The Fast and the Curious: Creative Improvisation as Action-oriented Abduction

Abstract

Improvisation, understood as a skillful activity, suggests a tension between the constitutive roles of habit and creativity. Cognitive and phenomenological accounts of skill acquisition point to the importance of developing habitual responses that facilitate fluid performance. Yet, habit formation is associated with repetitive behavior. Seemingly at odds with repetition, spontaneous invention is often regarded as the basis of improvising creatively. The puzzle – how habit and creativity relate in skillful improvisation – is resolved by drawing on a view of the brain as an embodied, predictive organ, entangled in cultural scaffolding, and perpetually engaged in inference. This view is in part rooted in the pragmatist tradition, as is a large body of work on music. By combining strands of theoretical and empirical research on the embodied, predictive, creative, and musical mind, an account of action-oriented abduction is developed and applied to an illustrative vignette. The resulting application goes beyond existing accounts to articulate an enriched sense of creative improvisation.

Introduction

Improvisation is commonly associated with musical performance, although the concept is also applied to other activities, including painting and even ordinary conversation (Sawyer, 2000a). This paper focuses on musical improvisation, but draws on both music-specific and domain-general research to sharpen the contours of a puzzle. Specifically, improvisation suggests a kind of fluidity when it is done skillfully. Such skill implies that learning has taken place, with accompanying processes of automatization and habit formation. But (and here's the puzzle), habits are strongly associated with repetitiveness, as exemplified by the phrase ‘force of habit’. Skilled improvisation, on the other hand, is typically associated with openness and creativity. ¹ What then is the relationship between habit and creativity in skilled improvisation?

We think this puzzle is important. It draws attention to something deeply significant, yet poorly understood: the capacity to produce behavior that is both fast and fluent, yet flexible, variable, and in some sense ‘creative’. The puzzle also raises interesting questions about the relationship of creative improvisation to the threshold of conscious awareness. To unpack this puzzle, we first review a range of theoretical and empirical literature, drawing together a set of core elements. We incorporate these elements into a novel theory of creative improvisation that we designate as action-oriented abduction (discussed in detail later on). After developing the theory, we apply it to a vignette that foregrounds the key insight provided by our approach, showing how it extends beyond extant theoretical formulations of creative improvisation. Moreover, by accommodating traditionally extramusical dimensions in a novel way, the theory we develop and apply also extends beyond the typical confines of adjacent work on embodied music cognition (e.g., Cox, 2016; Hubbard, 2017; Leman et al., 2018). Finally, we suggest some implications and considerations for future investigations of creativity in improvisation.

There is a large and growing literature on musical improvisation (e.g., Clarke & Doffman, 2017; Lewis & Piekut, 2016; MacDonald & Wilson, 2020), especially in the humanities and social sciences, but also in the life sciences (Tachibana et al., 2019; Vergara et al., 2021; Walton et al., 2015a; for a review, see Beaty, 2015). Neurocognitive and other biologically centered experimental studies focus mainly on specific aspects of improvisation (spontaneity, coordination, self-evaluation, etc.) by setting up tasks that can help isolate neural correlates or behavioral patterns that differentiate improvisation from other modes of musical engagement. By necessity, empirical research of that kind deals with observable components such as contextual and formal constraints, generation and perception of tonal and rhythmic organization, communication among co-performers, the effects of skilled expertise on the brain and on performance, and so on. Rarely do such studies begin with a comprehensive conceptual account of improvisation. As a result, there is now an assemblage of concepts linked to improvisation through these studies, but which cannot necessarily be seamlessly integrated with one another.

In this paper, we aim to elaborate a theoretical foundation that integrates empirical findings, while dissolving the conceptual tension between habit and creativity in skilled improvisation. To do so, we will use the lens of a recently emerging theoretical perspective that treats the brain as an organ of embodied prediction. This perspective shares common roots with a wide range of studies in music psychology and music cognition, namely, in the tradition of pragmatism, as we describe next.

Music, Mind, Brain, and Body: Pragmatist Roots

While we focus on improvisation, it is worth noting the broader theoretical foundation that grounds our proposal, in which music, mind, brain, and body are intertwined. Predictive accounts of mind and brain are increasingly becoming the norm in cognitive and computational neuroscience (e.g., Clark, 2016; Friston, 2010; Hohwy, 2013; Knill & Pouget, 2004; Parr et al., 2022; Rao & Ballard, 1999). Their representation is also growing in perceptual, psychological, and neuroimaging studies of music, which we refer to broadly as ‘biocognitive’ (e.g., Friston & Friston, 2013; Koelsch et al., 2019; Rohrmeier & Koelsch, 2012; Vuust & Witek, 2014). These biocognitive approaches contrast with a previous period of cognition-oriented but non-biology-based applications of probabilistic analysis to music (Temperley, 2001/2005, 2007/2010). Contemporary predictive accounts of mind and embodied cognition have been shown to strongly overlap with or even derive directly from the pragmatist tradition (Engel et al., 2016; Madzia & Jung, 2016; Williams, 2018). A predictive, embodied biocognitive approach has also been linked simultaneously to pragmatism and music (Schulkin, 2013, 2021).

Broadly speaking, pragmatism is a tradition of thought that, among other properties, suggests that the beliefs we adopt “directly or indirectly produce behavioural or epistemological consequences” (Almeder, 2007, p. 172); social values and history put these consequences into perspective, providing a standpoint for evaluating beliefs alongside considerations of truth or objectivity. ‘Classical’ pragmatism – through its primary figures, John Dewey, William James, and C.S. Peirce – pervades the relevant musical literature even though it is not often singled out therein as a philosophical tradition.

We highlight a few significant ways that Dewey, James, and Peirce are woven into musicological studies, even when unacknowledged. To begin with, Dewey heavily informs Meyer's (1956) influential work on musical meaning and emotion. Meyer in turn informs the equally influential work of Narmour (1990), Huron (2007), and others who deal with musical expectancy.

The notion of attention, nearly ubiquitous in contemporary neuroscientific perspectives on music, stems from James (1890; Bisley & Goldberg, 2010). He is also in part responsible for ideomotor theory (James, 1890; Lotze, 1852), found as well throughout the literature on musical performance (e.g., Koch et al., 2004; Stöcker et al., 2003; Waszak et al., 2005). Similarly, James (1884) introduced the theory that emotions relate to bodily self-perception, along with Lange (1885), now known as the James-Lange theory of emotions. Although famously critiqued by Cannon (1927), the James-Lange theory is still referenced throughout the musical literature on emotion and mood (e.g., Garrido, 2017; Huron, 2007).

