Interpretation in human-machine improvisatory performance

Interactive music, musical meta-creation, electro-acoustic improvisation, human-machine performance… each of these labels refers to a type of performance that involves the improvised collaboration between humans and computer music systems. I have recently been fascinated by the contradictions that abound in this space. For me, placing an instrumentalist in a performative context alongside an autonomous or responsive software system raises significant questions about performative agency, as well as the notion of artistic authorship. This has led me to begin thinking about how the term musical ‘interpretation’ might apply to this context, in addition to improvisation, which is undoubtedly fundamental to this type of performance practice.

At first, speaking of musical interpretation might seem like somewhat of an anathema in the context of free or non-idiomatic improvisation. Of course, by definition in a freely improvised performance musicians develop a musical structure in real-time without the presence of an overarching compositional framework. The notion of interpretation as requiring something to interpret – that is, a pre-existing ‘work’ in the form of performance instructions typical of a musical score – does not exist in any traditional sense. However, despite the best efforts of programmers/system designers to seamlessly embed such systems into previously human-only musical performance, it may be argued that the introduction of autonomous systems into this performance scenario does challenge traditional understandings of improvised performance practice. As such, I believe the term interpretation warrants some attention here.

In the broadest sense, interpretation in performance might be defined as the process of making sense of the artistic ideas of one or more authors, and bringing them to life in the form of a real-time performance. Such an understanding of interpretive performance is of course common to all forms of the performing arts, including dance and its connection to choreographers, theatre and film’s playwrights and screenwriters, and finally to music and its composers. For music as for other art forms, integral to the notion of interpretation is the medium in which artistic ideas are transmitted outside of the performative context itself, what we might call the musical text. It is the interpreter’s job to make decisions about how to most successfully bring to life a composer’s ideas within the constraints communicated by their text. A composer might be fastidiously prescriptive by offering many detailed symbolic instructions for the performer to follow (as in the complex traditional notation of Brian Ferneyhough, or the tablature notation of Aaron Cassidy), or they might outline a framework that guides a performer’s interpretive decisions (“Draw a straight line and follow it” – La Monte Young; or “Make a sound in a middle place, in some respect, of the sounds around it” – Wolff).

Regardless of the medium through which a composer’s ideas are expressed, however, the contents of the musical text not only outline a composer’s musical instructions to be executed, they also embody and express context-specific assumptions about the role of the performer in bringing the work to life. To give a few examples, figured bass notation assumes a level of contextual understanding about harmony on the part of the performer; tablature notation directs the performer towards the execution of physical action before sounding result; graphic notation often relies heavily upon the more subjective interpretive faculties of musicians; and textual instructions like those of La Monte Young and Wolff focus upon providing performers with a musical and interactive framework with which to construct a musical happening. Despite their disparate manifestations, all of the above types of musical text articulate the desired boundaries of a musical performance, and imply various levels of musical and interactive freedom and constraint for the performer. These examples all involve some combination of explicit and implicit constraints upon the performer, and most of them act as a form of instruction. It is within these sets of explicit instructions and implicit boundaries that the musician is said to interpret a musical work.

So what is it that qualifies a musical performance as interpretive as opposed to it being completely improvised? And how might we understand the role of implicit musical constraints in the face of the more explicit requirements found in a set of instructions? The distinction between the concepts of instruction and constraint is an issue that dominates discussions surrounding the relative openness of musical texts, and therefore their ultimate interpretation by performing musicians. Understanding a musical text as a set of explicit instructions provided to performers, the term interpretation implies a balance between fidelity to these instructions, and the injection of significant performative and stylistic understandings of the composer’s intentions by the musician. In addition, the relationship between adhering to and interpreting instructions is often heavily dependent upon the stylistic norms of the musical period in which the work was written (as is often the case in the performance of ‘classical’ music). In this context, interpretation defines the process by which any gaps in communication between composer and performer are made sense of in light of accumulated cultural and stylistic norms. In this respect, the musical text (score) communicates the requirements of the work as articulated by the composer, whilst the musical and cultural context in which the work resides provides further constraints on any reasonable and/or correct interpretation of it.

