At a first glance, it might seem unlikely to find any common properties in robots' behavior, cell dynamics and baroque music performance. Yet, from a systemic perspective they all share an essential and crucial emergent phenomenon: their dynamics is the result of the interaction between their structure and the environment. In this perspective paper, we discuss this property and build upon it to outline a guiding principle for artificial creativity.
Prelude
“[. . .] every behavior is emergent. You cannot program a behavior into the system. You
need the physical body and the minute you have a physical body you have an interaction
with the environment.” (Rolf Pfeifer [
The fundamental role of embodiment in cognition is currently acknowledged among
cognitive scientists and AI scholars [
In all the three cases mentioned above, the observed dynamics is the emergent result
of the interactions between the system and its environment. Here we refer to the broad
definition of emergence as a characteristic of a system whose behavior cannot be solely
reduced to its properties: also its interaction with the environment (i.e. what is external
to the system and interacts with it) must be taken into account for describing and possibly
understanding its behavior [
Emergence has been widely advocated as an enabling condition for creativity, artificial
or not. As for artificial creativity, a case in point is the use of evolutionary computation
techniques for generative art [
In this section we discuss the main features of three paradigmatic examples of emergent behavior. We rfist illustrate the case of embodied robots, followed by cell dynamics and we conclude with baroque music performance. The aim is to conduce the reader to appreciate the common core of these examples so as to identify a unifying abstract model that can be used also as a guiding principle for artificial creativity.
As the description of the following three examples is necessarily succinct and we only address and discuss the features relevant for the perspective we propose, the indulgent reader will forgive us for discarding other important elements of the systems we discuss.
In a keynote speech titled “The Emergence of Cognition from the Interaction of Brain,
Body, and Environment” [
We can state that the behavior of a robot emerges from the interaction among robot’s
control program, robot’s body and the environment. In this context, emergence is not an
all-or-nothing property, but it can be observed at varying degrees: from simple behaviors
arising from the interaction between a simple pure reactive system and its environment [
For simplicity we consider only two entities: the controller and its physical embodiment,
which includes both the body of the robot and the environment. The controller is an
implementation in some formal language or mathematical notation of a structured set
of instructions that maneuvers the actuators, which in turn act on the efectors of the
robot.2 The key point here is to observe that the environment, robot’s body included,
constrains the physical outcome of the instructions of the controller. Therefore, we
have a source of indications (provided by the controller) that manifest themselves3 by
being filtered and transformed through (physical) constraints. The role of constraints
in creativity will be discussed in section 4. Here we emphasize the fact that the role of
physical constraints, rather than being a limiting factor, somehow enables the actions
of the robot and drives them towards specific spaces of possibilities. For example, a
rubber foot on a wooden surface enables a diferent space of possibilities than that of a
plastic foot, which exerts negligible friction on the terrain. Therefore, a crucial role of
the environment on robot’s behavior is that of providing enabling constraints [
So far we have considered a single direction in the causal chain observed in the system:
the control program sends signals to the actuators, which maneuver the efectors that in
turn act on the physical world. However, the efects of controller’s actions mediated by
the body of the robot and the environment have a feedback on the robot itself through
1The absence of emergence characterizes those cases in which the robot interacts with an almost completely
deterministic environment and robots’s behavior is fully and completely preprogrammed in the controller.
2Actuators are such things as motors, while efectors are the parts that directly interact with the
environment [
In the highly influential book titled “The music of life” [
We first observe that the participation of a portion of DNA to the dynamics of the cell depends on the state of the cell itself and the current external stimuli. For example, we consider a gene as active if the information it provides is currently used for producing a protein. Notably, one gene can code for a set of diferent proteins, therefore taking part to diferent functions, and more genes can code for the same protein.
The activation of a gene also depends on the proteins that are present in the cell, which are in turn afected by the signals that the cell is receiving. Indulging to a computational metaphor—which must be used cum grano salis —we could say that the instructions contained in the genome are read, combined and executed depending on the current state of the system and the inputs it is receiving.
Rather than being the “program of life”, which prescribes the overall dynamics of the system, the genome is a source of descriptions that are interpreted inside a larger context including the environment. In addition, a sequence of instructions can be read and executed in possibly infinite ways depending on the machine we choose to run it. As for recipes describing a procedure to cook a dish, so in cell dynamics there are some implicit rules that the genetic mechanisms exploit. Changing these rules means changing the resulting behavior.
Lastly, a remarkable property of cells is that the chain of causal actions starting from
genes up to the whole organism is not simply unidirectional, but filled with feedback
loops. Notably, the organism itself can constrain and condition gene expression. This is
a wonderful example of downward causation [
While an analogy between robot behavior and cell dynamics is probably not particularly surprising, pushing the analogy to baroque music performance might sound rather bizarre. Yet, this case completes the picture of our perspective and provides a bridge towards artificial creativity.
