Angus Lee L.R.A.M | May 2024 | Hong Kong
Parts of this text were edited and reworked from an earlier conference paper titled In modo molto giocoso: Play, Creativity and Subversion in Music (2024), which was presented at the HKU international symposium, AI, Music, and Creativity: At a Crossroads, in March 2023.
“Can machines think” was the formulation that opened Alan Turing’s seminal 1950 article, “Computing Machinery and Intelligence.” 1 What I propose in this article, by way of appropriating Turing’s question as a springboard, is to substitute “think” with “play”, and ponder – in a deliberately naïve fashion – what “play” means when deployed in conceptualising musical creation and performance. We might pursue this line of inquiry by focusing on the question of agency, that is to say, who is it that is playing. Does the constitutive difference in agential subjectivity (i.e. machinic or human) lead to qualitatively different forms of music- making, and with what criteria might we meaningfully assess and compare them? To lay the foundation for our analysis, we might begin by reading the question “can machines play” in a number of ways; yet, to do so, we must return to the question of “who”, and contextualise it by asking the question of framework, that is to say, what is it that is being played? 2
In the first instance, since the raison d’être of any given machine resides in its capacity in efficiently and optimally fulfilling specific functionalities, we might postulate that the operational logic underlying a machinic subjectivity is teleological – that is, goal-oriented – in nature. Considering “play” in tandem with “games”, we might observe that machinic entities are generally highly adept at forms of “play” that concerns the strategic annihilation of an opponent. One might recall here the examples of AlphaGo or Deep Blue,3 machinic entities that defeated their real-world human opponents. In the games of both go and chess, a potently calculative mind, one that not only enumerates the possible combinations of moves ahead but also, by way of pattern recognition, deduces and predicts what response the opponent might offer, is vital for securing victory. The historical defeats of Sedol Lee (go) and Garry Kasparov (chess), therefore, seem to confirm the fact that human forms of play are – paraphrasing philosopher Miguel Benasayag – prone to erroneousness in their being underscored by affectivity and corporeality. 4 Not only, then, can machines play, but that, given the circumstances, they will play to win.
In asserting that machines can play, however, one is already subtly rewriting the premise of the initial question; to put it differently, such a response sheds the question of its veneer of rhetorical naiveté, unveiling instead its hidden, injunctive nature: it affirms the machinic capacity for play as it questions the conventional – and assuredly anthropocentric – definition of “play”. In affirming machines as capable of playing, one is simultaneously adopting a skeptical stance as to whether humans might remain capable of playing, given the latter’s substantively altered definition after embracing a machinic – teleological – form of operational logic. If “play” can be accounted for as nothing more than a series of deductive and predictive manoeuvres, it risks creeping dangerously close to being synonymised with execution: a notion that thrives on rebelliousness, invention and creativity thus becomes inscribed within the etymological fabric of its opposite, privileging instead conformity, precision and utility. If a muted, subcutaneous hypothesis were to run parallel to our mediation on “play”, it would necessarily take the disguise of yet another tentative line of query; rather than seeking out the manners in which musical forms of “play” may be proclaimed as antonymous to its counterpart in games, we might profit instead by venturing to expose the secret affinities that unite the two in allegiance.
According to its conventional definition, composition implies a measured, calculated, and strategic putting-to-use of musical materials; this is exemplified by the terminology characteristic of the field’s pedagogical parlance, such as when one speaks of an economic use of musical material, a functional structural development, or an optimal formal design befitting the nature of its material. Any elements of “play” that orbit a composition, therefore, inevitably succumb to latter’s gravitational pull; viewed through a teleological framework, even musical events intended as ruptures within a composition’s unfolding logic are denied their centrifugal, differentiating tendencies, and are instead subsumed and appropriated as an organic developmental stage of the composition itself. What of the performing musician? Since the musical notation primarily functions as a repository of information, a musician’s vocation would merely consist of ensuring said information’s functional execution. A more “informed” musician might contextualise an interpretation by studying a work’s historical, stylistic and aesthetic coordinates, though still ostensibly operating from within a work’s limit. The playing – interpretation – of a work becomes the actualisation – execution – of a text. If the formulation above appears alien from our common perception of what composition and performance are, might it not be clear that we are hypothesising on what composition and performance might be from a machinic perspective?
