level perspective as a simplification has the problem that, unless a fourth metacognitive level is considered (e.g., Frese & Zapf, 1994), it does not make any statements about how the switch between the levels takes place during action regulation. It appears, this task is located on the middle level (Hacker & Sachse, 2013), but it is not satisfactorily specified. How metacognitive processes can mediate between the levels we elaborate when discussing the tripartite model of the mind below.

Instead of describing intuitive use as “unconscious”, it may be better referred to as taking place at “fringe consciousness”, as researchers working on intuition outside of HCI have described it (e.g., Mangan, 2015; Norman, 2017; Norman et al., 2010; Price & Norman, 2008; Reber et al., 2004; Zander et al., 2016). Thus, the basic prerequisite of intuitive use can be regarded as a predominantly unconscious cognitive information processing on the basis of action-relevant prior knowledge on the fringe of human consciousness. This area of fringe consciousness will be discussed in more detail below in the context of preattentive metacognitive processes.

2.2. Processing fluency as a metacognitive feeling with many correlates

O’Brien et al. (2008, 2010) have introduced three metacognitive feelings that influence how information processing advances at the levels of ‘well-learned activities’ and ‘figuring out what to do next’. Recent research in cognitive social psychology indicates that one metacognitive feeling might be sufficient to switch between different levels of processing, i.e., processing fluency.

Processing fluency describes the ease with which people process information. It consists of two dimensions. The first is objective fluency. It describes the mental efficiency of information processing as it can be objectively assessed. The second is the feeling dimension of fluency (i.e., subjective fluency). It is the subjective experience of effortlessness with which a person can process information (Reber et al., 2004). The subjective experience of processing fluency can result from a variety of sources: ease of perception, ease of memory encoding and retrieval, ease of linguistic processing at the phonologic, lexical, syntactic, and orthographic levels, as well as ease of reasoning. Processing fluency can be produced, for example, by concept priming and stimulus repetition (Alter & Oppenheimer, 2009). Design characteristics, e.g., symmetry, prototypicality, simplicity, contrast and clarity can influence processing fluency (Reber et al., 2004).

The relation between objective and subjective fluency is not as straightforward as might be assumed. Subjective fluency is not just the conscious reading of objective mental workload, for example. Subjective fluency depends on the anticipation of processing ease or difficulty and the interpretation of a fluency experience relies on past experience and the current context (Oppenheimer, 2008). Thus, it makes a difference whether objective or subjective fluency is measured, and whether it is directly measured as “ease of processing” or as one of its derivatives (see below).

Despite its different origins (e.g., perceptual, linguistic, conceptual), the subjective feeling of fluency is thought to be a single experience of ease or difficulty that can influence further judgements through cognitive heuristics. Cognitive heuristics are implicitly learned correlations between a feeling of fluency and its origins. Stimuli are fluent, for example, when they have been processed frequently, recently and/or for a long period of time. As a result, fluency becomes a useful cue for frequency, recency and duration of exposure. These again are cues for familiarity or social consensus (Alter & Oppenheimer, 2009). As there only is a single experience of fluency, it is difficult to know which source had been the most relevant and thus, by applying heuristics, a variety of judgment outcomes are likely (Alter & Oppenheimer, 2009; Oppenheimer, 2008). Thus, fluency, as a metacognitive feeling delivers summary information about cognitive processes and plays an important role in human judgement.

Previous research could show a variety of judgmental effects that are connected to fluency heuristics (Ackerman & Thompson, 2017; Alter & Oppenheimer, 2009; Morewedge & Kahneman, 2010; Price & Norman, 2008; Reber et al., 2002; Reber et al., 2004). According to these, the more fluent the mental processing feels,

• The more familiar people judge an object to be,

• The more people prefer and like the objects involved,

• The more beautiful these objects appear to people,

• The higher their confidence in their performance is,

• The lower people perceive their effort of learning,

• The more truthful people judge an information to be, and

• The more trustworthy a person or thing appears.

Fluency not only influences the outcomes of judgments, it is also involved in how information processing advances. A subjective feeling of fluency provides a positive affective impulse signalling that the current mental model is sufficient to cope with the situation and that there is no need to adjust it by using conscious processing. Fluency provides a global, ensuring and satisfying gut feeling, which provides a very condensed overview of non-conscious information processing and a strong action tendency without providing conscious access to the antecedents of that feeling (see Ackerman & Thompson, 2017; Gigerenzer, 2007; Hammond, 1993; Mangan, 2015; Price & Norman, 2008; Reber et al., 2002; Thompson, 2009; Thompson et al., 2011; Thompson & Morsanyi, 2012; Topolinski & Strack, 2009).

In the course of predominantly unconscious cognitive information processing (i.e., fringe consciousness), the feeling of fluency influences intuitive action, since fluency generates an affective directional impulse and thereby determines when the central executive switches between unconscious and conscious processing (Price & Norman, 2008). The feeling of fluency thus holistically reflects the depth of intuitive processing and the associated subjective satisfaction (Ackerman & Thompson, 2017; Price & Norman, 2008; Reber et al., 2002; Reber et al., 2004; Thompson et al., 2011, 2013).

Turning to the measurement of intuitive interaction, four conclusions can be drawn. First, processing fluency, conceptualised as objective and subjective mental effortlessness, also needs to be measured objectively and subjectively. Objective and subjective measures do not need to be the same. Fluency theory and research can hint at the links and dependencies between the two forms. Second, fluency as a metacognitive feeling is involved in determining the level of information processing. Thus, the level of fluency can reflect the level of unconscious information processing that is often declared central for intuitive use.

Third, the consequences of the subjective feeling of fluency can be manifold when fluency heuristics serve as a cue to other judgements. Comparing the various effects of fluency shown in social psychology studies with the different subjective measures proposed by previous research on intuitive interaction, we speculate that many subjective measures of intuitive use represent judgement effects derived from fluency heuristics. In the INTUI questionnaire, for example, the subscale effortlessness represents a direct measure of fluency. The subscales verbalisation ability, gut feeling and magical experience, we propose, are interpretations of the feeling of fluency by the user applying fluency heuristics. Depending on the context (e.g., entertainment vs. household systems) these interpretations then lead to different INTUI patterns (Ullrich & Diefenbach, 2011). Similarly, in the QUESI, the sub scale subjective mental workload is a direct measure of fluency, while perceived achievement of goals, perceived error rate, perceived effort of learning and familiarity report subjective interpretations (or correlations) of fluency. Although further research is needed to validate this conjecture, our proposal is that we need to distinguish between direct measures of fluency and derived measures that are subjective interpretations of fluency via fluency heuristics.