Peirce is commonly cited in the context of musical semiotics, in connection with questions regarding perception and mental activity (Cumming, 2000; Curry, 2017; Turino, 1999). Peircean semiotics, in particular, i.e., his theory of how meaning is conveyed, has been used to analyze affective or emotional aspects of music (e.g., Dowling & Harwood, 1986). Turino nicely sums up the way in which a conception of Peircean musical semiotics relates to culturally situated embodied cognition in general:

Peirce developed a theory of signs to understand how people are connected to, and experience, the world. Whereas [Saussurian] structuralism used language as the primary modelling system, Peirce defined the concept of sign in the widest, most flexible way as something that stands for something else to someone in some way, thus allowing for many different types of signs outside propositional language. (Turino, 1999, p. 222)

The analysis of multi-faceted experience that permeates both aesthetic contexts and everyday life is a strand that runs through all the major pragmatists (e.g., Dewey, 1934; see Johnson & Schulkin, 2020).

Notably, however, Peirce's notion of abduction is rarely invoked in research on music (for exceptions, see Moroni et al., 2005; Oliveira et al., 2010; Schulkin, 2013), even as it is found across a variety of general accounts of creativity (e.g., Anderson, 1987; Gonzalez & Haselager, 2005; Harnad, 2007). To our knowledge, abduction has not been linked specifically with musical improvisation in the manner we develop in this paper. ² Yet there is one instance of a connection made between the concepts of improvisation and abduction in a discussion of general creativity that also includes references to music; namely, in Ingold's opposition of the two concepts:

I am concerned to reverse a tendency, evident in much of the literature on art and material culture, to read creativity ‘backwards’, starting from an outcome in the form of a novel object and tracing it, through a sequence of antecedent conditions, to an unprecedented idea in the mind of an agent. […] To trace these connections is to perform the cognitive operation of abduction. […] [However,] the creativity of the work lies in the forward movement that gives rise to things. To read things ‘forwards’ entails a focus not on abduction but on improvisation (Ingold, 2010, p. 10)

The account we develop below takes a different tack, and ultimately reconciles this opposition of concepts. (We return to Ingold's opposition later on.)

A Puzzle: Habit and Creativity in Skilled Improvisation

Before turning to habit, we first consider in what respects creativity relates to improvisational skill. Someone who can improvise in the style of bebop saxophonist Charlie “Bird” Parker (frequently invoked in both creativity and improvisation research) would be clearly considered skilled, but what about creative? Boden (1990/2004) draws a relevant distinction here between psychological and historical (p- and h-) creativity. This allows the claim that creative processes might be involved in the mind, even if the results of the process (such as a musical performance) do not constitute a new contribution in sociohistorical terms. The distinction thereby facilitates a way to recognize creativity when someone improvises a new ‘Bird-style’ solo in the bebop idiom. In other words, reproducing a historical style could still be (p-)creative, as long as it relies on mental faculties other than reproducing a memorized sequence of notes.

One of the main ways in which researchers have dealt with idiomatic stylistic improvisation (e.g., emulating Bird's style in a bebop performance) is by identifying what could be called first-order melodic phenomena – either by positing an intrinsic generative musical grammar (Lerdahl & Jackendoff, 1983/1996) or by observing extrinsic melodic or harmonic transitions (e.g., Johnson-Laird, 2002; Norgaard et al., 2013). We call this kind of creativity ‘first-order’ because it relates to producing patterned sequential structures, even if these can be formally represented as ‘n-order’. (An n-order process is just a way of representing regularity, which can be the regularity of a ‘surface’ sequence modelled in relation to structurally ‘deeper’ sequences.)

Using an analogy frequently made between music and language, we note that speaking freely can still be regarded as a first-order sequence of words, even if the word sequence has a deeper n-order structure (whether grammar-based or relative to corpus statistics). Moving beyond the first order, what we would term further ‘higher order’ dimensions of the word sequence would include phonetic aspects (e.g., alliteration), semantic aspects (e.g., conceptual association), or broader semiotic aspects (e.g., allusion). The first-order sequence of words could be embedded in different contexts, such as a conversation at a bus stop or on a theatre stage. The word sequences could be spoken with different accents, or among groups with or without relevant background knowledge. These dimensions point to alternative sets of higher-order regularities, even given a common (first-order) base sequence of identical words (Linson, 2016).

Whether in spoken conversation or musical improvisation, a significant amount of practice is required for a first-order sequence of units to be produced fluidly. Fluid extemporizing, as when giving an impromptu speech, seems to fit the notion of (p-)creativity. Clearly, the ability to give such speeches is also regarded as a skill that benefits from practicing, which unavoidably instils habits. How could habitual actions be creative acts?

Below, we develop the argument that the cognitive basis of improvisation is not reducible to habits. This argument is at odds with the influential position in writing on music that suggests that improvisation is indeed mere habitual action, and therefore cannot be creative. That position is advocated by prominent figures including composer John Cage, who famously held that improvisation is best understood as recitation from memory, always familiar and never leading to new experience (see, e.g., Turner, 1990; see also Lewis, 2007, pp. 363ff.).

Rhyme Without Reason? Non-Random Behavior and Going with the Flow

One of the earliest contributions to empirical brain research on musical improvisation (Bengtsson et al., 2007) seeks to dissociate improvisation from recitation (reproduction from memory) using an analysis of neural activation patterns. Their research in part draws on theories of creativity in the psychological literature, such as divergent thinking. Divergent thinking relates to freely generating counterfactual thoughts and actions (Campbell, 1960). Counterfactuals can play a role in finding multiple ways of solving a problem rather than converging on a single correct response.

A relevant interesting finding by Bengtsson et al. (2007) is that improvisation paired with the effort to memorize the improvised melody (for subsequent performance from memory) did not fundamentally differ in neuronal activation patterns from improvising without any memorization requirement. Yet both conditions differed from the activation patterns observed when performing the melody from memory. Crucially, they also found an important neuronal difference between improvisational and random behavior that suggests a neural contribution to ‘sculpting’ (Frith, 2000) the coherence of what is being performed. This is opposed to coherency being purely a property of the observed result. (Coherence is not used here in the sense of formalized musical structure, but rather, as one might describe coherency of thought or behavior.)

The sculpting of coherence does not preclude bottom-up sensitivity to remain responsive. Moreover, a role for top-down processes does not suggest that anything has been planned explicitly. The execution of a memorized plan is a general form of the performance of a memorized melody; the latter is differentiated from improvising in the above-referenced neuroimaging study. An alternative appeal to a different sense of planning that can emerge improvisationally from a top-down and bottom-up dynamic interplay is found in the notion of planning as inference (Botvinick & Toussaint, 2012; Kaplan & Friston, 2018), which we return to later on.