It follows from this that the more unequivocal a text appears as a form of instruction, and the more significant the weight of cultural conventions associated with that text, the less interpretive flexibility is offered to the performer in creating their own personal ‘reading’. As more ambiguous forms of musical text, graphic scores are often provided to performers with no accompanying instructions or performance direction. Although such scores may not communicate explicit performance requirements of the work to the performer, given their identity as scores these objects still exist as forms of musical text to be interpreted. Graphic scores range from a set of symbolic instructions that leave levels of freedom within certain musical parameters, to objects of immense interpretive flexibility (such as in Cornelius Cardew’s Treatise, for example). In the latter case, the notion of interpretation rests upon the performer developing explicit rules and procedures in order to realise the musical work. Arrangements of graphic symbols such as those found in Treatise therefore become the catalyst for the creation of a personalised musical and sonic grammar in response to the text. Rather than acting as direct instructions, such scores provide a set of loose constraints that anchor a musician’s interpretive decisions; choices that are both guided by their own personal musical vocabulary, and the cultural and stylistic context of experimental musical performance.

Discussing the role of the score and of the composer in contemporary music-making, saxophonist Evan Parker posits that a musical score might be considered as either one of two contrasting types of text: in some cases as a representation of an ‘ideal’ performance, and in others as a ‘recipe for possible music-making’ (Bailey 1993, pp. 80-1). Speaking as an improviser, Parker’s opinion on the role of the composer and the prescriptive nature of some musical scores is articulated in his characterisation of the composer’s role as score-maker: ‘… if anyone in the production of a music event is dispensable, it is the score-maker, or the ‘composer’ as he is often called.’ (Bailey 1993, p. 81) Although the perspective articulated here is largely negative, Parker notes that the more musical scores resemble the ‘recipes’ he describes, the more there is a gap between the score as ‘ideal’ and any live performance of it. It is from within this gap that the real-time act of performance resides, leaving open for consideration further context-specific ‘ingredients’ of musical performance in addition to the instructions outlined by the composer. These may include the choice of specific musicians to perform the work, where the work will be played as well as how the audience might react.

Speaking of the difference between requirements and constraints in the context of musical performance, Bown has noted that performances of graphic scores may also include forms of extra-musical constraint that should be considered either directly curatorial, or at least requiring a form of interpretation by a group of musicians (Bown, pers. comm., 28 September 2014). Such constraints might include the decisions and wishes of a concert promoter (venue decision and therefore acoustics, performers asked to ‘keep it short’), discussions regarding musical/sonic materials between players, to the presence of ‘garish paintings’ hanging on the walls of the concert space. Such instructions and environmental constraints will of course require a form of interpretation on the part of the musicians that will directly or indirectly affect the resultant musical outcome of the performance. As Bown implies, such constraints often also apply to freely improvised music, no doubt affecting the way in which musicians interact with each other on stage, despite the lack of a unified performance framework as embodied in a score.

If such extra-musical constraints may act as objects of interpretation to a group of improvising musicians, might they therefore be considered as forming an interpretive framework on par with the ambiguous instructions of a graphic score in which musicians engage? My answer to this question rests upon an understanding of composition and curation as intentional activities of humans acting outside of the real-time context of musical performance. The framing of a performance context might be achieved through the existence of explicit requirements (notes on a score, imposition of formal boundaries etc.), but equally through placing musicians in a performance context that might constrain or alter their natural performance dynamic in some intentional way. However, the inherent collectivity and self-organisational properties of improvised music-making, free of intentional constraints on musical structure and materials, preclude it from being understood as an interpretive activity.

Because of a decided lack of musical text to be interpreted in this context, any extra-musical constraints that might influence an improvised performance should be considered part of the social and cultural context of improvised performance itself, and not intentionally imposed as objects of interpretation on an improvised performance. Musicians that engage in improvised performance bring with them their own personal histories of engagement with the practice, and although the boundaries and constraints may change from performance to performance, these form part of the context of improvised performance that has developed over time. This context both normalises and frames the performance activity itself. As such, although external constraints may be ‘interpreted’ by a group of improvising musicians – and may therefore affect the musical discourse – these constraints should be seen as manifestations of the context in which an improvisation takes place, and not part of a set of intentionally curated decisions in which the musicians must engage. The social and cultural context enables improvisatory music-making to occur, rather than directing it as such.