Fourth, the social psychological research on fluency suggests further derivatives of fluency, hitherto undiscussed in research on intuitive use (Alter & Oppenheimer, 2009; Winkielman & Cacioppo, 2001). As discussed above, these include aesthetics and beauty, safety and credibility—fields that have not yet been related with intuitive use.

2.3. Default-interventionist theories of information processing

The discussion of intuitive use as subconscious information processing (vs. non-intuitive use as conscious information processing) often refers to different dual processing theories from different branches of psychology. Since the 1970s, there has been a vast number of theories that consider human information processing from a two-process perspective. Such theories have been developed in cognitive psychology (e.g., Shiffrin & Schneider, 1977; Sloman, 1996; Stanovich, 1999; Stanovich & West, 2000), social psychology (e.g., Chaiken & Trope, 1999; Chen & Chaiken, 1999; Epstein, 1994), and decision psychology (e.g., Hogarth, 2001; Plessner & Czenna, 2008; Tversky & Kahneman, 1974), in the domains of problem solving (e.g., Epstein, 1994; Sloman, 1996), behavioural or attitude change (e.g., Chaiken, 1987; Chen & Chaiken, 1999) as well as in judgement and decision making (e.g., Evans, 2011; Hogarth, 2001).

Although the consensus seems to be that there are two different processes (rather than one or more), the resulting terminology is highly ambiguous (see Table 2). A summary description of intuitive actions as “unconscious” has been discussed as being problematic because unconsciousness is only vaguely defined and the associated characteristics can vary greatly (cf. Evans, 2010). Further, a dichotomisation of intuitive versus non-intuitive use based solely on unconscious vs. conscious processes can give the impression that intuitive actions do not allow for conscious processing at all and that an intuitive action cannot vary on a continuum between unconscious and conscious cognitive information processing (see above).

Table 2. Alternative terms for intuitive and reflective processing, found in various two-process theories (cf. Horstmann, 2012; Stanovich et al., 2014).

Two-process theory	Type 1 process	Type 2 process
Bargh and Chartrand (1999)	Automatic processing	Conscious processing
Bazerman et al. (1998)	Want self	Should self
Bickerton (1995)	Online thinking	Offline thinking
Brainerd and Reyna (2001)	Gist processing	Analytic processing
Chen and Chaiken (1999)	Heuristic processing	Systematic processing
Epstein (1994)	Experiential	Rational
Evans (1984)	Heuristic processing	Analytic processing
Evans and Over (1996)	Tacit thought processes	Explicit thought processes
Evans and Wason (1976)	Type 1 processes	Type 2 processes
Fodor (1983)	Modular process	Central processes
Gawronski and Bodenhausen (2006)	Associative processes	Propositional processes
Haidt (2001)	Intuitive system	Reasoning system
Hammond (1993)	Intuitive	Analytical
Johnson-Laird (1983)	Implicit inferences	Explicit inferences
Kahneman and Frederick (2002)	Intuition	Reasoning
Kahneman (2003)	System 1	System 2
Klaczynski (2000)	Heuristic processing	Analytic processing
Lieberman (2003)	Reflexive system	Reflective system
Loewenstein (1996)	Visceral factors	Tastes
Metcalfe and Mischel (1999)	Hot system	Cool system
Norman and Shallice (1986)	Contention scheduling	Supervisory attentional system
Pollock (1989)	Quick and inflexible modules	Intellection
Posner and Snyder (2004)	Automatic activation	Conscious processing
Reber (1989)	Implicit cognition	Explicit learning
Shiffrin and Schneider (1977)	Automatic processing	Controlled processing
Sloman (1996)	Associative system	Rule-based system
Smith and DeCoster (2000)	Associative processing	Rule-based processing
Stanovich (2005)	Autonomous processes	Decoupled simulation
Strack and Deutsch (2004)	Impulsive system	Reflective system
Thaler and Shefrin (1981)	Doer	Planner
Thompson et al. (2011)	Heuristic processing	Analytic processing
Toates (2006)	Stimulus-bound	Higher order
Wilson (2004)	Adaptive unconscious	Conscious

2.3.1. Characteristics of dual processes

The labels used by different theories (Table 2) usually refer to only one aspect, giving the impression that the process can be reduced to this attribute (e.g., automatic, associative, heuristic) and that only this attribute is sufficient to describe intuitive processing (Evans & Stanovich, 2013; Stanovich et al., 2014). Some neutral labels such as system 1 and system 2 (Kahneman, 2003; Stanovich, 1999) try to avoid this but suffer also from ambiguous terminology. The terms could imply that people have two separate systems for different types of cognitive processing that do not share the same cognitive resources and are distinct at the neurological level (Evans, 2008, 2011; Evans & Stanovich, 2013; Stanovich & Toplak, 2012). To counteract such misinterpretations, the literature now refers to Type 1 and Type 2 processes (Evans & Stanovich, 2013; Stanovich et al., 2014). These terms will also be used in the following.

Regardless of the specific labels, two-process theories agree that they assign distinct properties to each of the two processing types (Evans & Stanovich, 2013; Stanovich et al., 2014; Zander et al., 2016). The majority of authors characterise Type 1 intuitive cognitive information processing by fast, unconscious, automatic and parallel processing with a high processing capacity under low cognitive effort. In contrast, Type 2 reflective cognitive information processing is characterised by slow, conscious, controlled and sequential processing with a low processing capacity under high cognitive effort (Evans & Stanovich, 2013).

Evans and Stanovich (2013) summarise the associated characteristics of Type1 and Type 2 processes (Table 3). The two types of information processing are predominantly characterised by objective features (e.g., fast, low demand on working memory)—as they can also be found in the definitions of the QUT and IUUI research groups. Subjective consequences of intuitive use, as proposed in particular by the INTUI research group, were missing from this list, although a large number of theories emphasise the importance of affective information as preattentive metacognitive cues (e.g., Ackerman & Thompson, 2017; Hogarth, 2001; Kahneman, 2003; Reber et al., 2002; 2004; Reyna, 2008; Thompson, 2009). Since the associated preattentive affective directional impulse is realised with the help of Type 1 processes, affective information plays no role for Type 2 processes (Ackerman & Thompson, 2017; Horstmann, 2012; Thompson, 2009). Accordingly, we also included affective characteristics in Table 3, following Kahneman (2003) and Horstmann (2012).

Table 3. Characterising features of Type 1 (i.e., fully intuitive) and Type 2 (i.e., fully reflective) processes (cf. Evans & Stanovich, 2013; Horstmann, 2012).

