Adaptive Specialisation and Music Games On Networks

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Proc. of the 13th International Symposium on CMMR, Matosinhos, Portugal, Sept.

25-28, 2017

Adaptive Specialisation and Music Games on


Networks

Stefano Kalonaris

Queen’s University Belfast


Sonic Arts Research Centre
skalonaris01@qub.ac.uk

Abstract. This paper suggests a novel paradigm for structuring real-


time musical interactions, bridging the seemingly distant areas of game
theory and freely improvised music. It does so by employing adaptive
specialisation, in the context of networked music performance for large
groups of free improvisers. The proposed model is described as a Bayesian
game, in which players have a dynamically and stochastically assigned
private personality. Such personality is the object of their mutual infer-
ences as to whether to trust their connected neighbourhood and request a
joint musical assignment for such locality. This, in turn, entails implicit
negotiation over a choice of available relational functions. The author
asks whether such interaction model constitutes a useful framework for
organising musical relationships while questioning ideas about freedom
and form in music improvisation.

Keywords: Adaptive Specialisation, Markov Networks, Free Improvi-


sation, Networked Music Performance, Bayesian Games.

1 Introduction
Free improvisation started to develop around the early 1960’s, within commu-
nities of improvisers in the USA and Europe. It is a musical expression that
claims to be referent-free and to represent unbound and unconstrained extem-
poraneous musical creation, where players are deliberately not conforming to
any particular idiom or language [1]. Free improvisation presents unique chal-
lenges when performed in large groups and some of its eminent practitioners
have pointed out that such a practice is often more challenging and less satis-
factory when it is scaled to a size comprising more than 4 or 5 players [2]. In
order to address this issue, many diverse and idiosyncratic strategies that allow
for musical coordination have been employed. These might be directed frame-
works [3], the memorisation of a codebook of signs/cues [4] or they might be
inspired by board games and/or following a set of rules as in John Zorn’s Co-
bra or Game Pieces [5]. This paper proposes a novel model for large groups of
free improvisers where musicians are linked via User Datagram Protocol (UDP)
connection, exchanging information, requesting assignments and shaping clus-
ters of musical interaction in real-time. This paradigm was firstly employed at

420
Proc. of the 13th International Symposium on CMMR, Matosinhos, Portugal, Sept. 25-28, 2017

the end of 1970’s by the League of Automatic Composers. Towards the end of
the 1980’s the group re-named to The Hub and explored remote performances
over the internet. Such an extravaganza posed considerable problems in terms
of latency and bandwidth, which were partly addressed by the work of Chris
Chafe and his SoundWire group, based at Stanford’s CCRMA [6]. Within the
scope of this paper, the network is viewed as the expressive medium over which
distribution of agency and decision-making processes amongst and between free
improvisers take place. Decision-making processes are also paramount in other
situations that require strategic reasoning, such as games, social and economic
relationships. Viewing free improvisation as ”a forum in which to explore various
cooperative and conflicting interactive strategies” [7, p. 184], can help to link
the process of free music improvisation to the branch of applied mathematics
referred to as game theory, which is divided into two main branches: cooperative
and non-cooperative. These are normally highly stylised and abstracted, dealing
respectively with decision-making processes in the context of forming coalitions,
in order to maximise the profits for the parties involved, or in the context of
strategic situations where the individual interests and gains are paramount [8].
Although seemingly distant, free improvisation and game theory are linked by
their common stress on interactive strategies and distributed action [9]. Fur-
thermore, free music improvisation entails a high degree of dynamical trading
of roles, which allows the participants to shape interactions and to react to un-
foreseen circumstances with split-second decision-making wizardry. This trait of
free improvisers which makes them extremely versatile in negotiating real-time
objectives and roles, is akin to what economists have termed adaptive and/or
subjective specialisation [10].

