The Intersection of Artificial Intelligence and Classical Music Composition

The Intersection of Artificial Intelligence and Classical Music Composition: A Comprehensive Inquiry into Creativity, Authenticity, and New Methodologies Abstract This article undertakes an exhaustive and multi-faceted investigation into the burgeoning, complex, and often contentious intersection of artificial intelligence (AI) and the deeply rooted tradition of contemporary classical music composition. It seeks to systematically unpack the profound aesthetic, philosophical, and methodological questions that emerge from this synthesis, advancing the central thesis that AI’s role critically transcends that of a mere computational tool to become a powerful conceptual catalyst that is actively reshaping the very foundations of artistic practice. The inquiry commences by delineating the fundamental duality of AI’s application in music, meticulously distinguishing between the paradigm of autonomous content creation embodied by Generative AI and the synergistic, collaborative paradigm of Assistive AI, which is architecturally designed to foster human-machine co-creativity (Atanacković 2024). It then plunges into the core of the aesthetic and philosophical debates, performing a deep, critical analysis of the “AI composer bias”—a potent psychological phenomenon that reveals latent audience prejudices rooted in culturally ingrained, historically contingent perceptions of human effort, creative intentionality, and artistic authenticity (Shank et al. 2022). Through a granular, technically-informed analysis of specific algorithms—ranging from large-scale transformer models like MuseNet and GPT-2 to real-time interactive systems like Somax 2 (Somax2 2023) and novel neural audio synthesizers like DDSP—and their detailed application within the compositional portfolio of Robert Laidlow, this paper identifies, categorizes, and explicates a new set of emergent compositional methodologies, including “interlocking,” “hidden layers,” and “collaging” (Laidlow 2022). Furthermore, it critically examines the “conservative bias” inherent in AI models trained on vast corpuses of historical data, juxtaposing this powerful mimetic tendency with the subversive, norm-defying, and system-breaking teleology of the musical avant-garde (Valkenburg 2024). The analysis then pivots to explore how direct engagement with algorithmic logic inspires radical new formal paradigms, such as a data-driven “musical structuralism” and the novel, non-linear temporalities of “algorithmic time,” which fundamentally challenge the linear, narrative-driven forms that have dominated Western music (Laidlow 2022). Ultimately, this article posits that AI’s most transformative and enduring contribution to music may not lie in the sonic artifacts it produces, but in its profound capacity to function as an idea—a conceptual framework that compels composers to critically rethink the very nature of musical material, structure, temporality, and authorship. The paper concludes by situating AI within a broader socio-ethical context, framing it as a cultural “hyperobject” that necessitates a critical re-evaluation of authorship, copyright (Davis 2023), intellectual property, and the precarious future of creative labor in an increasingly automated world.

1. Introduction: A New Paradigm in Musical Creation and Thought Throughout its intricate and layered history, the evolution of Western classical music has been a story of constant, dynamic negotiation with technology. Each technological leap—from the codification of neumatic notation in the medieval era that allowed for the preservation and transmission of complex polyphony, to the development of well-tempered tuning systems that unlocked the full spectrum of modern harmony, the mechanical perfection of the pianoforte that birthed a new, expressive virtuosity, and the seismic shifts introduced by magnetic tape, electronic synthesis, and digital audio workstations that dissolved the very distinction between composition and sound recording—has acted as a powerful catalyst, fundamentally altering aesthetic priorities, compositional methodologies, and the very sonic possibilities available to the artist. Today, artificial intelligence represents the latest, and arguably most profound, of these technological ruptures. It stands as a paradigm shift that promises not merely to augment or streamline existing creative workflows but to challenge and potentially dismantle the most foundational axioms of Western art since the Enlightenment: the primacy of the individual authorial genius, the nature of originality, and the very locus of the creative act itself. The integration of AI into the composer’s sanctum has ignited a fiercely polarized discourse, a dialectic that oscillates between utopian visions of a technologically-liberated creative renaissance and dystopian fears of artistic deskilling and human obsolescence. This inquiry is dedicated to moving beyond such speculative binaries. Its aim is to ground the discussion in a meticulous, evidence-based examination of how contemporary composers are actively, critically, and creatively engaging with AI—not as a monolithic, deterministic force, but as a diverse and evolving ecosystem of tools, systems, conceptual frameworks, and unexpected collaborators. By analyzing concrete compositional strategies and their tangible aesthetic consequences, we can begin to chart the contours of this nascent artistic landscape and understand how it is reshaping the future of musical thought. 