Baroque music is usually placed in a time frame spanning one century and a half,
approximately from 1600 to 1750. As for all artistic epochs, chronological and geographic
boundaries are blurred, nevertheless it is possible to identify some features of baroque
music that represent hallmarks of this musical style. There are beautiful introductions to
baroque music that we suggest for a musicological, aesthetic and cultural illustration of
this music style [
As for many music styles, baroque music performance involves the interpretation of a score, which represents the music written in a conventional notation. Moreover, every execution is characterized by some emergent phenomena [23]. In the case of baroque music the relation between the score and the music executed is particularly important [24] and it is in fact the trigger for the main emergent behavior we can observe in this music.
A distinguishing trait of baroque music is basso continuo, often simply named continuo.4 The continuo part is mainly played by instruments such as organ, harpsichord, harp, lute, and theorbo and has the role of leading and sustaining [25] the upper parts.5 In a nutshell, basso continuo consists in an extemporaneous completion of a melodic bass line with chords, arpeggios, counterpoint improvisations and other ornaments.6 Here the main element of our analogy comes into play: these improvisations on a ground bass are not free, but they have to be performed according to rules, conventions and preferences of the period. These conventions were both explicit, e.g. detailed in printed books or manuscripts, and implicit, e.g. transmitted from teacher to student by imitation. The same also holds for higher voices, such as the violin: the part written in the score is just a skeleton of the music to be played.
Contemporary performers of baroque music call historically informed performance
practice a philological approach to early music, which includes the interpretation of
the score through the conventions that are supposed to be used in that time [
The discussion of this example is out of the scope of this paper, but it is important to
remark that the original part establishes a structure around which diminutions are played
(e.g. a long note is substituted by several shorter notes). The notation is descriptive and
not prescriptive [
The analogy with robot behavior and cell dynamics is clear: a sequence of instructions (the score) is “run” by a machine that implements specific rules. If the rules change, also the final outcome of the execution does. It is not possible to predict the overall emerging behavior just by knowing the sequence of instructions. As for robots and cells, the music performed is brought to life by constraints and preferences that constitute its environment. Therefore, these constraints enable the music to exist. Furthermore, as an improvisation attitude is fundamental to baroque music performance, these constraints are also a trigger to create new impromptu executions. Finally, we observe that also in music performance we observe downward causation phenomena: once a choice is made according to the constraints, e.g. a specific chord in basso continuo, the current status influences back further choices.
A systemic view of the mechanisms we have exemplified in the previous section is illustrated in figure 2. The system has two basic components: a sequence of instructions and the environment. The interplay between the two gives origin to the dynamics of the system. The dynamics is also the result of feedback loops and downward causation links (represented by the two arrows between the two components).
Changes in just one of these components impact the whole emergent dynamics. Artificial creativity approaches usually do not adopt this “dual system” perspective and focus only on the artifacts, keeping the environment implicit in the interpretation of the descriptions of the artifacts, if not completely neglecting its role. We are not claiming that the approach we illustrate is brand new, as there are applications in computational creativity with similarities with the perspective we propose. For example, in artificial music composition, a Markov model of the music is built from a given repertoire and new pieces of music are generated by adding some constraints deriving from the original or a diferent repertoire [ 28, 29]. In this way, the descriptive part can be kept constant—i.e. the Markov model—while the creative impulse is given by the constraints. However, to the best of our knowledge, there are no works that explicitly adopt the view we discuss and focus on the automated elaboration of the constraints in the environment.
The essence of the computational mechanism we propose to investigate consists in providing a formal, somewhat incomplete, artifact description that is kept constant; this description is interpreted (in a sense, completed) by and “environment” that provides the rules for the actualization of the formal description. This environment is formalized as well and subject to changes operated by an algorithm, such as a genetic algorithm. The scheme of figure 2 is potentially subject to infinite implementations. Here we outline two possible instantiations that come from our current ongoing work, in two representative ifelds, namely music and robotics.
One of the most fascinating areas for artificial creativity is the production of an artifact belonging to one artistic field from an artifact belonging to another one. For example, the generation of a picture starting from a text [30]. Similarly, we can produce music from text. In previous unpublished work, we developed a system that takes a text as input, e.g. a poem, and produces a polyphonic music in form of a canon. In this case, the environment is composed of the algorithmic choices that translate the text into music. The environment is a code—in semiotic terms—that makes it possible to generate the musical artifact from the text. Instead of being defined by hand by the designer, this code can be parametrized and produced algorithmically by means of an optimization method—or any suitable search technique. Relevant parameters can be the number of voices, the starting measure for each voice, the relation between the symbol in the text (e.g. a letter or a word) and pitch and duration of notes. The algorithm iteratively defines this code and produces the music; the iterative changes in the code can be driven by human evaluation of the music produced or a combination of objective functions (e.g. musical properties such as distribution of intervals and generic functions based on information theory measures). In this way, the description provided by the text is translated into a piece of music according to rules that are subject to changes.