Of course, the most fundamental and defining constraint of improvised performance is the contribution of the individual performers to the musical outcome. From a freely improvised perspective, the most important ingredient of any performance is the ability for improvising musicians to be free to make their own decisions on how to affect the musical discourse, without having to defer to a set of instructions that might constrain these decisions. A form of musical activity that Parker has outlined as his ‘ideal music’, free improvisation therefore not only eschews the presence of a pre-defined musical text, it also places specific emphasis on the reliance upon the interpersonal relationships between like minded musicians improvising ‘freely in relation to the precise emotional, acoustic, psychological and other less tangible atmospheric conditions in effect at the time the music is played’. (Bailey 1993, p. 81)

So, acknowledging these specifics of improvised performance practice, what of the inclusion of autonomous systems to improvisatory practice? Can we assume that an improvised duo or ensemble performance would function in an identical way with the added presence of an autonomous musical system? To answer this question we must concern ourselves with the ontology of such systems, and the way in which musicians may perceive them in performance. With the ability to generate independent musical material as well as to adapt to their performance environments, it might not be so unusual to think of such systems as ‘virtual performers’ in their own right. In some respects, one might argue that the presence of such systems in performance, although synthetic, would not change the nature of the musical context given that a successfully cognate system would act in the same way as another musician. Successful systems of this type would therefore share authorship of a musical performance with their co-contributing human improvisers. From this perspective, the presence of such an autonomous machine might simply form part of any ‘less tangible atmospheric conditions’ Parker has referred to – its status as a machine contributor resting as a mere technicality in the context of freely improvised performance.

Contrary to this assumption however, it is my opinion that the introduction of such a system does alter the way in which the performance context is perceived, thereby changing the nature of the practice itself. The presence of a generative, responsive or autonomous musical system in performance brings into focus the relationship between system designer and any human musician engaged in this practice. It is my view that such types of human-machine performance take place as part of a unique form of socio-technical curation. The autonomous machine significantly affects improvisatory performance practice due its existence as a programmed entity. It is precisely because of the specific creation of a non-human actor (the software) to be engaged with in performance – a context traditionally reserved for human musicians – that the dual notions of interpretation and constraint in performance become relevant.

However, at issue when relating these concepts to improvisation-centred human-machine performance is the very notion of a ‘work’, the detachment of the developer/designer from the instantiation of the work in performance, and the lack of a perceivable musical text in either traditional or non-traditional terms. Questions we may ask therefore are where the work resides in human-machine performance, whether or not we can attribute the proposed musical framework (inclusive of the software) to a single author, and whether or not the software itself may be considered a musical text. Such questions lie at the heart of the concept of authorship in human-machine performance, problematising the notion of autonomy of both a software system and the human improviser in such scenarios.

Firstly, regardless of any manifestation of autonomy on the part of a musical system, one cannot discount the fact that such systems are designed by human actors – namely their programmers or system designers. Taking a broad look at the practice, it is first and foremost these actors that propose the musical scenario of human-machine performance to improvising musicians, either explicitly (via invitation) or implicitly (through software distribution). Regardless of how this proposition is made, in this context the system designer, acting as author, has proposed a musical framework to be navigated and explored in performance. The placement of one or more human improvisers in such a performance context should therefore be understood as a non-trivial act of curatorial authorship. The programmer is no longer simply the author of a piece of software, but is also partly responsible for the musical and interactive framework in which both human and machine engage. From this perspective, although a system may act in an unsupervised manner, the programmer shares authorship over the musical discourse through their software’s interactive behaviour, acting as a part of this curated performance framework.

Secondly, neither this broad curated framework nor any musical system itself exists in an aesthetic vacuum. Human actors developing such systems indelibly hold personal musical goals and preferences, many of which are subtly (and sometimes overtly) manifested in the behaviour of their systems in performance. Although our musical systems are often capable of surprising yet musically coherent results, their capabilities are inevitably understood as the results of the programmed decisions of some human author external to the present interaction. Whilst neither the programmer nor the machine may have made specific requirements of a musician’s performance, the programmed dynamics of such software systems will ultimately contribute to framing the musical and interactive boundaries of any human-machine improvisation.