Type 1 processes	Type 2 processes
Fast	Slow
Unconscious	Conscious
High processing capacity	Low processing capacity
Parallel	Sequential
Automatic	Controlled
Associative	Rule-based
Low demand on working memory	High demand on working memory
Contextual	Abstract
Affective	Non-affective
Independent of cognitive abilities	Dependent of cognitive abilities

2.3.2. How processes combine

Concerning the interaction between the two processes a distinction can be made between pre-emptive theories (only one type of processing occurs exclusively), parallel-competitive theories (both types of processing are simultaneously activated) and default-interventionist theories (Evans, 2007). The latter is discussed here.

Default-interventionist theories (Evans, 2006, 2011; Glöckner & Betsch, 2008; Kahneman & Frederick, 2002; Stanovich et al., 2014; Thompson, 2009) postulate that Type 1 intuitive processes are always activated first by default. Type 2 analytical processes can intervene in a second step, if necessary (Evans & Stanovich, 2013; Horstmann, 2012; Stanovich et al., 2014). Various phenomena in human decision-making, action regulation and associated information processing are best explained by default-interventionist theories, and they represent the current consensus in the psychological research literature (e.g., Evans, 2007, 2009; Evans & Stanovich, 2013; Frankish, 2010; Horstmann, 2012; Kahneman, 2011; Stanovich et al., 2014; Thompson et al., 2013).

Default-interventionist theories explicitly distinguish between the two types of information processing and the actual information processing that takes place during action regulation, which varies on a continuum requiring both types of processes (e.g., Evans, 2007, 2010; Stanovich, 1999, 2011). The cognitive continuum results from the possibility of Type 2 processes to intervene, if the situation seems to require it, and thus to take over cognitive information processing. An intuitive action is predominantly regulated by Type 1 processes, whereas a reflective action is predominantly regulated by Type 2 processes (Evans, 2010; Evans & Stanovich, 2013; Stanovich et al., 2014). In order to describe the dynamic interplay of Type 1 and Type 2 processes, the dichotomous distinction between Type 1 and Type 2 processes is no longer sufficient, which is why meta-cognitive processes must exist in addition to these processes.

2.3.3. From dual processes to a tripartite model of the mind

Default-interventionist theories assume a tripartite cognitive hierarchy of consciousness (Ackerman & Thompson, 2017; Evans, 2009; Evans & Stanovich, 2013; Stanovich et al., 2014). In the tripartite model of the mind, Type 1 processes form the autonomous mind and are also referred to as the autonomous set of systems (Stanovich et al., 2014). Within Type 2 processes, a distinction is made between an algorithmic mind and a reflective mind. The functions of the individual instances are shown in Figure 4. We also link these to the levels of action regulation from Hacker’s (1986) action regulation theory discussed above (Figure 3).

Figure 4. Tripartite model of the mind according to Stanovich et al. (2014), supplemented with corresponding levels of regulation from Hacker’s (1986) action regulation theory. See text for a description of the processes involved.

According to the tripartite model, two outputs are generated by the autonomous mind, which act as inputs to both the reflective and algorithmic minds (arrows A in Figure 4; Stanovich et al., 2014). The first output is the content of the initial action tendency based on the mental model, and the second is an accompanying metacognitive feeling of fluency in the retrieval of that action tendency (Ackerman & Thompson, 2017; Evans, 2010; Evans & Stanovich, 2013; Reber et al., 2002; 2004; Simmons & Nelson, 2006; Stanovich et al., 2014; Thompson, 2009; Thompson et al., 2011). Type 2 processes are triggered to intervene when Type 1 processes are perceived as non-fluent (Ackerman & Thompson, 2017; Barrouillet et al., 2011; Evans & Stanovich, 2013; Stanovich et al., 2014; Thompson, 2009).

The generation of these two outputs of the autonomous mind can be understood as the retrieval of long-term memory content. Fluency signals to the two upper instances the extent to which the long-term memory content is suitable for coping with the current situation (i.e., action-relevant prior knowledge is sufficiently accurate and differentiated). If the metacognitive feeling of fluency is strong, the autonomous mind can solve the problem on its own on the basis of the available mental model (i.e., prior knowledge was correct and differentiated enough). In this intuitive decision, the initial action tendency of the autonomous mind is accepted and no intervention occurs (Figure 5). A weak sense or absence of subjective fluency signalises the need for intervention and the need for changing the mental model to solve the action problem (Fazendeiro et al., 2005; Reber et al., 2002; 2004; Stanovich et al., 2014; Thompson, 2009; Thompson et al., 2011; Zelazo et al., 2007). In this case, the initial action tendency of the autonomous mind is inhibited (arrow G in Figure 4).

Figure 5. Preattentive metacognitive processes in intuitive use following the double-response paradigm of the tripartite mind (see Ackerman & Thompson, 2017; Thompson et al., 2011) and the fluency-affect model of intuition (see Topolinski, 2011; Topolinski & Strack, 2009).

A number of research papers consider that the feeling of fluency arises, for example, from the objective fluency of the implicit retrieval of long-term memory content at the fringe of human consciousness (Figure 5; Ackerman & Thompson, 2017; Price & Norman, 2008; Reber et al., 2002; Reber et al., 2004; Topolinski, 2011). Objective fluency is expressed in the form of an unconscious affective directional impulse, which leads to a metacognitive feeling of fluency (Ackerman & Thompson, 2017; Harmon-Jones & Allen, 2001; Reber et al., 2004; Thompson, 2009; Thompson et al., 2011; 2013; Topolinski, 2011; Topolinski & Strack, 2009; Winkielman & Cacioppo, 2001).

High objective fluency provides the metacognitive cue that, for example, the memory content has already been successfully retrieved and used to solve a similar action problem more than once and is therefore likely to be correct for the current action context (Reber et al., 2004; Thompson, 2009; Thompson et al., 2011; Whittlesea & Leboe, 2003). Objective fluency describes in condensed form the effortlessness or simplicity with which the initial intuitive response tendency comes to the autonomous mind (Alter & Oppenheimer, 2009; Reber et al., 2002; 2004; Simmons & Nelson, 2006; Thompson, 2009; Thompson & Morsanyi, 2012).

Yet, the feeling of fluency can convey the illusion that this content has already been experienced before, even if this is not the case (Jacoby & Whitehouse, 1989). Another source of fluency is the ease with which stimuli can be perceived (Alter et al., 2007; Alter & Oppenheimer, 2009). If perception is disfluent, people tend to rely more on analytical Type 2 processes. Although researchers distinguish several types (i.e., origins) of subjective fluency, no difference has been empirically established between these types, as these exert similar effects on human decision-making (Graf et al., 2018; Schwarz, 2004, 2015; Winkielman et al., 2003).