2 Adaptive Specialisation

Adaptive specialisation is here referred to as the identification of a specific


role within the context of barter societies. In a network of agents engaging in
barter relationships without a social convention, the ability to predict an equi-
librium (intended as no agent having an incentive to change her behaviour) is
null [9]. Two conditions are necessary for the emergence of adaptive special-
isation, according to economist Robert Gilles. The first condition is that the
society/network be small, meaning that there can only be few agents interacting
and negotiating their mutual actions. The other condition is that, in such net-
work, the agents abide by the chosen barter convention as their decision-making
framework. In a scenario of this type, given that an agent has sufficient infor-
mation about the other agents, she will actuate a best response strategy [8] as
defined in the context of game theory. In the particular case where two agents
have identical sets of skills and demands, this reduces to a coordination game [8].
It can be difficult to map the concept of adaptive specialisation to the creative
setting of free improvisation, unless it is understood to be a process where the
assignment of the roles is not predetermined or immutable but is instead guided
by necessity and continuous contingency. All players are dynamically adopting

421
Proc. of the 13th International Symposium on CMMR, Matosinhos, Portugal, Sept. 25-28, 2017

and trading a plethora of roles, thus being required to supply di↵erent sonic
structures/textures, and to be proficient in all of these roles, be these defined a
priori or emerging ad hoc. Rather than labor and role, it is more appropriate
in this context to substitute these terms with process and function. This simple
shift allows to map roles to strategies linked to particular music processes, as
outlined in the work of Thomas Nunn, improviser and researcher of free improvi-
sation [11]. Regarding the content and flow of free improvisation, Nunn identifies
seven types of what he calls Relational Functions, namely: solo, support, ground,
dialogue, catalyst, sound mass and interpolation. It is the author’s contention
that Nunn’s Relational Functions constitute a solid basis upon which to build
the proposed model of interaction for free improvisers, as outlined in the next
section.

3 Proposed Model

Adaptive Markov Network for Free Improvisers (hereafter referred to as AMN4FI)


is a novel model for structuring large groups of free improvisers, liable of being
classed as a game piece [12, 13]. AMN4FI’s architecture follows from the author’s
previous experiments with Markov Networks [14], thus being implemented with
a centralised paradigm (one-to-many: server to clients). Formally, AMN4FI is a
synchronous repeated Bayesian game [8] over a Markov Network [15, 16]. There
are N players, each connecting to other players and playing a stage game. Each
of them can only play and musically interact with the players who are connected
to her in a local cluster. In a typical stage game, each player’s maximum links
(degree) is capped at N-2. This maximum number of connections does not have
to be exhausted. Players are arranged on a ring lattice, both in the physical
space and in the graphical user interface (GUI), with each player being assigned
a number from 0 to N-1. A typical physical configuration can be seen in Figure 1.
Once connections between players are made and the various local clusters have
been formed, any player within her cluster can send an assignment request via
the server. The latter sends out a prompt to the cluster and a countdown of 10
seconds starts. If no player cancels the request, the server then dispatches the
joint assignment for that cluster, in form of a list of three possible Relational
Functions. At this point, the players who are members of the same cluster ne-
gotiate the roles in real-time, in a non-verbal and non-explicit way. Practically,
players make inferences and assumptions about their cohort and actuate their
best response with respect to the musical scenario they are facing, as well as
in accordance to their own personal preferences with respect to the Relational
Functions they have available. Players are thus allowed to revoke edges between
one or more players they are connected to, and instantiate new ones (always
within the max number allowed), choosing from a list of available players.
All players have the same action space A = {connect, disconnect, send joint
assignment, veto joint assignment, accept joint assignment}. AMN4FI can be
played in two modes: ’Pure Adaptive’ and ’Adaptive + Bayesian’. In the latter,
an additional set ⌦ is used to model the player’s uncertainty about the other