Designer goals and preferences are also manifest in the manner in which improvising musicians are integrated into interactions with such software systems. In some cases, software behaviours and dynamics may have been explained to a musician prior to an interaction, whilst in others they may be left to the improviser to be discovered during performance. However, even if a musician is asked simply to ‘play normally’ as if improvising with another human, or advised not to try and provoke a direct response from the software, certain interactive preferences are revealed to the performer prior to an interaction. In the context of freely improvised musical performance, spontaneity and the ability to provoke and surprise are important elements of the performance practice. The choice not to reveal the underlying capabilities of an autonomous system can be seen as a manifestation of such an attitude to performance. As such, assumptions and expectations of both improvisation style, as well as a musician’s ability to act as if the machine is human are further revealed.

Another important factor in considering the nature of human-machine improvisation is the awareness and understanding of a machine’s capabilities. Just what abilities the machine has, whether directly responsive, analytical, generative or otherwise will no doubt affect a musician’s perception of their relationship to the machine in performance. In this respect, initially unfamiliar performance paradigms encountered as a result of human-machine performance context can be conceptualised as a form of material with which the human improviser brings their own musical self to bear. In addition, whereas human musicians conceivably approach an ensemble improvisatory context aware of the both the musical potential and cognitive faculties of their fellow human interlocutors, the same cannot reasonably be said of human-machine contexts.

In human-human improvisational contexts, although the possibility of encountering surprising musical materials and forms is always apparent, such materials are subconsciously rationalised and understood by musicians in terms of the skillset and abilities of their decidedly human counterparts. By contrast, any relationship that develops between a human and a computer system during performance might be characterised as one of navigation, exploration and discovery. The very programmed nature of a system’s interactive and sonic behaviour promotes a mode of interaction whereby the constraints of any given system are revealed during the act of performance. Programmed constraints, as experienced by the musician in performance, manifest themselves as both musical and interactive constraints on improvised performance. The more a musician spends time with such a system, the more such constraints are revealed. As such, these flexible and relational musical systems promote exploratory modes of interaction due to their inherent curiousness as programmed, interactive entities.

So does this mean that the software system itself becomes a type of musical text that the musician interprets? Although a musical text in the form of explicit requirements discussed previously might not exist, as we have seen, in communicating constraints upon a performance scenario and by embodying design decisions of human authors, these artefacts certainly contribute to the creation of an interpretive framework in which the musician engages. Even if one disregards some of the more obvious compositional decisions made by a system developer (such as parameter mapping, choice of sonic vocabulary etc.), we can also understand the choice of a particular generative algorithm as well as the type of audio analysis and feature extraction used in such software as aesthetically motivated. Although these decisions may seem fine-grained and technical in nature, they can reveal some of the context-specific assumptions of a system designer, and therefore further point towards the software system as providing constraints on performance.

It may therefore be argued that the development of the software itself is akin to the creation of a type of musical text not dissimilar from the graphic and open-scored works discussed previously. However, rather than embodied in an external graphical or textual representation, the constraints on a musician’s improvised performance with such machines are interactively instantiated through performance with the software itself. Certain modes of performance will ultimately reveal themselves as implicitly more ‘reasonable’ or ‘sensible’ than others, and may also have been communicated as such by the developer prior to an interaction. As such, imbedded in the software as text are the context-specific assumptions and expectations of a type of performance practice envisaged by the author of the software, as revealed by both the sonic and interactive constraints experienced in real-time by the human performer.

In such musical contexts, whilst the musician is certainly improvising with the machine, I would also argue that they are also working within a type of interpretive framework. Any musical performance with such systems may be seen as an instantiation of the combined musical ideas of the system developer, the musician navigating this space of ideas, and the interactively instantiated responses of a machine to the actions of both the performer and the overall musical environment. It is precisely because such systems are imbedded with these subjective attributes that performers in this context are necessarily engaging in a form of interpretation. Although their moment-to-moment performance may be freely improvised, the framing of the interactive context cannot suggest anything other than an interpretive framework. From this perspective, any consideration of a musical text must take into account the entire performance scenario.

The musical text is, in effect, the boundaries and constraints of such a human-machine musical interaction as influenced by the machine’s perceived capabilities. Navigating these possibilities in a truly interactive sense is the task laid out for the musician. Interpretation can therefore be characterised as the navigation a space of potential relationships between human and machine agency, a context envisaged and brought forward by a system developer to a live performance context.