The different consequences of the feeling of fluency can be illustrated using the subscales of the QUESI:

1. The scale perceived mental load directly reflects the level of subjective fluency.

2. Fluency can manifest in a feeling of familiarity, which provides the user with a cue that the retrieved memory content is the content sought (Ackerman & Thompson, 2017; Metcalfe et al., 1993; Price & Norman, 2008; Schwarz, 2004). The strength of this feeling depends on the level of objective fluency (Graf et al., 2018; Schwarz, 2004; Schwarz et al., 2007).

3. The feeling of error describes the subjective experience that something has gone wrong and that one has made a mistake (Ackerman & Thompson, 2017; De Neys et al., 2011; Fernandez Cruz et al., 2016; Gangemi et al., 2015). The higher the fluency, the weaker this feeling (Gangemi et al., 2015). In QUESI, this is reflected in the subscale perceived error rate.

4. Similarly, a feeling of rightness is determined by fluency and describes the degree to which the first solution that comes to mind feels right (Ackerman & Thompson, 2017). In QUESI this is reflected as higher ratings on the subscale perceived goal achievement.

5. The feeling of knowing provides the user with a judgement about how probable a previously not yet recalled memory content is correctly recognised (Hart, 1965; Koriat, 2000; Metcalfe & Wiebe, 1987; Singer & Tiede, 2008). According to Price and Norman (2008), this feeling can be likened to the tip-of-the-tongue experience (Brown, 1991). It also involves an assessment of learning, which tells the person how likely a piece of content they have just learned is to be recalled correctly at a later time (Veenman et al., 2006). The higher the objective fluency, the better the feeling of knowledge (Koriat, 1993, 2000) and assessment of learning (Matvey et al., 2001; Mueller et al., 2013). In QUESI this is reflected as a rating on the subscale perceived effort of learning.

Depending on the strength of the feeling of subjective fluency (see arrow A in Figure 4), the reflective mind as a metacognitive monitoring process can determine the extent to which the response of the autonomous mind is satisfactory (Stanovich et al., 2014; Thompson, 2009; Thompson et al., 2011). A strong sense of fluency signals to the reflective mind that the action can be solved by the autonomous mind alone. Therefore, no analytical processing is necessary; the initial intuitive response tendency of the autonomous mind is accepted. A weak feeling of fluency acts as a trigger for executive processes causing mental load.

If the feeling of fluency is weak, the reflective mind instructs the algorithmic mind to override the autonomous mind’s action tendency. This ability is also referred to as cognitive decoupling, as it allows people to construct and coordinate mental simulations to represent diverse decision possibilities (Evans, 2007, 2010; Evans & Stanovich, 2013; Stanovich et al., 2014). In order to perform such cognitive decoupling, the reflective mind must first initiate the three basic executive functions shifting (arrow B), updating (arrow C) and inhibition (arrow D in Figure 4), and then have the algorithmic mind perform them.

For the response of the algorithmic mind, processing must be shifted from unconscious to conscious, i.e., information processing must be taken over by Type 2 processes, (arrow B in Figure 4). With the help of these Type 2 processes, a new mental model must be built up and a simulation must be initiated by the reflective mind on the basis of explicitly recalled long-term memory contents and acting out alternative answers for the underlying action problem (arrow C in Figure 4). This action requires the content of the working memory to be modified and updated. Finally, the initial action tendency of the autonomous mind can be overwritten and thus suppressed (inhibition, arrow D in Figure 4).

To solve the action problem, the algorithmic mind subsequently performs a cognitive decoupling either (1) completely (i.e., consciously), (2) partially (i.e., dynamically alternating between unconscious and conscious processes at the boundary of human consciousness) or (3) not at all (i.e., unconscious). The first type of cognitive decoupling (arrow E in Figure 4) implies the complete simulation and explicit construction of a mental model (Stanovich, 2009, 2011; Stanovich et al., 2014). This form of cognitive decoupling corresponds to the intellectual level of action regulation (Frese & Zapf, 1994; Hacker & Sachse, 2013; knowledge-based in Rasmussen, 1983).

The second type, which is also called serial associative cognition (arrow F in Figure 4), implies an explicit cognitive simulation that is only partially carried out, since a mental model is in principle available for solving the action problem (i.e., a user has prior knowledge, but this knowledge needs to be adapted). The correctness and differentiation of the model is not completely sufficient for unconscious processing, but in contrast to complete decoupling, fewer conscious processes of adaptation are required (Stanovich, 2009, 2011; Stanovich et al., 2014). In the context of action regulation (Frese & Zapf, 1994; Hacker & Sachse, 2013), this type of partial cognitive decoupling corresponds to the perceptual-conceptual (rule-based) level and thus to processing at the boundary of human consciousness.

Cognitive decoupling can also be dispensed with altogether and the initial action tendency of the autonomous mind can be accepted instead (arrow H in Figure 4). In the context of action regulation (Frese & Zapf, 1994; Hacker & Sachse, 2013), this type of cognitive decoupling corresponds to the sensorimotor level (skill-based in Rasmussen, 1983). For more details regarding cognitive decoupling see Stanovich et al. (2014).

The extent of cognitive decoupling required for solving an action problem can be determined by the basic executive functions involved (i.e., inhibition, updating and shifting), which implicitly underlie all other higher executive functions (Jurado & Rosselli, 2007; Miyake et al., 2000). The extent of cognitive decoupling can be read from the cognitive load on working memory during action regulation (Evans & Stanovich, 2013; Stanovich et al., 2014). The more cognitive decoupling is required to solve a problem, the higher the load on working memory. For this reason, default-interventionist theories postulate that the load on the working memory that results from cognitive decoupling alone is sufficient as a dichotomous objective characteristic to distinguish between Type 1 and Type 2 processes. The involvement of Type 1 and Type 2 processes in action regulation can be read off from measures of cognitive effort (Evans, 2011; Stanovich, 2011).

The extent of cognitive effort exerted is thus a distinguishing feature between Type 1 and Type 2 processes. The other features that various researchers have associated with the two-types distinction in Table 3 then are typical correlates of cognitive effort. The regulation of information processing based on the metacognitive sense of fluency causes variable cognitive effort and this reflects the associated dependence on Type 2 processes (Ackerman & Thompson, 2017; Markovits et al., 2015; Thompson, 2009; Thompson et al., 2011). As this is not equal to the subjective feeling of fluency, it needs to be measured separately.