1.1 The Foundational Duality of AI in Music: Generative Autonomy and Assistive Synergy To holistically comprehend the multifaceted impact of AI on musical composition, it is imperative to first recognize that its application is not uniform. Rather, it manifests along a functional and philosophical spectrum defined by two primary poles: the generative and the assistive. This distinction, as articulated by researchers like David Atanacković (2024), is not merely technical; it delineates two fundamentally different visions of the human-computer creative relationship, one centered on algorithmic autonomy and the other on synergistic collaboration. • Generative AI: The Pursuit of the Autonomous Agent: This paradigm represents the ambitious, long-standing quest for algorithmic autonomy in art, a tradition that stretches back to historical curiosities like the Musikalisches Würfelspiel (musical dice games) attributed to Mozart and finds its more rigorous 20th-century footing in the stochastic, probability-driven compositions of Iannis Xenakis and the pioneering computer-generated Illiac Suite by Lejaren Hiller. Modern Generative AI, however, operates at a level of complexity orders of magnitude greater. Systems in this category, often built upon sophisticated deep learning architectures like Generative Adversarial Networks (GANs) or the currently dominant transformer models, are designed to create new musical content de novo. The process begins with training on vast, curated corpora of existing music—for instance, the complete MIDI catalog of J.S. Bach’s chorales or thousands of hours of Romantic-era string quartets. Through this intensive training, the algorithms learn not explicit, human-readable rules of music theory, but the underlying statistical probabilities and complex, high-dimensional relationships that govern a given style’s patterns, harmonic syntax, rhythmic conventions, and formal structures. The ultimate objective is for the model to internalize this stylistic “DNA” so thoroughly that it can generate novel compositions that are not only stylistically plausible but also exhibit internal coherence and, in some cases, a semblance of creative invention. The ambition here is profound: to position the AI not as an instrument, but as a synthetic composer, an autonomous agent capable of authoring complete works in its own right (Laidlow 2022). • Assistive AI: The Co-creative Paradigm: In stark philosophical contrast, Assistive AI operates under a paradigm of human-machine partnership. Its design ethos is not to supplant the human composer but to augment and amplify their innate creative faculties. This approach is fundamentally aimed at enhancing human “creativity, efficiency, and knowledge” (Atanacković 2024, 15), functioning as a force multiplier for human imagination. Assistive systems operate as intelligent, interactive collaborators, capable of performing a diverse range of sophisticated tasks on demand: proposing harmonically rich continuations for a melodic line, generating a cascade of compelling variations on a core thematic motif, intelligently orchestrating a piano sketch for a full symphony orchestra, or providing an intuitive interface for exploring and sculpting complex, multi-dimensional timbral spaces. This model is architecturally designed to foster a dynamic of co-creativity, where the human artist remains the undisputed director and final arbiter of all aesthetic judgments. They guide the process, prompt the system, curate its outputs, filter its suggestions, and thoughtfully integrate the most promising machine-generated elements into their overarching artistic vision (Laidlow 2022). This creates a powerful, iterative feedback loop where human intuition and aesthetic sensibility guide the raw computational power of the machine, forging a symbiotic partnership that aims to achieve results neither could accomplish alone. This functional duality thus reframes the central debate from a simplistic “human vs. machine” contest to a more sophisticated and productive exploration of the new creative ecologies that emerge when human cognition is deeply intertwined with intelligent computational systems. 1.2 The Philosophical Divide: AI as Genius Tool or as Existential Threat? The integration of AI into a domain as profoundly associated with the ineffable qualities of the human spirit as classical music has inevitably created a deep philosophical schism, a debate that touches upon the very definition of art. The discourse centers on the ontological status of AI-generated music and whether a non-conscious algorithm can ever be considered a legitimate source of meaningful artistic expression. The arguments on both sides are deeply felt and draw from long-standing traditions in aesthetics and philosophy of mind. The critical perspective is deeply rooted in a humanist and intentionalist theory of art, which posits that a work’s value is inextricably and necessarily linked to its origin in a conscious, feeling, and thinking mind. From this viewpoint, music is fundamentally a form of profound communication—an intricate vessel for the transmission of human ideas, emotions, worldview, and lived experiences from one consciousness to another. As one online commentator eloquently captures this sentiment, audiences