Acknowledgements:

The thoughts above have been galvanised by some fruitful discussions with a number of fantastic musicians and researchers recently. Many thanks to Owen Green, Ollie Bown, Bill Hsu, Mark Summers, Aengus Martin, Arne Eigenfeldt and Pierre-Alexandre Tremblay for taking the time to discuss this topic.

References:

Bailey, D. 1993, Improvisation: Its Nature And Practice In Music, Paperback, Da Capo Press, Cambridge, MA.

Revisiting Symbiosis: a tentative manifesto?

Parasitic Relationships No.2 (Symbiosis Sate), 2013: Amber Stucke

A little while back I sketched out some rough thoughts about the metaphor of symbiosis in art and performance. These were really just some nascent musings thrown out there on a whim, coupled with the curious discovery of Leonel Moura’s Symbiotic Art Manifesto. I’ve read back on this post with curiosity over the past couple of months, and recently I’ve revisited some of these concepts to try and tease out what I was trying to say in a little more depth.

Moura and performativity:

In the previous post I mentioned Leonel Moura’s symbiotic art manifesto. What I questioned was how his ideas might be able to intersect with real-time performance, and with interactivity. The word ‘symbiotic’ in Moura’s lexicon describes the relationship between the artist and the machine/robot the artist has made to create artworks. From his perspective, the symbiotic relationship that is so vital for this new type of art exists between the artist as designer/creator and the machine/robot developing the artwork in real-time. In this context, the human needs the machine’s agency in order to render artworks, and the machine needs the developmental guidance of the creator (the structure) with which it is to then make artworks. This metaphor of symbiosis therefore relies upon mutual dependencies that exist between the designer/artist and the automaton/artist, the former as the ‘preconceiver’ of a possibility space for machine performativity, and the latter being the machine that navigates this space autonomously. This is a high level view of dependency in human-machine art making, but as I mentioned previously, it lacks a concern for an intersection with human agency during the performative act itself. What is missing therefore is a consideration for the real-time performance aspect of the creation of the artwork – the rendering of the artwork as a performance.

Looking at human-machine symbiosis during the performative act, the real-time act of decision-making by the machine (both with respect to its internal state and influence from the environment) can be viewed as a symbiont connected to its environment. In Moura’s work, as in the work of many procedural and generative artists, the machine’s automation could be said to act as the artist in proxy, with the results of this autonomous procedure being considered the final artwork. For those generative and/or metacreative works that exist in time (music, animation etc), it is the process of negotiation and generation – the performative act – that is presented as an artwork. For those works in which the environment involves the input of a human interlocutor, this performative act becomes more complex, and the notion of a symbiotic relationship extends into the act of performance itself. As such, mutual dependence between the automaton/machine, the human performer and the possibility space proposed by the designer all factor into our understanding of symbiosis in art. Because of the introduction of the performer to the interdependent relationship between artist and machine, the designer’s relationship to the situation is in some ways diminished, and in others it is strengthened by the real-time encounter.

What I am trying to understand is the entangled relationship between human and machine performances, both outside of and influenced by our focus upon making machines that participate in musical performance autonomous in some shape or form. Where symbiosis comes in is through an understanding of the bi-directional nature of adaptation and learning in the performance/development of such technological systems. For me, what is most important is finding ways of conceptualising interactivity in this performance practice that directly acknowledge the potential for mutually adaptive relationships to develop between humans and machines. To date, I feel that many approaches in this area have downplayed the need to understand the system’s affect on the human, and to focus upon developing systems that display autonomous behaviour for musicians to engage with as if the machine was another human player. This is why Moura’s ideas have struck a chord with me. Although in his manifesto he focuses on the machine’s ability to make art autonomously, he is also concerned with the new relationship that exists between these technologies and human creators. When considering this concept of developing artists rather than art, he asks the question: “what do we become ourselves?” In doing so Moura acknowledges how this technology puts this mutualistic relationship in sharper focus. As he states, “The life of the artist/machine is interlinked to the life of the artist/human.”