In intuitive use, the autonomous mind signals a high level of fluency, minimising the need for cognitive decoupling and the associated cognitive effort. Only in unforeseen new situations (e.g., planning a new action, responding to unexpected feedback from the system) cognitive decoupling needs to be employed. In contrast, a non-intuitive action is characterised by a weak feeling of fluency, a higher need for cognitive decoupling and increased cognitive effort. Based on the feeling of fluency and the cognitive effort exerted, an action can be more or less intuitive on a continuum, which can be shown by measuring these variables directly or in form of their correlates.

3. Defining core features and correlates of intuitive use: a new measurement definition

After reviewing the different perspectives on intuitive use and embedding these into recent developments in cognitive, work and social psychology, we are able to formulate the defining characteristics of intuitive use with the goal of measurement and evaluation.

The first requirement is high effectiveness. It derives from action regulation theory that assumes that all actions are goal-directed. Goals can be pursued consciously or subconsciously and they can also be more abstract goals (O’Brien et al., 2010) on the motivational or needs level (e.g., need for social affiliation, need for entertainment; Desmet & Fokkinga, 2020). Thus, if goals are not fulfilled when interacting with a system, we cannot speak of intuitive use. This may be seen as a basic pragmatic feature of intuitive use and has been a characteristic of previous definitions and measures of intuitive use (e.g., QUT, IUUI, GT).

The second requirement is low cognitive effort. This is derived from the discussion of dual-process theories, particularly of the stance of Evans and Stanovich (2013) who reduced the available lists of characteristics of Type 1 and Type 2 processes to defining characteristics and their typical correlates. These lists have many characteristics in common with what HCI researchers have defined as characteristics of intuitive use (e.g., fast, subconscious, implicit). Evans and Stanovich (2013) see as the core defining features of Type 1 processes that they do not require working memory and that they are autonomous (i.e., require no cognitive control). Hence, in the tripartite model of the mind, the working of the autonomous mind leads to a low level of cognitive effort, and when the responses of the autonomous mind need to be overridden, cognitive effort is high.

While autonomous processes need sophisticated laboratory experiments to be detected, low cognitive effort that directly follows from autonomous processing has a long history of measurement in HCI (Young et al., 2015). Cognitive effort, or mental workload, is understood as the used amount of working memory resources. When information processing is relayed to Type 2 processes of the reflective or algorithmic minds, processes like updating, shifting and inhibition induce working memory load, hence cognitive effort that can be measured using a variety of techniques. Low cognitive effort, associated with Type 1 processing, then corresponds to intuitive use.

The third requirement is a strong metacognitive feeling of fluency. Most HCI researchers employ a subjective component when measuring intuitive use. In line with the tripartite model of the mind and research on processing fluency we propose that subjective fluency is underlying the many subjective measures of intuitive use. Subjective fluency accompanies Type 1 processing and is used as a metacognitive indicator influencing the type of further processing. When the feeling of fluency is high, there is no necessity to involve Type 2 processing, thus, indicating intuitive use.

In summary, the defining features of intuitive use are its high effectiveness, low cognitive effort and a strong metacognitive feeling of fluency. Cognitive effort is measured objectively; the feeling of fluency is measured subjectively. Typical correlates of low cognitive effort (e.g., fast, unconscious, experience-based, automatic interaction) and of strong subjective fluency (e.g., feelings of familiarity, of rightness, of certainty) can be used as proxies or additional information, but are not themselves necessary features to describe intuitive use. Thus, direct measures should be preferred.

We assume that all three defining features need to be achieved together to claim intuitive use. No single measure of effectiveness, cognitive effort or subjective fluency can fully represent overall intuitive use. Thus, intuitive use is not complete when only two characteristics are fulfilled (e.g., low cognitive effort and high subjective fluency with low effectiveness would not be considered intuitive use). The relative importance of the three criteria, however, depends on the specified context of use and the reason why a researcher is considering intuitive use. Thus, there is no general rule for how single measures should be chosen or combined. It is only necessary to use at least one measure for each of effectiveness, cognitive effort and subjective fluency.

4. Discussion

The new measurement definition of intuitive use and the theoretical account we give have several implications for measuring intuitive use that are discussed in the following. We also discuss the limitations of our approach and the resulting opportunities for future research.

4.1. Implications of the new measurement definition

The new measurement definition of intuitive use has a number of implications, summarised in Table 4. We will discuss these in the light of previous work.

Table 4. Implications of the new measurement definition of intuitive use (overview).

	Implication	Insight/examples
1	Reduces the number of previously proposed measures of intuitive use to measures of the three defining characteristics.	High effectiveness, low cognitive effort and a strong metacognitive feeling of fluency
2	Allows to differentiate previously used measures into measures of defining features and typical correlates.	Temporal efficiency, verbalisability, many subsclaes of the QUESI or INTUI questionnaires, for example, are typical correlates, not defining features of intuitive use.
3	Allows to search for other operationalisations of the defining characteristics not used before in intuitive interaction research.	Physiological measures of cognitive effort, the Single Ease Question and other operationalisations in the literature on processing fluency come into view.
4	Differentiating the defining characteristics of intuitive use from typical correlates sheds new light on differences between research groups.	Differences often arise from looking at different conditions under which typical correlates of intuitive use operate.
5	Summative and formative evaluation is supported.	While quantitative measures allow for summative evaluation, their continuous measurement can identify points of failure during interaction which are important for formative evaluation.
6	Intuitive use varies on a continuum.	There is no pure intuitive or unintuitive use, neither in single operations nor in whole interaction sequences.
7	Clarification, integration and updating of underlying psychological theory.	Extending dual processes with the tripartite model of the mind. Inclusion of processing hierarchies in the model. Clarifying the role of processing fluency.

First, the new measurement definition helps to reduce the large number of recently applied measures of intuitive use to a small number of defining characteristics:

1. Effectiveness. According to ISO 9241-11, effectiveness is the “accuracy and completeness with which users achieve specified goals” (ISO, 2018, p. 3). Effectiveness is typically measured objectively via success rates, error rates or the number of helping hints needed in a usability test. Objective effectiveness is currently a measure in QUT video coding, the CHAI coding procedure, and other evaluations (e.g., Macaranas, 2013). Although effectiveness can be measured subjectively as, e.g., with the QUESI subscales perceived goal achievement and perceived error rate, these ratings could be biased by a strong sense of fluency without users objectively achieving their goals. However, if the users’ goals are experiential, e.g., having fun or experiencing suspense, then subjective measures of attaining such goals are appropriate measures of effectiveness.

2. Cognitive effort. Cognitive effort here refers to the used amount of working memory resources. Cognitive effort is higher the more Type 2 processing occurs, involving the activation of the algorithmic and/or reflective minds. It is low when Type 1 processing by the autonomous mind prevails. All classic methods of objectively measuring cognitive effort may be considered. In intuitive use research, IntuiBeat has been employed, for example, as a secondary task measure (Reinhardt, 2020).