A study of minimal cognition in improvisation (Linson et al., 2015) provides an example of coherency being purely a property of the observed result, rather than part of an active neural contribution that we are fleshing out here. In that study, an artificial agent performing with human improvisers relied on their collaboration to establish coherence, as it lacked its own mechanism to achieve this. The agent could be understood as ‘off-loading’ musical-behavioral organization onto the human improvisers, in the absence of any higher-order computational analysis in its system architecture. At the same time, the agent was highly responsive to its human co-performers in a ‘bottom-up’ fashion, without any recourse to stored sequences (i.e., no ‘recitation from memory’). As judged by the experts who performed with the agent, a coherent human-computer improvisation – rather than random behavior – emerged from most of the interactive performances in the study. This suggests that the improvisations were shaped by the skilled experts’ ability to dynamically establish coherence. A precondition for such coherent interaction in a free improvisation was mutual human-agent responsiveness (see Konvalinka et al., 2010), but not mutually memorized plans or material.

Another early neuroimaging study on musical improvisation (Limb & Braun, 2008) takes a closely related approach to Bengtsson et al. (2007). Limb and Braun (2008) identify in musical improvisation an “absence of central processes that typically mediate self-monitoring and conscious volitional control of ongoing performance” (p. 1), in contrast to (e.g.,) performing a composition (they use etudes as a control for the study). Thus, their study similarly supports the idea that musical improvisation does not need to rely on a memorized plan. They also connect their findings to more general psychological research on creativity, in particular, on the notion of spontaneity. As they point out, spontaneous artistic activity in general is “frequently described as occurring in an altered state of mind beyond conscious awareness or control” (Limb & Braun, 2008, p. 1), which is befitting of other art forms that are not commonly referred to as improvisation, such as automatic writing in the Surrealist tradition. (The theme of automaticity recurs below in terms of a creative cognitive ‘mode’.)

Spontaneous musical performance that is coherent rather than random clearly relates to improvisation, but it is less clear why it should relate to creativity. At the same time, if a musician were to spontaneously produce a coherent series of notes that is improvisational, yet perceived by an audience as ‘uninspired’, it would be consistent with Boden's notion of p-creativity (and our description above of creative first-order sequences). This resonates with the position that “some improvisations are relatively boring, repetitive, and structured; others are inspired, fresh, and original” (Sawyer, 2000b, p. 181).

If skilled improvisation is consistent with creativity, how does it relate to habit? A preliminary answer is offered in the notion of ‘flow’, which refers to a feeling of timelessness and effortlessness in a challenging activity that involves a full range of embodied cognitive skills (Csikszentmihalyi, 1990). Flow states have been described in terms of neurocognitive mechanisms (Dietrich, 2004) that have been linked to musical improvisation (Landau & Limb, 2017; McPherson & Limb, 2013). Dietrich (2004) suggests that flow is related to implicit knowledge, habit, and skilled performance more generally. Dreyfus (2002) integrates some phenomenological and neuroscientific accounts of skilled performance with a focus on fast, intuitive maneuvering, for example, in chess: “Excellent chess players can play at the rate of 5 to 10 s[econds] a move and even faster without any serious degradation in performance. At this speed they must depend almost entirely on intuition” (Dreyfus, 2002, p. 372). His description of ‘intuition’ seems indicative of flow states, especially related to the use of implicit knowledge described by Dietrich (2004). Both point out a widely held view that mental appeals to explicit knowledge inevitably slow down performance (even given gradations between implicit and explicit; see, e.g., Clark, 2016; Schulkin, 2020). The overcoming of such slow-downs is central to Sudnow's (2001) widely cited phenomenological account of developing improvisational skill on the piano.

A related take – weaving together embodied cognition, flow states, habit and skill – is found in Sutton et al. (2011). While critical of some aspects of Dreyfus’ theory of expertise (see also Torrance & Schumann, 2019), they share his view that “fast and rapidly-changing dynamic domains like open-skill sports or improvisatory jazz make intellectualist approaches particularly hard to credit” (Sutton et al., 2011, p. 87). They formulate an approach they call Applying Intelligence to the Reflexes (AIR) which, as its name suggests, is concerned with intelligence; they do not address creativity directly.

Similarly, in an application of AIR to expert musical performance, Geeves et al. (2014) address neither improvisation nor creativity directly. However, they provide a general description that is near to, and in key respects in agreement with, the account of creative cognitive phenomena in skilled improvisation that we are progressively elaborating here:

The AIR approach proposes that expert skill in music performance is constituted by a dynamic responding that involves retrieval of the most effective (combination of) learned material given the unpredictable, contingent contextual demands with which a musician is faced during performance, and the integration with and expansion of this material in line with these demands. The AIR model posits that … an expert's attention is freely allocatable to addressing whichever variables, at whatever level, present themselves as most important in a given moment. This, as opposed to attention being so automated that it becomes stuck to predetermined variables within a particular level of encoding. (Geeves et al., 2014, p. 682)

Their conceptual framework (retrieval of learned material, level of encoding) differs significantly from our own, but the general picture of dynamic attention allocation remains roughly compatible with our approach.

A Curious Thing

Although chess is not quite a rapidly-changing dynamic domain in the way a ball game or musical performance is, it does provide a bridge from the idea of divergent thinking (described above as a form of creativity) to what is known in chess as the ‘unexpected move’, indicative of creative play. Kind (2022) counts being unexpected or surprising alongside being spontaneous as a core aspect of general creativity. If we treat habit as being underpinned by implicit knowledge – as when someone learns to play a musical instrument with proficiency – we can recognize that some proficient players can enter into a flow state; in such a state, a player can produce a spontaneous-yet-coherent performance that is not a recitation from memory, and that can sometimes lead to surprising results. This could be viewed initially as a reconciliation of habit and creativity in skillful improvisation – but it is not yet robust enough to withstand a counterexample.

Certainly, the kind of skill required for getting into a flow state amounts to an accumulation of habits, on the one hand. On the other hand, maintaining a flow state also entails a high degree of responsiveness to whatever is encountered. Such fluid performance can lead somewhere unexpected; that is, it can produce surprises. However, it appears exactly this kind of scenario can unfold during what could be called wandering. Assuming it is apt to call wandering a form of improvising, does it warrant the designation of creativity?