Designing for the (un)known

Whilst artificial intelligence and the study of autonomy is a perfectly viable field of study for musical interaction, it is my view that in seeking ‘natural’ behaviours for such systems we must also acknowledge our personal biases towards certain types of musical interaction. That is, we must acknowledge exactly what we deem to be natural. The irony in many biologically inspired generative contexts is that whilst we may be seeking to develop new modes of musical interactivity, the modes of generativity that are sought are often justified because of their ability to act in ways that approximate us as humans. This is indeed a curious paradox. What this kind of developmental trajectory presupposes is that human performance is itself predictable, that the machines we build should be immediately understandable, easy to interact with and adaptable to our needs. What I find fascinating however, having developed and performed extensively with such a system myself, is that the musician who begins performing with these systems is not the same after each successive interaction. Musicians adapt to different performance situations, they learn to embrace new performance scenarios and their conceptions of an ‘ideal’ performance practice expand through interacting with our systems over time. This suggests that if we are to be engaging in the development of new forms of performance practice, we should also consider the way in which our machines shape performers, as much as we train our machines to be just like ourselves.

Also curious in the design of such systems is the focus upon modelling, and or approximating human performance dynamics in the machine performer. Some approaches take as their starting point an established performance practice such as free improvisation, then attempt to understand its inner workings and then use these understandings as a basis for design. As an example, some approaches have emphasised that the practice of collective, non-idiomatic musical improvisation is global in nature, with improvising musicians reacting to the overall sonic environment and not responding directly to the material provided by individual musicians. Implicit in such approaches is an overall concern for designing for musical improvisation as a known and understood musical practice. Because such forms of musical improvisation might be thought of as non-referential, and that the global sound environment plays a larger part than the sum of its parts, consequently any musical system designed to take part in such an encounter should adhere to this understanding of the performance practice.

Here I take a broader view of the role of the design of musical systems. Although I understand human improvisatory practice as being the foundation of any human-machine performance dynamic, I also understand that such a dynamic is in itself fundamentally different precisely because of the presence of the machine collaborator. As such, I believe that any artistic practice that seeks to explore novel interaction paradigms should also be questioning how the role of the machine will ultimately alter understood modes of practice. Although we may be comfortable modelling our machines on types of behaviour closest to our own experience, we must also understand that the very context of human-machine performance practice changes that experience, often even before a live performance has actually begun.

Metaphors for interactivity:

One thing that I have found curious in this area of research and artistic work, as well as in other areas of human-computer interaction, is that the adoption of metaphors often plays a central role to both design and evaluation of our musical systems. Metaphors are often used to describe interactive traits of musical systems and environments – control metaphors related to instrumental performance; biological systems metaphors such as stigmergy that describe/define group dynamics; categorisations of interactivity into differing modes such as conversational/ornamental/instrumental interaction; understanding systems and interfaces as prostheses etc. On the one hand such metaphors serve define desired modes of interaction to aid design, but they are also ways of understanding interactive traits perceived when observing humans interacting with machines. For the former, they provide ways of conceptualising approaches to designing, situating a complex developmental project within a broad interactive/relational framework. They become anchor points with which to ground future design aims. For the latter, they enable a retrospective understanding of the interactive traits exhibited by the interaction domain itself, understanding these situations without imposing any preconceived notions of interactivity on the system.

Whilst some of these metaphors of human-machine interaction have been developed through observation of real-world settings, others remain speculative understandings of the space in which humans and machines interact and are borne from personal artistic experiences and/or aesthetic leanings. As such, what fascinates me is that such speculative understandings say as much about the aesthetic preoccupations of the designer/researcher as they do the phenomenon itself, as I believe they do in Moura’s work. It is my view that the choice to develop systems that can participate with humans in musical performance is in itself a speculative position, and so it is from this vantage point that I situate my ideas of symbiotic human-machine performance practice.

Symbiosis – what is it really?