3. Feeling of fluency. Subjective fluency is the subjective ease with which information is processed. In intuitive use research it has been measured including questionnaires like SMEQ, but also specific subscales of dedicated questionnaires can be used as operationalisations, e.g., the subscale effortlessness of the INTUI or the subscale subjective mental workload of the QUESI. (Note that the NASA TLX is too broad to cover cognitive effort alone.)

Second, many of the currently proposed measures of intuitive use are, according to the new measurement definition, typical correlates of intuitive use, but not defining features (Table 5). Measures of constructs like high temporal efficiency (QUT), low explicit verbalisability (QUT, INTUI), or unconscious processing (IUUI, QUT) are correlates of cognitive effort. By reducing the characteristics of intuitive use to just the one objective characteristic of cognitive effort, ambiguities and controversies in measuring and interpreting the unconscious application of prior knowledge, low explicit verbalisability or the high temporal efficiency of cognitive information processing can be avoided.

Table 5. Defining characteristics and correlates of intuitive use according to the new measurement definition of intuitive use.

Defining characteristics of intuitive use	Typical correlates of intuitive use
High effectiveness	--
Low cognitive effort	Predominantly unconscious cognitive information processing based on prior knowledge that is relevant to action
	Time-efficient cognitive information processing
	Low verbalisation of information processing
Strong metacognitive feeling of fluency	Perceived low mental demand of cognitive information processing
	Perceived high achievement of goals
	Perceived low temporal demand on cognitive information processing
	Perceived high familiarity
	Perceived low explicit verbalisability
	Perceived low effort of learning
	Perceived low error rate
	Feeling of Knowing, of Directness, of Progress
	Magical experience
	Gut feeling
	Intuitiveness

Regarding subjective measures, many of the subscales of the INTUI or QUESI questionnaires that are not directly linked to subjective fluency can be seen as typical correlates of subjective fluency. The same goes for the NASA TLX overall score. Again, hitherto unused measures come into view; for example, the SUPR-Q (Sauro, 2015) which measures standard fluency outcomes, as established in the fluency literature: subjective usability, trust, appearance/attractiveness and customer loyalty intention (cf. Alter & Oppenheimer, 2009; Cyr et al., 2006; Im & Ha, 2018; Schwarz et al., 2021), but have not been proposed as a measure for intuitive use.

While the defining characteristics of intuitive use help to operationalise the essential characteristics, measuring the typical correlates of intuitive use can nevertheless provide important information. Correlates will be useful to measure, for example, to determine the consequences of fluency and its respective judgments via fluency heuristics. Our take is that although these measures should not be confused with the defining characteristics of intuitive use, they are useful to predict important outcomes of intuitive use or can be used as proxies when the essential characteristics cannot be measured.

Third, pointing out the defining characteristics of intuitive use allows to search for other operationalisations that have not yet been employed in intuitive interaction research. Physiological measures, for example, seemingly a standard in human workload studies, could be used to objectively measure cognitive effort. Relevant physiological parameters are heart rate variability calculated from electrocardiograms (ECG), event related potentials in brain activity (ERP), electrodermal activity (EDA), or, in ocular activity, blink rate, fixation length and pupil diameter (for reviews see Charles & Nixon, 2019; Cowley et al., 2016). Although early experimentation with physiological data to measure intuitive use (EDA and heart rate) led to “disappointing” results (Blackler, 2006), it is possible that newer instruments would be more reliable. Similarly, the Single Ease Question (SEQ; Sauro & Dumas, 2009) is a direct measure of subjective fluency that has not yet been employed in intuitive interaction research. It consists of a single item with a semantic differential response format: “Overall, this task was…” very easy (coded as 1)/very difficult (coded as 7). Also, looking into the psychological literature on processing fluency will reveal alternative operationalisations of subjective fluency (e.g., Graf et al., 2018; Kostyk et al., 2021).

Fourth, differentiating the defining characteristics of intuitive use from typical correlates sheds new light on the potential controversies between research groups. While fast interaction, for example, is important to the QUT research group, this is rejected by the IUUI research group. The new measurement definition can clarify this: speed of interaction is not a defining feature, but a typical correlate of intuitive use. Speed is likely to indicate Type 1 processing when looking at the latencies of movement onset, for example. When considering task completion times, however, speed may be dissociated from intuitive use. Installing an application using the command line may be fast but mentally demanding, whereas following a step-by-step installation wizard is slower but less mentally demanding. Further, when people trade off speed against accuracy, speedy interaction jeopardises effectiveness as a characteristic of intuitive use.

Distinguishing between defining features and typical correlates may also account for the observation of INTUI-patterns. The INTUI subscale effortlessness measures subjective fluency as a defining feature of intuitive use. The other subscales verbalisability, gut feeling and magical experience then are typical correlates. As there is no perfect correlation between the judgments of fluency and these other scales, the correlation between these differs according to the situation or product used (Ullrich, 2014).

Fifth, this new measurement definition of intuitive use can be used as a basis for the summative evaluation of intuitive use, that is a mostly quantitative assessment to compare between systems and/or benchmarks and to take stock whether a design project has reached its goals. The new measurement definition can also be a good basis for the formative evaluation of intuitive use, which deals with detecting and solving problems that impair intuitive use. This entails, for example, identifying the critical moments during interaction in which there is a need for cognitive decoupling and how this manifests in an increased demand on working memory and a weak sense of fluency. Identifying these moments via continuous measurement can give the evaluator important clues to usage problems and how to solve them. Then, the evaluator can obtain qualitative feedback from the user at such points in time to narrow down on the source of the problem (cf. the differentiation in a summative and formative evaluation mode when using IntuiBeat in Reinhardt, 2020).

Sixth, as action regulation proceeds during the interaction with a system, Type1 and Type 2 processing will be intermixed and vary at different time-points. There will unlikely be pure Type 1 or pure Type 2 processing during the interaction with a product. Likewise, intuitive use will vary on a continuum and the extent of intuitive use can be measured by the extent of effectiveness, cognitive effort, and the feeling of fluency throughout the interaction with a system. Measures can be chosen by the researcher as appropriate—whether they are interested in the intuitive use of single operations or whole interaction sequences.