seek out music “for and by people, expressing human ideas with intention” (number9muses 1y ago). An AI, as a non-conscious, non-sentient entity, is axiomatically devoid of beliefs, desires, passions, memories, or the subjective, phenomenal experience (qualia) of the world. It cannot possess genuine intention. Therefore, its output, no matter how structurally sophisticated or superficially beautiful, is relegated to the status of a hollow simulacrum, a highly complex “remix or montage” of the statistical patterns it absorbed from its training data. The argument logically follows that an AI-generated Mozart symphony would be a technically flawless but ultimately meaningless artifact, a brilliant imitation of syntax without a grasp of semantics. It could replicate the style of Mozart, but it could never access the subjective wellspring of joy, sorrow, love, and loss that animated the original composer’s work and gave it its enduring meaning (number9muses 1y ago). This argument finds a strong parallel in John Searle’s famous “Chinese Room” thought experiment, which argues that a system manipulating symbols according to a set of rules does so without any real “understanding” of the meaning of those symbols. Conversely, the defense of AI’s role in composition necessitates a fundamental reframing of music’s function and a repositioning of the AI as a revolutionary new class of artistic instrument. This perspective challenges the romantic prerequisite of music as a “soul scanner,” proposing instead a more formalist or functionalist understanding of it as a meticulously engineered “vibration pattern” designed to elicit specific and predictable responses—emotional, mnemonic, or kinesthetic—in a human listener (EntrepreneurOld7792 6mo ago). Within this framework, the AI is not a ersatz composer but an unprecedentedly powerful and versatile tool, an extension of the human composer’s will and imagination, whose legitimacy is no more inherently questionable than that of a Steinway grand piano or the Ableton Live digital audio workstation. The ideas realized through the AI are not considered the AI’s; they are understood as the human operator’s. The final musical product is seen as a direct and unmediated reflection of the myriad creative decisions made by the human: the choice of a specific algorithm, the careful curation of the training data, the artful crafting of prompts and parameters, and, most importantly, the rigorous selection, editing, and contextual arrangement of the generated output (DerIntrigant 1y ago). In this model, creative agency remains unequivocally and firmly in human hands. The AI serves as an incredibly advanced conduit for expression, a “thought partner” that can expand the composer’s palette of possibilities and lead them down creative paths they might never have discovered on their own. This debate echoes the historical anxieties that surrounded the invention of photography in the 19th century, which was initially dismissed by many as a purely mechanical process incapable of “true art,” a charge it eventually overcame through the demonstrated intentionality of the artists who wielded it. 1.3 Perceptions of Authenticity and the Empirically Demonstrated “AI Composer Bias” The philosophical skepticism surrounding AI-generated music is not merely an abstract, academic debate; it manifests as a consistent and measurable psychological phenomenon that has been termed the “AI composer bias.” A pivotal empirical study conducted by Shank et al. (2022) provided definitive evidence for this bias by demonstrating that listeners’ aesthetic judgments are profoundly and systematically influenced by their beliefs about a work’s provenance. In their experiments, when listeners were informed that a piece of music was composed by an AI, they consistently rated it as significantly less likable, less enjoyable, and less emotionally moving than when they believed the exact same piece of music was composed by a human (Shank et al. 2022). This finding is critical because it isolates the source of the negative judgment not in the objective sonic properties of the music itself, but in the contextual information surrounding its creation. This bias appears to be driven by a confluence of deeply held cultural narratives and cognitive heuristics concerning the nature of art and the meaning of authenticity. Two primary psychological factors can be identified: • The Effort Heuristic and Procedural Value: Humans possess a powerful cognitive bias that leads them to assign greater value to outcomes they perceive as requiring more effort, struggle, or time to achieve. The creation of art, particularly within the classical tradition, is culturally coded as an arduous process involving years of dedicated training, intense intellectual labor, emotional turmoil, and personal sacrifice. This “story of creation” imparts a procedural value to the artwork. AI, in stark contrast, is perceived as a machine that produces complex music instantaneously and effortlessly, a “push-button” solution. This perceived lack of effort triggers the heuristic, leading to an automatic devaluation of the final product, as the work is seen as fundamentally “unearned” (Shank et al. 2022). The music is divorced from the romanticized struggle that, in the minds of many listeners, gives it its value and meaning. This resonates with Walter Benjamin’s concept of the “aura” of a work of