One might hyperbolically assert that, in its appropriating exactitude as a prime objective, machinic performers would always be more “faithful” to the musical score than any human performer: given that a machinic reading of music is, by default, informational in nature, the notation is recognised – but, one might hypothesise, cognised – on a literal plane, with the musical symbols translated, so to speak, into numeric instructions as part of an operational sequence of actions. Such an operational sequence does not require interpretation but merely execution: for the machine, execution is performance, the latter being qualified chiefly by its exactitude and velocity. With regard to velocity, in particular, we might observe that while any self-respecting human musician would (rightfully) refuse to perform without sufficient prior preparation, machinic performers prove highly adept in this particular arena. Since notational symbols in music can be deciphered as numeric information, a machine recognises a musical text by literally “counting” it, and it is this numericised “counting” – as opposed to recounting – that enables a machine to not only execute a musical text with exactitude and faithfulness, but also with minimal lag between “input” (the learning of the text) and “output” (the execution of the text). If a machinic performer, therefore, appears more efficient than a human performer, we may attribute it to the fact that, for the machine, music-making qua execution consists merely of measuring, quantifying and counting.
1 Alan Turing, “Computing Machinery and Intelligence,” Mind, Vol. 59, No. 236 (October 1950), 433–460.
2 We must equally differentiate, on a related tangent, that which is being played and that which is being played with, as they clearly refer to wholly different modes of operative thinking.
3 See Melanie Mitchell, Artificial Intelligence: A Guide for Thinking Humans (London: Pelican Books, 2020).
4 Miguel Benasayag, The Tyranny of Algorithms: Freedom, Democracy and the Challenge of AI (Paris: Europa Editions, 2021), 15.
Under this numerico-informational conceptualisation, the musical notation is essentially understood as consisting of a finite number of facts; subsequently, through analysing and compiling them as data, an accurate picture of the musical “reality” may be derived. To postulate something as factual, however, implies that the conclusion is not only arrived at by means of analytic procedures, but that it should also be qualitatively definitive, stopping, perhaps, only a few steps short of what might be called “truth”. According to this deterministic – if not scientistic – logic, the notation’s musical “reality” contains all that can be and needs to be known by a performer. One might register – though certainly not without suspicion – the affinity that links this stance to the opening propositions of Ludwig Wittgenstein’s Tractatus, that not only is “[t]he world […] all that is the case,” but that it is “determined by the facts” in their “being all [sic] the facts”. Such a reductive reading that takes Wittgenstein’s statement as substantive validation for a machinic reading of music, however, must be rejected. If that the notation-in-itself is the absolute, self-sufficient “reality”, any attempt in re-presenting it risks compromising substantive integrity of the music; and if there is no going “behind” and “beyond” the notation, then we would reach the tautological conclusion that any performative intervention in music – human or machinic – would not only be functionally superfluous, but indeed logically impossible; or, as we might, again borrowing from the Tractatus, concede to the fact that what we cannot execute without substantially altering, “we must pass over in silence”.