One can draw on experience (habit, skill) to wander fluidly around a new city and maintain spontaneity, responding to anything that crosses one's path. If this leads to a new route though the city that no one has taken before, it could even be regarded as h- and not merely p-creative. But this kind of ostensibly creative act still falls short of the enriched notion of creativity under discussion. Peirce's (1974, 6.452–465 [1908]) concept of ‘musement’ is close to wandering, in that it is designated as a purposeless activity that is enjoyable and playful. (We return to the concept of ‘play’ in section "Getting into the Swing of Things".) He does regard musement as creative, in contrast to his notion of inquiry, which is hypothesis-driven – while also indicating that novel hypotheses can be formed during musement (Cooke, 2018).

Crucially, the account of improvisation being developed here does not square neatly with Peirce's concept of musement, because an important aspect of creativity in musical improvisation is, as Landau and Limb (2017) put it, “producing music that ‘works’” (see also Loui, 2018, which refers to the highest computational system goal as “successful musical improvisation”). We regard this kind of improvisational production as being curiosity-driven, combining implicit hypothesis formation with hypothesis testing (Gregory, 1980), described further in the next section. Spontaneity, responsiveness, and hypothesis formation in improvisation could be tied to musement; however, in the midst of improvisation, with the tight coupling of perception and action in a flow state, novel hypotheses are not only formed, but also tested: ‘What will happen if I do that? Will it work?’ Thus, improvising creatively as defined here combines aspects of musement and inquiry.

Curiosity-driven creativity arises in both improvisational and non-improvisational activities. A scientist may be trying to develop a new measurement technique, inspired by some established fact, such as the flickering frequency of a star. They continue to draft and revise equations and diagrams, laboring to view their selected problem space from different vantage points, until they are struck by a solution. Such a solution is clearly a creative application of some set of ideas that could exemplify both p- and h-creativity. Some would call the result a creative ‘leap’, in that it seems to bring about a non-incremental change to the status quo. While such a scenario can unfold in the absence of improvisation, an improvisation can involve a similar kind of creative leap.

Creative ‘Leaps’: Predicting the Unexpected

To understand the notion of implicit hypothesis above, research on active vision is illustrative. As Friston et al. (2012) describe, saccades are a form of evidence-gathering that relates to counterfactual hypotheses. For example, in recognizing a face, one may guess before the face is fully taken in that it is either familiar or unfamiliar. These competing hypotheses are explored by saccades that sample different regions of the face, which provide evidence for one of the two possibilities. Eventually, a sufficient accumulation of evidence for an answer will thereby reduce the demands on further sensory sampling.

A common example given for a language-based scenario of this kind is predictive text in mobile devices. The software is constantly formulating a hypothesis about what the speaker will type next. When they type exactly what is expected, it is by definition unsurprising. When surprise is understood as an information-theoretic quantity, we can speak of just how much or how little surprise there is based on the number of expected possibilities. For example, if someone says, “I’ve got a visitor this weekend …”, a listener might narrow the scope of possible next words to, say, a family member, with a concrete prediction of ‘sister’. Hearing ‘cousin’ instead will be more surprising than hearing ‘sister’, but far less surprising than hearing ‘Martian’. As the latter is not in the category of family members, it is far more improbable.

Continuing with the above example, consider that in psychological creativity research, flexibility and persistence have been identified as key components of creative cognition (Nijstad et al., 2010). The two aspects have been investigated using experiments related to concept categories, where ‘within-group fluency’ – e.g., iterating the category of family members with ‘sister’, ‘cousin’, etc. – is identified with persistence, and flexibility is reflected in category-switching: e.g., family members vs. fictional creatures. These senses of flexibility and persistence have been linked to corresponding neural substrates (Boot et al., 2017).

Category-switching, also related to divergent thinking, is a metaphorical form of a creative ‘leap’. If making such leaps indicates flexibility, then being in a creative mode – or mood – can be framed as predicting the unexpected. That is, one is poised to pursue or respond to highly improbable outcomes.

Kronsted and Gallagher (2021) bring out a related point with respect to dance improvisation, that they also link to (what amounts to) divergent thinking in problem-solving, such as reframing a math problem to find a new approach to a solution. They discuss a dance improvisation training exercise in which one is asked to “dance with” an inanimate object like a box or table. As a way to improve dance improvisation, the exercise “involves actively trying to bring out new affordances in the engagement between one's body and objects by shifting one's attention toward features of the object that are normally nonsalient to oneself” (Kronsted & Gallagher, 2021, p. 48). This attentional shifting yields moves that are otherwise improbable. Their description fits nicely with the above-mentioned notion that “an expert's attention is freely allocatable to addressing whichever variables, at whatever level, present themselves as most important in a given moment” (Geeves et al., 2014, p. 682). But, as Kronsted and Gallagher (2021) highlight, in improvisation, there is an additional active, exploratory twist.

(Good Old-Fashioned) Abductive Inference

Flexible category-switching can include a further kind of leap in which a third category comes to mind that is more general than the initial two, moving beyond their dichotomy. To continue with the above ‘weekend visitor’ example, perhaps the choice is not between family member and fictional creature, but instead expands to a narrative with a character that fits in a more general grouping of both categories at once (‘my cousin, the Martian’). Putting this in terms of a hypothesis that jumps from an ‘either/or’ to an ‘and’ – ‘there is a fictional creature that is also a family member’ – connects with what Peirce describes under the concept of abductive inference, also known simply as abduction. More precisely, Peirce is concerned with explaining an unexplained (or unsatisfactorily explained) phenomenon by using the imagination to bring together divergent elements in what Peirce calls ‘colligation’:

the different elements of the hypothesis were in our minds before; but it is the idea of putting together what we had never before dreamed of putting together which flashes the new suggestion before our contemplation. (Peirce, 1974, 5.181 [1903])

The new suggestion, that is, the novel hypothesis yielded in this flash of abduction, can arise in the sciences as well as the arts, with a key difference. In scientific abduction, there are additional facts – perhaps yet to be arrived at through new experiments – that constrain the novel hypothesis. In artistic abduction, the hypothesis also arises when bringing together elements not previously brought together in an external manifestation – say, an idea for a sculpture that merges a particular emotional quality with a particular formal one – but is not subject to the same kind of constraints as science (Anderson, 1987; Barrena & Nubiola, 2019). Instead, the artist's hypothesis is measured against what could be referred to as the ‘success’ of the realized work, even when defined differently by the artist and by other observers or audiences (the public, critics, fellow artists, commissioners, etc.). This brings us back to a point introduced above, that skillful improvisers aim to make music that ‘works’ (Landau & Limb, 2017; Loui, 2018).