422
Proc. of the 13th International Symposium on CMMR, Matosinhos, Portugal, Sept. 25-28, 2017

Fig. 1. AMN4FI: star configuration

performer’s type, or personality. This is a finite set, which describes the individ-
ual characteristics for each state, and it comprises of three items: {cooperative,
non-cooperative, chaotic}. These represent musical personalities and attitudes
to which players abide by. One of these types is assigned with equal probability
at the start of the game and changed thereafter at random intervals, by the
server. A finite state T indicates the signal that might be observed by player
i with respect to ⌦. The signal function will thus be ti : ⌦ ! Ti . A proba-
bility measure pi on the set of states ⌦, expresses the prior belief of player i,
for which pi (ti 1 (ti )) > 0, 8ti 2 Ti . Lastly, the preference of player i in regards
to the set of actions A with respect to the set of states ⌦ is defined as the
preference relation, akin to the concept of factors in a Markov Network. In this
game, every player can observe (listen to) the actions of all other players, yet it
is not possible to class it as a Bayesian extensive game with observable actions
in that the type !i 8!i 2 ⌦i keeps being stochastically reassigned at every pe-
riod. Players form local clusters, in accordance to either their preferences (if in
Pure Adaptive mode) or their inferences about the nature of their type (if in
Adaptive + Bayesian mode). In both cases, a joint assignment is represented by
a set of roles/musical functions being sent to a local cluster, and its acceptance
(or rejection) signifies the willingness to adopt them and negotiate over them.
These roles are directly mapped to Nunn’s Relational Functions. Additionally,
the following mappings are also used, to establish a meaningful parallel with
adaptive specialisation, as outlined earlier:
– social network ! group of improvisers
– social wealth/good ! desirable/satisfactory musical outcome/interactions
– specialisation ! Nunn’s Relational Function
The two conditions for the emergence of adaptive specialisation mentioned ear-
lier are met in that the network of players is relatively small (normally com-
prising less than twenty players), and that the barter convention adopted is

423
Proc. of the 13th International Symposium on CMMR, Matosinhos, Portugal, Sept. 25-28, 2017

egalitarianism. Furthermore, all players are assumed to be independent and self-


sufficient. However, having established the correspondence for the utility (map-
ping 2, above), this scenario does not reduce to a coordination game, as two or
more players might choose the same relational function or modify the chosen
role solely according to creative, artistic and aesthetics goals.

3.1 Graphical User Interface

AMN4FI implements a simple visual representation of the local musical relation-


ships, providing a system where players explicitly have to commit to, or refuse,
such relationships, thus shifting the focus towards sustained, multilayered mem-
ories of musical and social interaction. Concretely, at each game stage, player i
will connect to other players in the group by simply pressing the number corre-
sponding to them. In turn, they can accept the connection or disconnect from
that player if they wish so, for reasons concerning their inferences about the
type of the player or simple personal preference. Each player appears as a red
node on her own GUI and connected players show as green nodes. The GUI has
also several text/message areas, where players are able to view the play mode,
notifications and error messages from the server (for example, whether they have
exceeded the maximum number of allowed connections, or if they are attempting
to connect to themselves, etc), a window where a countdown of 10 seconds is
triggered every time a player in their cluster sends an assignment request, and
an assignment window displaying (once the request has not been vetoed) the
roles and relational functions available to the cluster. Figure 2 shows a typical

Fig. 2. AMN4FI: typical operation

succession of events when playing AMN4FI in Pure Adaptive mode, where:

1. Player 0 (in red) connects to Players 3 and 4


2. Assignment request to cluster [0,3,4]
3. Request is accepted and assignment is displayed

424
Proc. of the 13th International Symposium on CMMR, Matosinhos, Portugal, Sept. 25-28, 2017

4 Procedure
AMN4FI was performed on 01.07.2016 at Bath Spa University, at the I’klectik,
London, on 20.07.2016, and at the New River Studio, London, on 04.09.2016.
A total of 25 experienced free improvisers took part in these case studies, and
both evaluation via questionnaires and focus group discussions were carried out
on all occasions. The first of these experiments did not allow any familiarisation
phase nor rehearsal of the system. This was due to the nature of the event, which
comprised of a busy schedule of talks, performances and demos. These events
were scheduled back to back without any additional time allowance besides the
technical set-up. The second and third experiment, instead, allowed ample prepa-
ration and rehearsal time, and it was possible to apply a control condition for
the group of participants. During these instances, the players performed firstly
unconstrained, then once more following the architecture of the model here pro-
posed, under rehearsing conditions (without an audience, with the possibility to
start and stop if needed in order to clarify the a↵ordances of the system) and
then again in a public performance. The introduction of this control condition
helped the players to place awareness to the modus operandi of the system,
and the changes in their behaviour that this system might introduce. This was
particularly evident during the focus group discussion, which took place after
the rehearsal session, along with the completion of the evaluation form. The
groups then performed again with an audience and in a similar format, in that
both AMN4FI and system-free pieces were played. Both focus group discussion
and questionnaire completion were not repeated after the public performances.
All experiments were carried out with participants who are active practition-
ers of free improvisation, thus ensuring high ecological validity, involving het-
erogeneous communities of free improvisers, in di↵erent geographical locations.
Questionnaires comprised of 9 items, providing a multiple choice Likert scale in
6 levels (from ”strongly disagree” to ”strongly agree”, re-coded to 1-6, with the
exception of the item Proficiency, spanning from ”None” to ”Expert”). Items
in the questionnaire reflected the author’s interest in whether AMN4FI can in-
duce successful freely improvised music whilst coordinating/structuring musical
relationships in novel ways and suggesting a re-evaluation of the player’s beliefs
on freedom and form in music improvisation. In order to account for a relevant
concept of what constitutes successful free improvisation, questions were chosen
based on the opinion of expert practitioners and researchers [7, 17–19].
More specifically, items measured self-reported proficiency, satisfaction, per-
ceived sense of form, dynamics and texture sustain, sense of structure, clarity
of interpersonal musical relationships, novelty of musical interactions, (multi-
layered) distribution of interactions, and constraint experienced. Nineteen par-
ticipants in total returned completed questionnaires.

5 Results
Given the small size of the sample, these results can by no means claim to be con-
clusive or statistically significant, but they o↵er a useful insight into the practice

425
Proc. of the 13th International Symposium on CMMR, Matosinhos, Portugal, Sept. 25-28, 2017

of seasoned improvisers, and they can help assessing the suitability of AMN4FI
as a framework for interactive decision-making and systemic improvisation. It
is important to keep the opinions expressed by the participants as a paramount
factor for the analysis of AMN4FI, since it was developed by and for practi-
tioners. The distribution of the reported proficiency in free improvisation was
approximately normal, with a mean of 4.5 and a standard deviation of 1.098127.
Likewise, the distribution of the reported satisfaction with respect to AMN4FI
was also normal, with a mean of 3.833333 and a standard deviation of 0.150477.

Fig. 3. AMN4FI’s satisfaction per proficiency

Proficiency and satisfaction were not strongly correlated but a strong rela-
tionship between experienced sense of form and satisfaction emerged, with a
correlation value of 0.79. A multiple linear regression was calculated to pre-
dict satisfaction based on proficiency, form, sustain of interactions, structure,
interpersonal dynamics, novelty, layer interaction and constraint. A significant
regression equation was found (F8,8 = 3.634, p <0.05), and could account for
2
78.42% of the variance in the outcome (RAdjusted = 0.5684). Only form was a
significant predictor of satisfaction and could account for a unique proportion of
variance in the outcome of 15.86%.
It was often reported that using AMN4FI induced reflection on one’s be-
liefs about what is normally meant to collectively improvise. In recalling the
succession of events and interactions that took place, players were prone to re-
membering in detail the sequential unfolding of the formation and dynamical
shaping of the musical clusters they were part of. They often commented that
AMN4FI brought to awareness the fact that behind each green node there were
players and that, by virtue of having committed to those particular relation-
ships, they had to maintain focus with those people. Below is a set of responses