Seventh, our treatment of the underlying theory clarifies, integrates and updates previous accounts of information processing in intuitive use. By alluding to the subconscious or unconscious application of prior knowledge many previous accounts refer to a dual-process view of intuitive use. By looking into integrative work conducted in psychology, we could show that these dichotomies correlate with other proposed characteristics of intuitive use (e.g., fastness, automaticity). Following Evans and Stanovich (2013), we propose cognitive effort as the primary distinctive feature (instead of subconciousness). To include the levels of information processing often referred to in the previous literature, we saw the necessity to go beyond dual-process theories and proposed a tripartite model of the mind. In this model, metacognitive feelings of fluency play an important role in determining whether the autonomous, the algorithmic and the reflective mind co-operate. Adapting the model of Stanovich et al. (2014) we could thus integrate, in the formulation of Hacker (1986), the skills-rule-knowledge hierarchy by Rasmussen (1983), which many researchers in intuitive use refer to (QUT, GT, IUUI). We propose that intuitive use can be mainly found at the skill-based and rule-based levels, while QUT, for example, would constrain intuitive use to the rule-based level only (Blackler, 2006). We feel, however, that our view is more coherent with the dual-process distinction by Evans and Stanovich (2013) and is compatible with other research groups like IUUI and GT.

Similar to GT we place metacognitive feelings as important for shifting the type of information processing, but while GT propose three metacognitive feelings, we propose that one feeling of processing fluency is sufficient, based on research in social cognitive psychology that shows that although the feeling of fluency can feed on several sources (perceptual, memory retrieval, linguistic), it does not depend on these sources in further processing. According to the subjective interpretation of this feeling of fluency involving fluency heuristics, a number of other subjective characteristics of intuitive use can be measured as corelates using questionnaires, e.g., perceived error rate and familiarity (QUESI), gut feeling and magic experience (INTUI). Further, we have grounded the effectiveness measures that were often proposed by intuitive use researchers (e.g., IUUI, QUT) in action regulation theory, because we believe intuitive use is goal-seeking behaviour, while goals can be pragmatic or more experience-focussed as GT has pointed out and phenomenological approaches aim to deal with in more elaborate ways (e.g., INTUI).

Altogether, we deem our biggest contribution, therefore, is the definition of three defining characteristics of intuitive use that are open to different operationalisations according to the specific context of application. Thus, our theoretical approach and new measurement definition offer a synthesis and extension of previous theories of intuitive use proposed by different research groups and newer theories and findings in cognitive, social and work psychology.

4.2. Limitations

First, our new measurement definition clearly focuses on the measurement (not the design) of intuitive use and on the measurable consequences (not the preconditions) of intuitive use. Further work would be necessary to review routes into designing for intuitive use. HCI researchers have emphasised that intuitive use is based on the subconscious application of prior knowledge, as it can be found centrally in the definitions of the IUUI research group (e.g., Hurtienne, 2011) and the QUT research group (e.g., Blackler, 2006). A number of design principles can be derived from the prior knowledge criterion such as designing for user expectation with familiar features and using a consistent appearance, location and functionality of features in the user interface (Blackler, 2006). Another way is to design for knowledge structures that can be transferred from universally available world-knowledge, for example, population stereotypes, image schemas, and conceptual metaphors (see Blackler et al., 2019; Hurtienne, 2011, 2017). Design approaches using prior world knowledge can be found in the Computers Are Social Actors paradigm (Reeves & Nass, 1996), tangible interaction (Ishii, 2008; Shaer & Hornecker, 2010), natural user interfaces (Wigdor & Wixon, 2011) and Reality-Based Interaction (Jacob et al., 2008). Yet despite all these approaches it seems difficult to compile a complete catalogue of what knowledge structures are universally available and how to assess these. There are first approaches to measure single aspects of prior knowledge. There are the Technology Familiarity Questionnaire by Blackler (2006), Asikhia’s Q (2015) based on image-schemas and observational measures such as the QUT categories clear expectation stated or relation to past experience. Each of these, however, covers only a small part of the relevant knowledge and its salience for intuitive use in any given moment, and much of the relevant knowledge depends on the user’s prior experience and the context of use. Therefore, it is difficult to derive necessary and sufficient lists of criteria based on prior knowledge, either for designing intuitive use or for deriving valid measures that can universally predict intuitive use.

Second, we developed the new measurement definition of intuitive use out of theories of action regulation and information processing taken from the broad areas of cognitive, work, and social psychology. Thus, this work sits firmly in the second paradigm of HCI with a focus on cognitive psychology, experimental work and quantification, taking a third-person perspective on interaction and aiming at general applicability (Harrison et al., 2007). With an emphasis on situated and phenomenological perspectives in the third paradigm, the evaluation of intuitive use could also be understood as focussing more on the experiential and qualitative aspects, and aiming at understanding interaction from a first-person perspective. While the discussion of intuitive use has not yet fully arrived in the third paradigm, it is the INTUI group that highlights the experience side of intuitive use (Diefenbach & Ullrich, 2015; Ullrich, 2013, 2014). Yet, they also ground their approach in information processing theories of judgment and decision making. With their concept of INTUI patterns, however, they offer a more situated view. Depending on the interactive systems and their specific contexts of use, different experiential qualities become salient over others. Our new measurement definition reduces dimensions such as magical experience and gut feeling to correlates of subjective fluency, but they could be further developed into an own phenomenological understanding of intuitive use.

Third, while our theoretical framework captures relatively well the main strands of previous research on intuitive use, there is also a smaller strand that sees intuitive use not just as Type 1 processing, but would also include trial & error interactions that lead to fast learning. O’Brien et al. (2008, 2010) see Type 1 processing as only one form of intuitive use (performing well-learned activities). The second type is “determining what to do next.” This has more of a trial-and-error character, but in lenient learning environments is also deemed central for intuitive use. Spool (2005), from a practitioner’s perspective, allows that a “knowledge gap” between the user and the system can be closed with minimal cognitive effort, and Baerentsen (2000) allows for spontaneous learning in intuitive use. Translated into the framework of the tripartite mind presented above, it seems that intuitive use can also be considered when there is what was called ‘partial decoupling’ or action regulation at the rule-governed perceptual-conceptual level, where Type 1 and Type 2 processes intertwine. The idea of seeing intuitive use on a continuum between more and less rather than yes and no and the focus of our new measurement definition on the outcomes of intuitive use can account for this. Thus, for example, we are looking at the extent of cognitive effort indicating the amount of Type 2 processing (and decoupling) that occurs. We would expect higher intuitive use the more Type 1 processing occurs, but as processes involving partial decoupling will only use limited amounts of Type 2 processing, intuitive use might still be high in many cases. Learning may improve this situation further when new information becomes automatically processed over time. Further, implicit, that is effortless, learning would also be visible as a gradual increase in intuitive use. Further research could thus also look at the steepness of learning curves as another measure of intuitive use that has not been included in our framework. The performance differences between primed and unprimed use as proposed by Fischer et al. (2015b) might be seen as a first step into this direction.