If there is a role in creativity for predicting the unexpected by having flexible and environmentally responsive ideation and attention allocation, there is equally a role for the skilled judgment of whether a creative leap will lead to something that ‘works’. Anderson (2005) provides a Peircean way of linking these roles that echoes the active vision example above: “the inquirer's relation to ideas is in some ways analogous to perception of the world” (Anderson, 2005, p. 19). But crucially, “the good abductive reasoner must be receptive to ideas and be able to recognize their fittingness – their plausibility – in addressing the problem at hand” (Anderson, 2005, p. 19). In other words, learning from experience is vital to increasing one's grip on what might work – however improbable and risky – and what probably will not. In the arts, the improbable hypothesis arising in abduction thus relates to risk-taking in the sense of pulling off a feat, materializing an unlikely outcome; ‘I (or we) can make it work’.

Furthering these connections between abductive inference, perceptual hypotheses, curiosity, and flashes of insight is work tying them all together with formal neuronal process theory (Friston, FitzGerald, et al., 2017). In simulations that correspond to functional neuroanatomical mappings, following an initial drive to sample novel cues, a form of Bayesian model selection and reduction occurs that is akin to forming and testing hypotheses until one is confirmed, bringing about a “reduction in the latency of evoked responses following the acquisition of insight” (Friston, Lin, et al., 2017, p. 2667). That is, beyond mere repeated exposure (for instance, to a stimulus sequence), a sudden flash of insight that explains the generation of the sequence subsequently reduces latency in evoked neuronal responses, which can be measured electrophysiologically or in terms of behavioral reaction times.

In the example presented in Figure 1, simple rules illustrate the underlying principle that when the ‘hidden’ pattern is identified, it will alter the estimation of the most probable model that captures the sequence (see Romeijn, 2013). The new basis for estimation will increase responsiveness over and above associative learning (Friston, Lin, et al., 2017; see also Wagner et al., 2004). The discovery (or invention) of such a hidden pattern – i.e., abductive inference – certainly requires curiosity, but does not require speed, even if the insight itself comes in a flash. Repeated observation through experience allows the gathering of sensory evidence. But in a creative mode, such observation includes an exploratory attentiveness that permits the ampliative (additive) contribution of an imagined pattern that can integrate ideas with ‘facts’ about the material world. Curious, imaginative exploration allows the forming and testing of hypotheses about possible integrations (‘what imagined patterns assimilate the observations?’).

Figure 1.

Illustration of pattern completion possibilities. The patterns indicate listening to a clapped rhythm, as follows. Each row (A-D) is an independent sequence synchronized with the second-hand on a clock (column progression). Clock ticks with a clap are shown as a filled circle; clock ticks without a clap (a silent ‘rest’) are shown as an empty circle. Consider a fictive study for illustrative purposes: in a training period, a subject is played one of the four sequences with only a corresponding letter displayed. In a testing period, they are presented a letter and the corresponding clap sequence is heard up until t0 (vertical dotted line). At that point, from t1 onwards, they must clap along. A subject may be highly capable of memorizing the sequence associated with each letter, but the latency of their response can be expected to be reduced once they learn (abductively infer) a simple rule that corresponds to each letter; namely, A: 4 claps, 3 rests (repeat); B: 4 claps, 3 rests, 2 claps, 1 rest (repeat); C: 4 claps, 4 rests, 3 claps, 3 rests; D: 4 claps, indefinite rests.

The foregoing explanation of abduction can be read as Peirce's understanding of what mental faculties are involved (Anderson, 2005), or as a basis for a hierarchal Bayesian formalism that can model the neuronal processes behind perception and action (Friston, Lin, et al., 2017). In the latter, when the moment of insight occurs, also called the ‘a-ha’ or ‘eureka’ moment – in other words, the creative leap under discussion – an operative manner of assimilating experience (understood as a model) is suddenly replaced by an alternative. That is, a once improbable ‘organizing principle’ of observations crosses a threshold to become the most probable, amounting to a sudden and unexpected re-organization. (The foregoing example and description will be concretized with an application to musical improvisation in section "Creative Improvisation as Action-oriented Abduction").

We have not yet brought abduction into a positive correlation with improvisation. One interesting aspect of creative improvisation is the interplay between what takes place in and out of awareness, such as being surprised by a self-generated action. In other words, events may be consciously improbable, but unconsciously probable. This apparent contradiction could in part explain why improvisation is sometimes described as being beyond volitional control. A different understanding, however, is offered by the conceptual framework of predictive processing, as demonstrated in implementations of active inference. Explicit hypothesis formation and testing are mirrored formally by their implicit counterparts below the awareness threshold (broadly equivalent to personal and sub-personal levels, respectively). While there is a symmetrical description of explicit and implicit cognitive processes, empirically, these processes operate at different time scales. Alongside the implicit processes that support explicit awareness, others continue independently in parallel, especially in response to conditions that are changing faster than conscious registration can accommodate.

Indeed, a number of widely recognized properties of consciousness (Seth et al., 2005) summarized below appear to support the aspects of creative improvisation we identify here, relating to processes both above and below the awareness threshold. Given evolutionary considerations, it would be unlikely for these properties to occur naturally in the absence of at least primary consciousness. However, it may be feasible to artificially replicate the processes underpinning improvisation without necessarily engendering consciousness.

Of the properties listed by Seth et al. (2005), informativeness appears to be a criterion for the exogenous orienting of attention, which would allow improvisers to incorporate (or electively discard) any sensory data from the surrounding environment that is potentially relevant to ongoing action. At the same time, taken together, the rapidly adaptive and fleeting nature of conscious scenes, internal consistency and limited capacity and seriality could explain why some improvisational action unfolds in ways that seem to defy accurate self-report (the standard behavioral index for consciousness in humans). Specifically, if sensory data cannot be assimilated with internal consistency, it may not enter awareness, while nevertheless being informative enough for the attentional modulation of unconscious sensorimotor processes (Sumner et al., 2006). In such cases, perceptual awareness may track a salient event, thereby excluding from awareness a competing perception that still influences immediate motor initiation.

To sum up thus far, in trying to give substance to the idea that habit and creativity are reconciled in skilled improvisation, we determined a role for habit in developing fluid performance, including habits of action and habits of perception. These habits relate to fluency on an instrument, in an idiom, in interpersonal coordination, and the corresponding know-how to best allocate attention to serve action. We also determined that getting into a flow state indicates rapid responsiveness, with fast and implicit handling. Yet, a wandering improvisation in a flow state does not necessarily entail enriched (p-)creativity, even if it may lead to surprising and novel (h-creative) results, similar to uttering a random sequence of words extemporaneously. In contrast to wandering, exploration driven by curiosity can lead to creative leaps, but that kind of creativity could be the culmination of a slow process that is not necessarily improvisational.