426
Proc. of the 13th International Symposium on CMMR, Matosinhos, Portugal, Sept. 25-28, 2017

from participants, which might help showing whether AMN4FI has met its goal
in questioning unconscious improvisational and musical procedures/practices, as
well as in suggesting alternative ways to structure the sonic output at both micro
and macro levels.
I think it was quite a successful system to me, in terms of making me
think much more about who I was interacting with and how I was in-
teracting with them, which in a normal group of improvisation it kind
of happens, of course, but I do not really think about it so much [. . . ]
(participant, New River Studios, London, 04.09.2016)
[. . . ] I look at it as something positive, because it puts me outside of
these things that I do, that are more easy to do with my saxophone. Here,
with the computer and this program, this instrument, I have to think
with a bit of distance about what I usually understand as improvisation,
collective improvisation. (participant, I’klectik, London, 20.07.2016)
From some of the comments it appears that AMN4FI did suggest a sustained
focus on local interactions, for arguably longer than it would have otherwise hap-
pened if players had followed their long-established musical habits. At times, this
was perceived very clearly and participants were able to formalise this property
and reflect on it.
But there is a light switched on. Oh, yes yes, I had better carry on.
(participant, New River Studios, London, 04.09.2016)
I guess I kind of have to sustain it, you know, support that person for
longer which I probably would have not done, if it did not tell me to do
that, [. . . ] I consciously remember thinking I had better stick with it.
(participant, New River Studios, London, 04.09.2016)
More broadly, AMN4FI represented a useful way to access processes which
occur during improvisation and that might remain under cover or unconscious.
This is probably a factor that can potentially contribute to the initial awkward-
ness that some players experience upon their first acquaintance with the system.
It does mirror quite a lot of the social behaviours that happen in free
improvisation anyway, so you are right, maybe that is just uncomfortable
in itself. Perhaps, a resistance to the idea of being confronted with things
that are normally much more under the surface. (participant, New River
Studios, London, 04.09.2016)
When no familiarisation or rehearsal time was possible, as in the case of the
Bath Spa University instance, participants showed a much higher resistance and
questioned the need to have a system in place. It is the author’s contention that
this can be greatly alleviated, if not resolved, given sufficient time to practice
with AMN4FI and understand its a↵ordances. Four players in total (including
the author) participated in two of the three instances of AMN4FI and for them
the result of this mini longitudinal study was quite dramatic.

427
Proc. of the 13th International Symposium on CMMR, Matosinhos, Portugal, Sept. 25-28, 2017

[. . . ] we have already rehearsed that, and played a gig with this, so,
coming back to it for the second time, I feel much more comfortable
with using it as a system. (participant, New River Studios, London,
04.09.2016)

And I realised I had quite a lot of resistance to it at first, I did not see
the point, I am used to improvising without such tools, and then by the
end of the performance I realised that it made it really easy to form a
group in the space of no time at all, a whole bunch of people who did
not even know each other, well some did, but, you know, nine people
or something, and what came out was some very subtle and very struc-
tured free improvisation where nobody was in charge. It was completely
egalitarian. (participant, New River Studios, London, 04.09.2016)

The quote above not only highlights the change of perception with respect to
AMN4FI but also its e↵ectiveness in coordinating and aggregating large groups
of players according to a decentralised schema. This particular property of facili-
tating cohesive music making amongst potential strangers, was noticed by other
participants, too.