Here I’d like to outline what I understand about the concept of symbiosis, and how I see it relating to human-machine performance. Symbiosis is a biological process that describes the interaction and often interdependence between organisms in the natural world. There are three main types of symbiosis observable in nature: mutualistic, parasitic and commensalistic. Each one of these types describes the relationship that exists between different organisms, and is related to dependence which is either uni or bi-directional:

• Mutualistic Symbiosis – describes a situation where two organisms are dependent upon each other, with both benefitting from this interdependence
• Parasitic Symbiosis – is a one-sided relationship where one organism benefits and the other member is harmed by the interaction
• Commensalistic Symbiosis – where one organism benefits from the another, yet the other is neither harmed nor helped

Adaptation and dependence:

So, how might/do these ideas play out in human-machine performance? As I mentioned in my previous post, I came to an understanding of symbiosis as a metaphor for human-machine musical interaction due to its focus upon adaptation and dependence, in contrast with a strict consideration of machine autonomy. For me this is a fundamental distinction to make. In my view, the context of human-machine performance practice (including the process of developing musical systems that take part in performance) is something so specific and alien to traditional performance practices that it benefits from being understood, described and designed for in and of its own terms. What appeals to me with the metaphor of symbiosis is that it becomes a way of understanding and conceptualising a type of musical performance that is not possible outside of a human-machine context. Although the continued development of the autonomous capabilities of machines is a natural technical and theoretical trajectory for software development for many fields, I believe what this priority misses out in the context of human-machine performance is an overall way of understanding the mutual adaptivity of humans and machines in these hybrid environments.

For me, symbiosis as an interactive metaphor sits decidedly outside of the purely technological, acknowledging the mutual dependency and mutual influence between player and machine, machine and composer/developer, and researchers and their theories. The metaphor is both inherently interactive, but also has its genesis in the intersection between separate and contrasting organisms. As such, I use it to encompass the space in which human-machine performance takes place, as well as helping to define what approaches to developing technological systems that take part in these performances. However, this shift in focus towards dependence in an interactive relationship should not be misconstrued as a move away from developing the machine’s capacity for displaying agency in an interactive exchange. Instead, the process of symbiosis in performance should be understood as encapsulating the complex interaction between two autonomous entities that are both aware of each other, but also dependent upon each other in performance.

Defining symbiotic musical interaction:

So, taking the above definitions of symbiotic relationships, I would like to sketch out what I think characterises symbiotic interaction in human-machine performance practice, and how I see it reflected in my own work in this area.

For me, symbiotic musical performance means:

• Mutualism – mutual dependence between human and machine in performance
• That one’s actions may affect the evolution of the another’s material, though not as a direct reaction to or cause of the other’s responses
• Machine agency and autonomy are an ingredient of such interactions, though not the only ingredient

The development of interactive systems, whilst often focused upon maximising the autonomous decision-making capacity of the machine, should be thought of primarily as engaging human musicians in a mutually influential performance context with a machine. Human engagement with such systems fundamentally differs from human-human interactive contexts. In human-machine improvised performance, in my view performers are in some way dependent upon the interactive capabilities of the musical system. Musical systems are developed with specific stylistic constraints in mind, and performers interacting with such systems learn to interpret the nuances of these systems through repeated exposure to them in rehearsal and performance. They are dependent on the system in a broad conceptual sense, as it is often the case that the musical system’s interactive vocabulary (or indeed, sonic vocabulary) defines the range of possible musical territory to be covered in an interaction. Although a musician is clearly an autonomous entity, their musical input to the performance is shaped and altered by the context provided by the intersection with the musical system. This may in fact be thought of as a form of interpretative performance, not unlike traditionally understood notions of musical interpretation. This last point opens up a range of issues related to ownership and authorship of which I plan to write about in a future post, so I won’t go into these ideas further here.

From a system design perspective, it is possible to conceive of such systems as enabling either a mutualistic or commensalistic symbiotic relationship to occur in performance, as outlined above. The development of such a relationship need not seek to maximise the qualities of musical autonomy, however musical autonomy may emerge through cumulative interaction with the system over a period of time. The ability for systems of this type to exert agency in performance whilst still depending upon the sonic input from a musician provides a challenge for the system designer. In the context of human-computer improvisation systems, one thing that enables a musical relationship to develop between software and a musician’s performance is how a piece of software seeks to deal with information it receives in the present, as well as how it makes sense of the past. In human-machine performance practice, an interactive environment fundamentally shapes the performance of a human musician over time throughout an improvised session, as well as from session to session. Mutual dependence in this instance refers to the change brought about in the human’s behaviour towards the system, and the ability for the system to change along with the performer. One way of conceiving of machine dependence therefore in this context is through the material siphoned from the performer and used to further its own autonomous behaviour. Taking the history of past material as a space of possibilities provides the machine a sense of performative agency that is directly tied to the cumulative history of the human musician.