There remains a still outstanding need in the present account, to link the fast flow processes to the curious creative ones. To meet this need, we coin the notion of action-oriented abduction, which we connect to musical improvisation. We draw on literature not normally pressed into the service of describing musical improvisation, building on the idea of planning as (active) inference (Botvinick & Toussaint, 2012; Kaplan & Friston, 2018) and a hierarchical Bayesian account of musical percepts (Friston & Friston, 2013). But as a final step before unpacking these connections, it will be helpful for us to build up the context – the scaffolding – in which action-oriented abduction can take place.

Getting into the Swing of Things

We’ve considered a fast, wandering flow that could be construed as improvisational, but is not necessarily creative. We’ve also seen how curiosity can drive spontaneous exploration of the imagination and the world towards abductive surprises. The latter can be understood as creative, even if not necessarily improvisational. To integrate these two strands, we turn to how experience builds up our networks of mental associations in ways that affect our predictive neuronal machinery, based on overlapping and nested contexts. Key to this is the way our constructed environments scaffold and alter the scene statistics that predictive brains adapt to, and that in turn shape our actions and interactions (see, e.g., Clark, 2016, pp. 277–9). These environments include the many purpose-built worlds in which we learn and play – indeed, the full range of human culture, including (importantly) the arts. Here, we are particularly interested in continuities between play and art.

We previously used an example of the relation of habit and skill to describe fluency with a musical instrument as a basis for fluid, rapid responsiveness. An instrument is of course a human artifact, on which various techniques are learned, developed, and passed on. Once one becomes a trained instrumentalist, the neuronal patterns evoked in musical listening are transformed. Specifically, motor connectivity is enhanced along with greater motor and somatosensory integration than non-musicians, especially related to the upper limbs and torso (Alluri et al., 2017) – the bodily regions that tend to be heavily relied on for most conventional musical instrument performance. Even further sensitivity in listening to a specific instrument is made possible by expertise on that instrument, in contrast to other instruments (Margulis et al., 2009), which further emphasizes the specific statistical relationships in perception and action that, once learned, affect future experience.

A musical instrument, however, is just one among many different patterned elements of sociocultural scaffolding. For example, expertise in a musical idiom such as bebop affects listening to music in that idiom, revealing different responses to regularities than experts in other musical idioms (Hansen et al., 2016). Moreover, collaborative performances with other musicians requires attuning to regularities for coordination (Keller et al., 2014; Novembre et al., 2014; Walton et al., 2015b). Just as with the instrument example, undoubtedly more experience with specific players will result in improved coordination with those players, and likely translate into marginally improved coordination with unfamiliar players.

All these elements (and more) comprise designer environments, including those of play and art. In a recent predictive mind account of play, Andersen et al. (2022) propose that “play is experientially associated with a feel-good quality because the agent is reducing significant levels of prediction error (i.e., surprise) faster than expected” (Andersen et al., 2022, p. 2). This suggests that play is a practice that opens up a simulated world in which different regularities are salient, compared to the world outside of play. For example, driving a car on city streets requires being attuned to various patterns of road signs, other drivers, dashboard indicators and so on, for safely reaching a desired destination. On a racetrack, other patterns will be salient, and still others when one is controlling a car avatar in a videogame. Playing the videogame might offer ‘safe’ surprises, like a sudden obstacle that appears in one's path without the threat of physical harm, and for which skillfully performing a successful avoidance maneuver is enjoyable.

In the arts, enjoyable recovery from surprises (i.e., assimilation of experience following a reduction of prediction error) has been associated with physiological phenomena such as ‘chills’ and ‘goosebumps’ (Schoeller, 2015). These phenomena have been investigated in musical listening contexts, and are thought to arise from expectation violations in different aspects of musical regularity such as harmonic sequences or orchestration (Grewe et al., 2005; Guhn et al., 2007; Sloboda, 1991). Two points are worth highlighting here. The first is that, when understanding these chills-inducing moments as expectation violations, “it must be noted that such ‘expectations’ are, of course, dependent on a listener's enculturation, i.e., the listener's familiarity with a musical style and its conventions” (Guhn et al., 2007, p. 482). And the second is that “subjects often report the musical structures related to chills as being extraordinarily pleasant” (Grewe et al., 2005, p. 448). So, a clear connection between play and art can be understood in the context of learned statistical patterns among the scaffolds of designer environments, and how we react to and resolve swells of uncertainty with enjoyment.

Now's the Time: Active Inference and Creativity

Sensitivity to possible-yet-improbable future directions benefits not only from past experience, but from the ability to freely select from the past elements that form possible future trajectories, i.e., extending forward from a subset of the past, through the present, to a possible future. As Schacter et al. (2007) put it:

Given the adaptive priority of future planning, we find it helpful to think of the brain as a fundamentally prospective organ that is designed to use information from the past and the present to generate predictions about the future. Memory can be thought of as a tool used by the prospective brain to generate simulations of possible future events. (Schacter et al., 2007, p. 660)

This is a far cry from the narrow conception of uncreative, inflexible recitation from memory that has been misleadingly identified with improvisation.

Simulated futures, however, have at least two pitfalls. They are computationally (and thus metabolically and/or temporally) expensive, and, they have no guarantee of transpiring, given resistance from the world. These issues can be addressed simultaneously, continuing within the conceptual framework of active inference.

A musical performer cannot guarantee that they know what will happen next, even in light of extensive rehearsal. An instrument (reed, string, etc.) could fail unexpectedly, or one's own body may fail to respond as expected, e.g., due to fatigue or a suddenly arising injury (lip, finger, etc.). As Peirce notes, abductive insight is “extremely fallible” (Peirce, 1974, 5.181 [1903]). Nevertheless, experience allows one to form beliefs about what will happen next, and to act accordingly, until some new observation thwarts that belief (Kunde et al., 2004; see also Keller & Koch, 2008). In this respect, a player may form a relatively long-range and coarse-grained expectation about streams of notes that will be performed, in a way that provides a basis for advancing through a series of actions, a strategy referred to as planning as inference (Botvinick & Toussaint, 2012). The series of actions is not in fact planned explicitly, but identifying a distal goal implicitly yields an inferred plan.

The goals in question here have two aspects; namely, pragmatic and epistemic. On the pragmatic side, actions are selected to achieve preferred outcomes (i.e., maximise expected value), and on the epistemic side, to maximise expected information gain (Friston et al., 2017a; Schwartenbeck et al., 2019). Technically, this reflects the dual aspects of Bayesian or abductive optimality, by complying with the principles of Bayesian decision theory (Berger, 1985) and design (Lindley, 1956), respectively. The expected information gain is key here, and lends certain actions epistemic affordance that can be seen in the light of salience and novelty (Barto et al., 2013; Schwartenbeck et al., 2013). In other words, plans that lead to novel outcomes correspond to actions that resolve the greatest amount of uncertainty. It should be noted that action refers to both overt movement (and possibly autonomic reflexes) and covert (mental) actions such as the deployment of attention.