So it was almost instant gratification, [. . . ] a music accelerator. (partic-


ipant, New River Studios, London, 04.09.2016)

6 Conclusion
AMN4FI presented the players with a game situation designed to harness the
potential of a network formation based on inferential reasoning. Depending on
whether sufficient time for familiarisation and rehearsal was provided, AMN4FI
proved a useful framework for players not only to examine their long-established
behaviours and habits in collective free improvisation for large groups, but also
a novel way to structure interaction at a systemic level. AMN4FI showed that
it is possible to employ model-based strategies for large groups of free impro-
visers, without being explicitly normative or directive, but instead providing an
underlying construct which can inform the texture, form and architecture of the
musical output. A perceived sense of form or musical coherence appeared to be
at the core of the players’ evaluation as to whether they considered their impro-
visations to be satisfactory and/or successful. Notwithstanding some limitations
with regard to the design of the interface and the repercussions that this had on
the constraint imposed on the players, AMN4FI encouraged the questioning of
some of the fundamental axioms of free improvisation. Furthermore, AMN4FI
suggested modalities of interaction that were reported as unusual and novel, with
respect to the normal practice of the participants. Not only AMN4FI managed
to coerce structural coherence and multiple sonic layers, but it did so abiding by
a completely decentralised and egalitarian paradigm, where by all players could
be agents of change by explicitly prompting local clusters to take on musical
assignments and negotiate roles and functions.

428
Proc. of the 13th International Symposium on CMMR, Matosinhos, Portugal, Sept. 25-28, 2017

References
1. Bailey, D.: Improvisation: Its Nature and Practice in Music. Perseus Books Group
(1993)
2. Fell, S.: Report on the Composition of Improvised Music no.4, http://www.
brucesfingers.co.uk/texts/rubberneck28.html Accessed 23 February 2017
3. Butch Morris, http://www.conduction.us Accessed 23 February 2017
4. London Improvisers Orchestra, http://www.londonimprovisersorchestra.co.uk
Accessed 24 February 2017
5. Brackett, J.L.: John Zorn: Tradition and Transgression. Indiana University Press
(2008)
6. SoundWire, https://ccrma.stanford.edu/groups/soundwire/ Accessed 22
February 2017
7. Borgo, D.: Negotiating Freedom: Values and Practices in Contemporary Improvised
Music. Black Music Research Journal 22, 165-188 (2002)
8. Osborne, M.J., Rubinstein, A.: A Course in Game Theory. MIT Press (1994)
9. Gilles, R.P.: Wealth Creation and the Social Division of Labour: Networks, Institu-
tions and Trust. Lecture Notes, Queen’s University Belfast (2016)
10. Gilles, R.P., Lazarova, E., Ruys, P.H.M.: On Socio Economic Roles and Speciali-
sation. Economic Papers 25, 157-170 (2006)
11. Nunn, T.E.: Wisdom of the Impulse: On the Nature of Musical Free Improvisa-
tion, http://intuitivemusic.dk/iima/tn_wisdom_part1.pdf Accessed 20 Febru-
ary 2017
12. Havryliv, M.: Playing with Audio: The Relationship Between Music and Games.
Master’s thesis, University of Wollongong, Wollongong, Australia (2005)
13. Mckinney, C.: Metagaming: Experiments with Art and Games. Master’s thesis,
Mills College, Oakland, CA, USA (2009)
14. Kalonaris, S.: Markov Networks for Free Improvisers. In: Proceedings of 42nd In-
ternational Computer Music Conference, Utrecht, pp. 182-186 (2016)
15. Sucar, L.E.: Probabilistic Graphical Models - Principles and Applications. Springer
London (2015)
16. Koller, D., Friedman, N.: Probabilistic Graphical Models: Principles and Tech-
niques. MIT Press (2009)
17. Sato, S.: Motivic Improvisation: Approach and Analyses of Two Selected Works.
D.M.A. thesis, University of Washington, Seattle, USA (1996)
18. Stenström, H.: Free Ensemble Improvisation. Ph.D thesis, University of Gothen-
burg, Gothenburg, Sweden (2009)
19. Toop, D.: Search and Reflect: the Changing Practice of Improvisation. New Sound:
International Magazine for Music 65, 119-129 (2008)

429

You might also like