As a second example we mention a recent work in robotics showing that the same
controller can be dynamically adapted to accomplish diferent tasks just by adjusting its
connection points with the environment [31]. Here, being creative means being able to
adapt and solve a problem. In this work, robots are controlled by Boolean networks [
The analogy we have illustrated in the previous sections is instrumental for outlining the core elements of a general mechanism, whose potential we believe has not yet being fully explored; we hope that this perspective paper can promote explorations in this direction. Besides the implications for artificial creativity, the focus on emerging dynamics from the interplay between a descriptive system and its environment (including constraints and preferences) inevitably evokes some remarks on creativity in general.
The role of constraints in creativity is overwhelming. We have previously emphasized
the property of constraints to be “enabler”: rather than being a limiting factor, a
constraint triggers innovative solutions and, above all, canalizes the creative efort.
Uncountable are the examples in the art, from metric in poetry to tools used in painting.
This property finds a beautiful example in evolution: when a new species or a variation of
an organism appears, it perturbs the environment and introduces new opportunities (more
precisely, afordances ) for the other life forms. Once a new opportunity is exploited, a new
constraint is created that is in fact an opportunity for new actions and new evolutionary
steps [
The relation between human and artificial creativity deserves a final remark. The recent impressive results of AI systems in generating artworks, such as paintings [33] and music [34], might induce us to think that natural and machine creativity are substantially the same and explore the same spaces of possibilities. It is true that we are often surprised by artworks produced by algorithms, but this does not mean that the two forms of creativity are equal. They might be indistinguishable in practice by an observer, but there is a crucial distinction that has to be remarked: the space of possibilities in artificial creativity is always bound by some choices made by the designer and, furthermore, it relies on machine capability of identifying afordances, which is believed to be rather limited compared to human one [35, 36].
“–You adjusted him?– she shrieked. –But it was he who created my light-sculptures. It
was the maladjustment, the maladjustment, which you can never restore, that-that. . .”
(Isaac Asimov [37]).
“But what evolves cannot be said ahead of time: what evolves emerges unprestatably —I
know of no better word—and builds our biosphere of increasing complexity.” (Stuart
Kaufman [
The author thanks the anonymous referees for interesting comments and useful suggestions for improving the paper. 7“Il sonare senza trilli, over accenti, fuor che ne luoghi dove la prestezza sia tale che non li ametta,e` cosa insipida.” [20] B. Haynes, G. Burgess, The Pathetick Musician, Oxford University Press, 2016. [21] M. Bukofzer, Music in the baroque era-from Monteverdi to Bach, Read Books Ltd, 2013. [22] N. Harnoncourt, Baroque Music Today: Music as Speech, Helm, 1988. [23] G. Minati, Music and systems architecture, in: Proceedings of the 5th European
Systems Science Congress, 2002. [24] B. Kuijken, The notation is not the music: reeflctions on early music practice and performance, Indiana University Press, 2013. [25] A. Agazzari, Del sonare sopra a’l basso con tutti li stromenti, Siena, 1607. [26] J. Veihlan, The rules of musical interpretation in the baroque era, Leduc, 1979. [27] A. Roli, On the complexity of baroque music and implications on robotics and creativity, in: Systemics of Incompleteness and Quasi-Systems, Springer, 2019, pp. 109–116. [28] F. Pachet, P. Roy, Markov constraints: steerable generation of markov sequences,
Constraints 16 (2011) 148–172. [29] F. Pachet, P. Roy, B. Carer,´ Assisted music creation with flow machines: towards new categories of new, in: E. Miranda (Ed.), Handbook of Artificial Intelligence for Music, Springer, 2021, pp. 485–520. [30] DeepAI, https://deepai.org/machine-learning-model/text2img, Last visited October 2022. [31] M. Braccini, A. Roli, E. Barbieri, S. Kaufman, On the criticality of adaptive boolean network robots, Entropy 24 (2022) 1368:1–1368:21. [32] S. Kaufman, Humanity in a creative universe, Oxford University Press, 2016. [33] J.-W. Hong, N. Curran, Artificial intelligence, artists, and art: attitudes toward artwork produced by humans vs. artificial intelligence, ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM) 15 (2019) 1–16. [34] J.-P. Briot, F. Pachet, Deep learning for music generation: challenges and directions,
Neural Computing and Applications 32 (2020) 981–993. [35] A. Roli, J. Jaeger, S. Kaufman, How organisms come to know the world: fundamental limits on artificial general intelligence, Frontiers in Ecology and Evolution (2022) 1035. [36] S. A. Kaufman, A. Roli, What is consciousness? Artificial intelligence, real intelligence, quantum mind and qualia, Biological Journal of the Linnean Society (2022). URL: https://doi.org/10.1093/biolinnean/blac092. doi:10.1093/ biolinnean/blac092, to appear. [37] I. Asimov, Light verse, in: The Complete Stories, volume II, Collins, 1995. [38] G. Kapsberger, Libro quarto d’intavolatura di chitarone, Rome, 1640.