Sub-goals can be added on-the-fly and reached almost instantaneously, subsumed into the current trajectory for a distal goal (Kaplan & Friston, 2018). Crucially, in a highly volatile situation – as in some musical improvisations – the distal goal can itself be revised on-the-fly as ‘costs’ mount – i.e., a preponderance of evidence is gathered for alternative distal goals. In terms of a decision tree, the current distal goal branch is pruned (Huys et al., 2012), but sufficient flexibility makes possible the inference of a new branch with a ‘leap’ (as described above).

Given the surrounding web of regularities we described as scaffolding, a flexibly selected new way forward may, however, no longer coincide with the previous trajectory extending from past to present. Thus, using the same predictive neuronal machinery discussed throughout, a new (subset of the) past is selected or ‘predicted’ (Gershman, 2017). This dynamic process provides a basis for assimilating a newly imagined future with past and current observations and the extant regularities of the surrounding scaffolding.

During musical improvisation, any present moment – the ‘now’ – is not merely a ‘you are here’ dot on an abstract map or avatar in a game that can be shunted around with virtually no resistance. Moreover, the experience of the present is not organized around a monolithic domain of statistical regularities, as when a computer-generated musical performance has been trained on a corpus (a distribution of pitch transitions). Instead, there are many intersecting sources of expected regularity – sound-producing instruments, aspects of one's own body, co-performer behaviors, idiomatic patterns, idiosyncrasies associated with specific performers’ personalities, features of a performance space, audience responses, and more – that form the scaffolding of the situation. Within this context, possible subsets of the past are (neurocognitively) linked to possible futures in alternative trajectories that extend bidirectionally in time from the present. This conceptualization is consistent with recent empirical studies that focus on general mechanisms, independent of music or improvisation (Sharp & Eldar, 2024; Tarder-Stoll et al., 2024).

Nevertheless, aiming at a possible future in such rough terrain may preclude it from ever being reached, such that fast, responsive, flexible maneuvering is required. This includes anticipating unlikely as well as likely contributions from improvising co-performers, which also has been studied as a general neurocognitive process independent of music or improvisation (Ma et al., 2024). We are finally in a position to bring together all the constitutive conceptual elements of action-oriented abduction, and present a concrete application.

Creative Improvisation as Action-oriented Abduction

Returning to the example accompanying Figure 1 (above), we can consider a performer having played the pattern (with notes in place of claps) up until t₀, the present moment, common to patterns A-D. Whether or not they play a note at t₁ will prune the decision tree to A/B or C/D. If a note is played, the A/B branch is thereby selected; the subsequent node in the tree that will disambiguate between A and B does not occur until t₃. Crucially, already at t₀ one of the four patterns may be inferred in the form of abductive hypothesis generation – an imagined trajectory from past to future through the present (t₀) governed by a hidden pattern such as (A), a repeating sequence of 4 on, 3 off. But if at t₃ an unexpected glitch occurs in which the attempt to play a note fails, evidence for pattern B is instead produced, and a rapid flexible abductive inference of a new hypothesis shifts from being improbable to highly probable (i.e., from ‘B is an unlikely outcome’ to ‘B is the most likely outcome’). In essence, action-oriented abduction is a novel combination of ordinary abduction with flexible adaptive planning as (active) inference.

There are two major simplifications in Figure 1 that are convenient for illustrating the basic point, but that can be expanded upon theoretically. First, the sequence that immediately precedes t₀ and is in common with all four patterns is simplified. In realistic conditions, we can stretch back further and open the pre-t₀ time span to include any freely selectable elements that are ‘predicted’ as evidence for a new hypothesized trajectory that would extend through post-t₀ action. Second, temporary divergences from the timeline (absent in the figure) can be introduced at any point by incorporating evidence into hypothesized sub-goals that are subsumed into the distal goal.

On the perceptual side, using the same conceptual framework that has underpinned our framing throughout (active inference), Friston and Friston (2013) describe a non-improvisational case of listening to different melodies that can be categorized (recognized) among a family of predicted trajectories. Violations in the expected trajectory of one melody, i.e., surprises, are accommodated by selecting an alternative that corresponds to a different melody. Next, we introduce a historical example to illustrate the relevance of this idea to the present focus.

In an improvised group performance of “The Song is You” by Jerome Kern, Charlie Parker inserted a melodic fragment of “Over There” by George M. Cohan (Woideck, 1996). Listeners are first surprised by a different melody than the song being performed, which becomes salient (draws attention) as a mismatch between actual and expected trajectories. Just as the momentary divergence is resolved upon a listener's recognition of the second song, a further surprise is induced by observing the fragment's seamless harmonic and rhythmic integration with the on-going performance. Maintaining the distal goal of the main piece's continuity facilitates the continuing interaction with the players in the group improvisation, and thus preserves the overall success of what the audience perceives as the originally indicated performance.

However, there is more to the story, in relation to the multilayered scaffolding we introduced in the previous section, but beyond the bebop idiom and the sensorimotor musical inter-player coordination. Consider the huge variety of symbol-mediated loops into material and social culture. These loops have been theorized as enabling new forms of predictive processing (Clark, 2016, pp. 277–278, see also 2006, 2008). For example, a spoken description of a visual observation can serve as a string of public symbols launched into the world that supplies a possible auditory observation to other agents, with different ramifications depending on their current circumstance and background knowledge. An individual experience that is thus externalized may interact with the surrounding situation in any number of ways, for example, as follows.

Continuing with the historical performance vignette, we can see the effects of such externalizations and cascades of loops through layers of scaffolding. As Woideck (1996) has documented, Chan Parker, who was married to Charlie Parker, was in attendance at the performance referred to above. The saxophonist, from the vantage point of the stage, saw a mutual of friend of theirs enter at the back of the club, partway through the rendition of the main song. That was the moment at which he incorporated the melodic fragment from the secondary song into his saxophone improvisation, thereby drawing attention to the (unspoken) lyric “over there”. Based on the couple's common background understanding, the musical phrase implied a verbal phrase that in turn became an affordance for directing her attention to a salient extramusical visual observation at the back of the club. This rather complex phenomenon quite simply led to her turning around and seeing their friend.