Fourth, we did not build this research on a systematic literature review (SLR), but on the experience of one of the authors being the founder and long-term member of one of the research groups plus extensive research experience in the field by both authors. This, to us, seemed permissible for three reasons: (1) our primary interest was in the critical examination of theories and measures as well as their conceptual integration. Thus, we understand our approach as a literature review according to APA (2020, p. 8), not as a meta-analysis (cf. Boell & Cecez-Kecmanovic, 2015). (2) The term “intuitive” is often used in vague and fuzzy ways sometimes denoting merely a positive feature of the interaction (e.g., Turner, 2008; Wennberg et al., 2018). Although a standardised keyword search would produce many papers, only a small fraction of these would further the theoretical discussion. (3) Some theoretically important works would not have been found via a standardised database search, e.g., the working paper by O’Brien et al. (2010) or the dissertation by Ullrich (2014). Nevertheless, a quick search on Google scholar, using the search terms “theory ‘intuitive interaction’” showed that our review seemed to have caught the most important papers. Among the first 100 search results, 61 papers directly belonged to one of the four groups reviewed or to authors of “Other Groups”; a further 19 papers used intuitive interaction in a way that cites one or more of these groups or is compatible with their frameworks. Eleven articles used the term “intuitive interaction” as a rhetoric device but did not present or reference a definition of the term; five articles were outside the field of HCI; and for three articles a full text could not be obtained to clarify their status. One article was new and was concerned with phenomenological aspects going beyond those proposed by the INTUI group (Kaltenbacher, 2008). Future work could include an SLR employing more and finer search terms as well as more elaborated inclusion and exclusion criteria to validate the claims made in our review.

4.3. Areas of future work

We see at least three further areas of future work with regard to the theoretical framework and new measurement definition proposed here.

First, as we have derived the framework from previous work in HCI, many of the core assumptions nevertheless stem from research and theory in cognitive and social psychology, e.g., the tripartite model of the mind and the metacognitive role of processing fluency. One large area of future work in HCI will therefore be to empirically validate the framework and the proposed categorisation of certain characteristics of intuitive use as what we have termed “typical correlates”. One question is whether the subjective feeling of fluency can hold as the core common denominator for the range of diverse constructs from preference and liking to credibility and trust in the domain of user interfaces. A second question, discussed above, is whether certain aspects, e.g., the phenomenological side of intuitive use, needs to be further strengthened in the theoretical framework. A third question is to derive and validate new measures for the three components of intuitive use and make these applicable for the summative and formative evaluation of intuitive interaction. Finally, it will be important to determine the conditions under which the correlates of intuitive use indeed show their validity and when they do not.

Second, we have already discussed the need for further work on the preconditions of intuitive use, especially the areas of prior knowledge that are relevant to designing (and measuring) intuitive use. A number of approaches (e.g., technology familiarity, image schemas) already exist in that area that have not yet been exhaustively explored. But other modules of the autonomous mind may be interesting to exploit for designing intuitive interactions. Stanovich (2011) calls these The Autonomous Set of Systems that together comprise the autonomous mind. This set includes perceptual and linguistic modules, processes of behavioural regulation by the emotions and, in general, many processes of phylogenetic origin and of evolutionary adaptive value like face recognition and mate choice. We have only begun to understand the impact of these on design. Research fields such as persuasive interaction and consumer psychology may help us in transferring related psychological findings into application.

Third, the strong role processing fluency plays in the theoretical framework and in deriving typical correlates of intuitive use warrants to initiate a greater cross-talk between research into intuitive interaction and processing fluency. Apart from looking at the different types of fluency (e.g., perceptual, motor, memory retrieval, linguistic) and the many correlates of fluency that have not been considered in intuitive interaction research (e.g., credibility, trust, customer loyalty, aesthetics) an exchange with fluency research also offers theoretical and methodological benefits. Current fluency research looks at the circumstances under which fluency is correlated with other measures. The effects of fluency were shown to be influenced by mood (Koch & Forgas, 2012), lack of control (Blair, 2020), time dynamics (Cox & Cox, 2002; Landwehr et al., 2013) and the naive theories people hold about the correlates of ease of processing (Greifeneder & Bless, 2018), and it is unclear how domain-dependent such effects are. Fluency research also develops theoretical explanations of how fluency translates into its typical correlates. Examples include the Pleasure-Interest-Model of Aesthetic Liking (Graf & Landwehr, 2015) or whether fluency works as simple hedonic marker or amplifier of other effects (Landwehr & Eckmann, 2020). There is also research on measuring fluency e.g., physiologically (Topolinski et al., 2009) or subjectively (Graf et al., 2018) that could also provide useful inspirations for measuring the fluency component of intuitive interaction. Finally, there is research on algorithmically estimating the processing fluency of visual stimuli that can inspire similar approaches in HCI (Mayer & Landwehr, 2018).

5. Conclusion

In this article we have worked towards a new measurement definition of intuitive use. For this we have reviewed previous definitions and measures of intuitive use in HCI and the relevant research on action regulation and cognitive processing in current work, cognitive and social psychology. The new measurement definition is able to distinguish what currently appears as a multitude of definitions and measures into defining features and typical correlates of intuitive use. The defining features of intuitive use are high effectiveness, low cognitive effort and a strong metacognitive feeling of fluency during interaction. The new measurement definition points out what is important to consider; it is applicable to formative and summative evaluations, and it opens up new areas for research in HCI including taking a closer look on processing fluency research in psychology.

The new measurement definition is limited by a focus on the outcomes of intuitive use. It does not address the preconditions like the application of prior knowledge and other modules of the autonomous mind. The new measurement definition is thus meant to primarily be of use for evaluators of intuitive use, not so much for those looking for inspiration on designing for it. Our hope is that this work opens up new possibilities for further research on intuitive use in HCI by providing a new platform for the ongoing debate. Promising future areas are to develop a more phenomenological account of intuitive use, to further the cross-talk with psychological research on processing fluency, to empirically validate the proposed distinction between defining characteristics and typical correlates of intuitive use—as well as to explore the conditions under which these correlations occur.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Notes on contributors

Daniel Reinhardt

Daniel Reinhardt is senior user experience researcher at Kaiser X Labs in Munich, Germany. He holds a PhD in Human-Computer Interaction from Julius-Maximilians-Universität Würzburg, Germany. His research interests include the theory and evaluation of intuitive use, tangible and embodied interaction.

Jörn Hurtienne

Jörn Hurtienne is full professor in Psychological Ergonomics with Julius-Maximilians-Universität Würzburg, Germany. His research interests include design for intuitive use, designing with image-schematic metaphors, tangible interaction, embodied cognition and user experience design.