Symbiosis and _derivations:

My work developing and performing with the _derivations performance system I believe has embodied this conception of symbiotic performance. One of my central concerns for developing musical performance system was that the machine, although not directly controlled by a human performer, should be heavily influenced by a performer’s playing. Although I was interested in developing a system that could contribute to performance un-assisted, this autonomy would need to be emergent and tied to the specific context of the current performance. In much of my work leading up to _derivations I experimented with techniques for capturing both audio and analysis data streams that could then be used for re-generation. Beginning simply by live sampling snippets of my saxophone performance, I soon realised that I was drawn to using recorded data of many kinds as a means to generate complex behaviour using simple underlying rules. The possibility of re-organising, processing and re-purposing sampled data streams (including audio recordings) became a meta-compositional process, whereby my musical decisions were tied to finding ways of coherently generating new material from material siphoned throughout a performance with a human player. Such generative techniques, in comparison to other approaches to developing autonomous behaviour, are of course extremely dependent on outside stimulus. Without having interacted with a human performer, systems of this type have no space of possibilities with which to generate their own material. This is a complex and often contradictory approach for human-machine performance, which from a design point of view provided a very concrete set of constraints with which to work with during development.

As for human dependence on the machine, this is what I am currently in the process of analysing in much of my writing. The development process that led to the creation of _derivations in its current form involved much live performance. My dual role as a performer and developer on the project meant that the line between development and performance was continually blurred. In my development process I would often have a saxophone at hand ready to ‘test’ my latest changes to the system, enabling me to make on the spot decisions about the software as I continued to refine its capabilities. The affect on the system’s overall design is indisputable, and I would attest that the machine is very much ‘composed’ in this respect. However, reflecting on both this testing process as well as the very many performances and full length rehearsal sessions I engaged in with _derivations, what is more subtle is the way in which my performance style has evolved to encompass the capabilities and intricacies of the _derivations environment. The intimate knowledge I have as both developer and performer of the system enables me to shape my performances in a way that is particularly in tune with the system’s character. I cannot say for certain that I would perform the way I do with _derivations with any other human performer. Knowing the system’s ultimate dependence on both my current and past performance input, as well as how it makes connections between these states means that I am able to provoke and predict certain actions, but also to be genuinely surprised by others. My performance has evolved with the design of the system, but at the same time, I am in the moment of performance dependent on the interactive context that the system provides. This provides a set of constraints for improvisation that I find fascinating in performance. A composed space of possibilities that – for want of a better term – enforces a mutually dependent relationship to develop from moment to moment in performance.

Closing thoughts:

So to finish off these thoughts on symbiosis, below I outline what I see as some of my fundamental approaches to creating mutual dependence in human-machine performance. Although I call this a template, I must stress that these were not guiding principles outlined before developing a performance system, but rather tendencies I have observed through retrospective consideration of my own work that point towards the concepts I’ve outlined in this post. Considering mutual dependence as a design principle is sufficiently broad in many respects, but also very constraining in others. The methods one might use to generate relationships between captured and analysed performance material are many and varied. However, what I have found is that focusing attention on capture and storage, navigation and re-generation has provided an intriguing balance between transparency and opacity of a machine’s contribution to a human-machine improvised performance.

Template for the design of a musical symbiont:

• The machine is an empty vessel – it cannot act without having interacted with the human
• The machine and the human form a symbiotic system whose musical results are an emergent result of their interactive history
• The machine’s behaviour and/or content is in some way dependent upon the human’s current and past actions
• The autonomy of the machine’s actions is in some way derived from the interactive history with the human performer
• The machine displays agency by making connections between materials siphoned from human interlocutors
• The machine feeds off the performer, using captured memories of the encounter as fresh material for its contribution
• The human’s past becomes a space from which the machine draws upon to influence the future musical states
• The machine also listens to itself, depending equally on its own past as an indicator of its future state as it does outside stimulus (autonomy)
• The impetus for the machine contribution is tied to some present understanding of the human’s actions