To contextualize the above example, the performer introduced a newly abducted sub-goal, a hypothesized trajectory linking a selected past with a possible future via present action. The selected past flexibly joined (colligated) multiple categories with differing sets of statistical regularities, including a familiar lyric from the repertoire, the on-going performance of another song, and the momentary appearance of someone off-stage. The sub-goal was the detour from the primary melody that could evoke the desired head turn. This sub-goal was instantaneously incorporated into the trajectory of the distal goal (performing the main piece). Accomplishing this incorporation entailed the harmonic and rhythmic integration of the interjected phrase so as to not derail the main song for the ensemble and audience, producing evidence of continuity, making it ‘work’.

Under time pressure, this performance demanded a fast skillful response, governed by habitual sensitivity in perception, retrospective and prospective imagination, and habitual fluency of action. The nature of action-oriented abduction can be tied to the question: ‘what imagined trajectories within the scaffolding are available to act on that would assimilate the evidence from this (or an alternative) subset of the past with the present and with a possible future?’ (see Figure 2). The fast response was driven by curiosity, including an openness to flexibly allocate attention to take in dynamic surroundings. At the same time, there was also an active, exploratory probing of possible future outcomes that could be worked out in practice.

Figure 2.

Integrated retrospective and prospective trajectory space. The uppermost diagram (I) illustrates how action selection (a-e) at the present moment (t₀) simultaneously selects a trajectory between various subsets of the past (P1-P4) and various imagined futures (F1-F4). The annotations in (II) and (III) refer to (I), and serve to connect it back to Figure 1 (filled and empty circles). The middle diagram (II) shows a common lefthand (shaded) sequence of filled and empty circles forming different patterns depending on whether a filled or empty circle is selected. The lowermost diagram (III) shows a similar pattern, but for a case in which two different past sequences could equally lead to a common future sequence (shaded).

In the scenario described above, a group improvisation ensemble remains on the proverbial same page, while an embedded divergent interaction takes place with one performer and an audience member. Similar divergences can arise wholly within an ensemble. Linson and Clarke (2017) discuss a group improvisation scenario in which the ensemble members pursue different performance strategies related to diverging analogical associations (e.g., extending a musical pattern based on its rhythmic versus timbral properties). In such cases, parallel series of associative chains per performer may develop into independent ‘local’ contexts that render some branches of subsequent associations more probable than others (Bar, 2007). But, while each performer's trajectory may in key respects diverge from one another, skilled group improvisers will act to subsume the trajectories into an emerging (i.e., abductively inferred) piece of music that is collectively negotiated, with the common distal goal of producing a successful group performance.

While we have focused on music, other improvised arts are also relevant. We briefly mentioned dance above, and it is also worth considering ensemble improvisation across practices within an art form, e.g., combined music and dance. In flamenco, for example, there are idiomatic tonal patterns in the music, and idiomatic movement patterns in the dancing, that are part of a sequentially organized tonal and rhythmic performance structure. Ensemble performers use a blend of improvisation and orienting structural reference points to achieve fine-grained inter-performer coordination. An intricate performance thus includes many possible points of failure that depend on individual musical, physical, and social proficiencies (Maduell & Wing, 2007).

In light of these possible failures, there is a common distal goal of producing an ensemble performance that ‘works’. At the same time, not only do flamenco styles relate to a set mood, but performances also unfold with an emotional-dramatic arc that remains open to unexpected twists and turns introduced by individuals. Granted, first-order coordination via perception-action loops could produce a technically rigorous ensemble performance in the absence of creative abduction. But alternatively, on our account of creative improvisation, the sub-goals within each performer's role are continuously dynamically updated, while being subsumed into the overarching progression of the ensemble performance. The performers – whose perspectives overlap to varying degrees – are attuned to these intertwining trajectories, but not only for coordination. They are actively inferring both their own and the group's trajectories in real time, i.e., as action-oriented abduction.

Conclusion

We have now described action-oriented abduction and applied it to creative improvisation, elaborating on the integration of fast flexible interactions and driving curiosity. While there remain a range of challenges to investigating the biocognitive dimensions of improvisation and creative cognition in general (McPherson & Limb, 2013; Vuust & Kringelbach, 2017), we have articulated a theory of creativity in improvisation that, in our view, goes beyond what is assumed in most empirical study and formal analysis on this topic. Specifically, our pragmatist-inspired, predictive, embodied approach gets at an enriched sense of creative improvisation that is irreducible to mere combinations of spontaneous invention, syntactical variation, and dynamic coordination.

A related approach uses an integrative review with a neuroanatomical focus to elaborate a model of musical improvisation as a special case of general creative cognition related to divergent thinking (Loui, 2018). With a different starting point and progression, that model arrives at many of the same key components as ours, including fluency, flexibility, cohesion, and intrinsic reward (which for our purposes appears in the guise of curiosity and epistemic gain). That account highlights, for example, that compared to other listeners, those with training in improvisation express a greater preference for progressions with unexpected chords (Loui, 2018), only without addressing why this should be the case. An explanatory theoretical foundation is provided (we claim) by work on the brain as an organ of embodied prediction, encompassing not only perception and motoric action, but also memory, imagination, and mental action, wholly entangled in webs of cultural scaffolding, including but not limited to musical idiom.

Returning finally to Ingold's conceptual opposition, recall that abduction was framed as a culmination of the past in the present, which he regards as (detrimentally) reading creativity ‘backwards’. In contrast, his preference for improvisation was based on its relation to reading creativity ‘forwards’, from a present that gives rise to things in the future. From the account we presented here, it should now be clear that understanding an enriched sense of creative improvisation as action-oriented abduction reconciles this opposition.

Creativity through abduction in improvisation involves the continuous exploration of latent patterns in the past and present. These patterns are selectively externalized through their projection into a prospective future that may itself never transpire. Put differently, creative improvisation as action-oriented abduction produces a once unexpected future that emerges from a dynamically predicted past, both of which are responsively reconfigured on a rolling basis.

The curiosity of inquiry, the sensitivity and responsiveness to multiple intersecting contextual scaffolds simultaneously, can produce hypotheses of trajectories that could materialize a latent pattern that can ‘work’. For the most part, traditional musical composition can also be creative in this way, evolving over multiple revisions independent of performances, occasionally informed by rehearsals. The context within the piece and beyond it can come to influence its internal musical organization, driven by the curiosity, imagination, and experimentation of the skillful composer.

Distinctively, however, a much different pace is set when the creative process takes place without the benefit of revision, under the demands of a time-pressured situation, relative to opportunity costs. This existential situation continues to yield unexpected observations that may be dealt with immediately or ignored – perhaps only momentarily, then dealt with later on. Self-reflective intervention can certainly occur in stretches of improvisational performance, just as decisions in composition can be made quickly without being revisited. Still, a viable model of creative skillful improvisation must (like ours) be capable of fast responsiveness, driven by curiosity, and open to multiple kinds of influence from the immediate situation and the wider world.