Since machinic forms of musical play – executions, for want of a shorthand – are clearly incompatible with their human counterparts, one might inquire – in a bid to escape this Wittgensteinian lacuna – in what way, then, might we consider play as more akin to the interpretation, the opposite of execution? I recall here Italian conductor Riccardo Muti’s remark that “[b]ehind the notes there is the infinite… the universe of possibilities”. Despite its truistic nature, the remark remains useful for our analytic purpose, since it assumes the premise that, firstly, not only is a notation’s factual execution not the endpoint in a sequence of action, but rather only the beginning. Clearly, the word “beginning” here hints at the notation being interpreted as a polymorphous, loose set of pre-conditions, out of which an innumerable, quasi-infinite number of ideational elaborations may be spawned; while such elaborations refer to the notation as the source, the latter is understood as suggestive, not imperative, in nature, speaking in terms of possibilities instead of prohibitions, and instead of rooting out the unforeseen and the unexpected, it remains hospitable to them. The notation ceases to be a singularity, but, rather, a modularity: where the machine sees unity, the human sees divergence; whereas the machine pursues the notation as the goal, the human pursues the notation as a process; where the machine sees what a notation is, the human sees what a notation can be. Whereas a machinic execution is teleological in nature, a human interpretation would be ontological in nature, the latter being distinguished in its ability in asking not only “what is” and “what can be”, but, in a playful – if not subversive – tenor, the question of “what if?” In other words: in music, human performers are not only capable of the recognition of meaning in a musical text, but, crucially, the cognition of its potential. If a machinic execution is fulfilled by “counting”, then human interpretation is fulfilled by “recounting”, that is, to give an account of the text. Marvin Minsky, in all likelihood, does not realise the poignancy of his remark that “easy things [for humans] are hard [for machines]”; as we progressively uncover the limits of machinic forms of play, the key to answering to the question “can machines play?” also become progressively clear.
At this point, I would like to contextualise my abstract formulation by offering but one example, which, incidentally, illuminates the paradox in musical interpretation that the death of a work – its becoming obsolete, “overplayed” or “out of fashion” – can often be traced to a misguided form of fidelity and filiation (to remain “faithful” to the composer’s style, to the composer’s intentions, to the indications on the score, etc.); on the contrary, given the circumstances, one might come closer to the “spirit” of a work by “betraying” the notation. Glenn Gould’s (in)famous rendition of Mozart’s Piano Sonata no. 11 in A major, for instance, springs to mind here, and we shall limit our discussion to the work’s first movement; to set the scene, allow me the luxury of a brief descriptive diversion. In short, Mozart defied a number of the conventions characteristic of the genre of sonata in a single, masterful stroke: not only was this first movement, marked andante as a tempo directive, not a fast movement by any stretch of the imagination, but that it is not composed in sonata form, but rather consists of a set of theme and variations. Furthermore, the distribution of tempo variance throughout the movement is highly irregular and disproportionate: the leisurely andante tempo is maintained throughout the theme and the first four variations, which would amount to well over half of the movement’s duration; the fifth variation sinks to an adagio before rebounding to a spritely allegro that concludes the movement. There is a third, final structural oddity: each of the seven sections of the movement lasts exactly 18 measures, each subdividing into an eight-plus-ten two-part structure that is, as Mozart instructs, to be repeated; compounded with the highly repetitive nature of the musical material, this sonata’s first movement appears extraordinary cumbersome by Mozart’s standards. In this case, what might an adequate interpretative response that rises to the structural challenges set forth by this piece be? ′
The article will continue in the next issue of Vantage.
5 Ludwig Wittgenstein, Tractatus Logico-Philosophicus, trans. D. F. Pears & B. F. McGuinness (London & New York: Routledge, 2001), 5.
6 Wittgenstein’s original formulation in proposition 6.54 is “[w]hat we cannot speak about we must pass over in silence”. See Wittgenstein, Tractatus, 89.
7 Conducting Mahler, directed by Frank Scheffer (Allegri Film, 1996), 12:43 to 12:46. https://www.medici.tv/en/docu mentaries/conducting-mahler.
8 This analogy is borrowed from Byung-chul Han, who posits that “[d]igital culture is based on the counting finger. In contrast, history means recounting. It is not a matter of counting, which represents a posthistorical category”; furthermore, since “[t]oday, everything is rendered countable […] [to] be transformed into the language of performance and efficiency”, all that “resists being counted ceases to be”. See Han, In the Swarm, 35.
9 Mitchell, Artificial Intelligence, 357.
10 Mozart’s formal treatment of the movement is, one must admit, no less intriguing than Gould’s interpretation of the piece.