art, which he argued stems from its unique existence in time and space, its history, and its physical creation—qualities that a digitally generated, infinitely reproducible AI work is perceived to lack. • The Intentionality Stance and the Search for Meaning: Humans are social creatures who are hardwired to seek out the “mind behind the message.” When engaging with an artwork, listeners instinctively adopt what philosopher Daniel Dennett calls an “intentional stance,” searching for the meaning, emotion, and narrative intent of the creator. The knowledge that a work was generated by a non-conscious algorithm effectively short-circuits this fundamental interpretive process. It thwarts the search for meaning, leading to a perception that the music is an empty shell—technically proficient but emotionally and semantically void. This demonstrates that aesthetic appreciation is not a simple response to sensory data but a complex cognitive act, heavily mediated by our “theory of mind”—our assumptions and beliefs about the mental state and intentions of the entity we believe to be the creator (Shank et al. 2022). We listen not just to the music, but through the music to the mind behind it. Crucially, the study by Shank et al. (2022) revealed that this bias is highly genre-dependent. It is at its most potent and pronounced in genres like classical music, where the cultural expectation is for a profound, personal “authorial voice” and a deeply felt human narrative. The entire tradition is built upon the cult of the genius composer. Conversely, the bias was found to be significantly weaker or even non-existent in genres like experimental electronic music, where algorithmic processes, technological mediation, and the de-emphasis of a personal authorial voice are already core components of the established aesthetic. This finding strongly suggests that the AI composer bias is less a reaction to the sound of AI music and more a reaction to a perceived violation of genre-specific cultural contracts between the creator and the listener. 1.4 The Emergence of New Methodologies for a Co-creative Process In response to both the glaring technical limitations (such as a lack of long-term coherence) and the unique creative affordances of current AI systems, a new generation of composers is moving beyond the naive paradigm of simple, one-shot “generation.” They are developing sophisticated and nuanced strategies for integrating AI into their creative practice. The work and writings of composer Robert Laidlow, in particular, provide a valuable taxonomy of these nascent methodologies. These techniques are not mere workarounds; they represent genuine aesthetic strategies that transform the AI from a volatile black-box generator into a flexible, responsive, and integral part of the compositional apparatus (Laidlow 2022). • Interlocking: The Cyborg Duet: This technique is a direct and elegant solution to one of the most persistent failings of many generative models: their inability to maintain structural coherence and narrative logic over extended periods. Instead of abdicating creative control and tasking the AI with composing an entire piece, the composer engages it in a turn-based, iterative dialogue. The process resembles a duet: the human composes a musical phrase or section, which is then fed into the AI as a prompt. The AI generates a series of possible continuations. The composer then acts as a curator, selecting the most musically interesting or surprising output, editing and refining it, and then composing their own response to the machine’s idea. This cycle repeats, creating a tightly woven fabric of human- and machine-generated material. This method allows the human composer to maintain firm control over the piece’s large-scale architecture and narrative trajectory while strategically leveraging the AI for moments of unexpected melodic invention, harmonic novelty, or textural surprise, resulting in a true “cyborg” composition (Laidlow 2022). • Hidden Layers: Algorithmic Scaffolding and Conceptual Genetics: This represents a far more abstract, subtle, and structurally profound method of AI integration. In this approach, the AI-generated material is deliberately withheld from the audible surface of the final composition. It is not intended to be heard directly by the listener. Instead, it functions as a form of conceptual scaffolding or a genetic blueprint that informs the work’s deepest structural levels. For example, a lengthy harmonic progression generated by an AI could be extracted and used as the foundational framework for an entire symphonic movement, with all the surface melodic and rhythmic material being composed by the human. Alternatively, a complex rhythmic pattern generated by an AI could be used as a temporal template to govern the durations and proportions of large-scale formal sections. In this methodology, the AI’s contribution is embedded in the work’s very DNA, acting as a secret, organizing force that shapes the music from behind the scenes (Laidlow 2022). • Collaging: The Embrace of Generative Entropy: This method takes what is often considered a technical flaw of AI models—their tendency towards incoherence, fragmentation, and non-sequitur—and transforms it into a deliberate aesthetic choice. The composer generates a large number of short, discrete, and often unrelated musical cells using one or more AI models. They then layer and superimpose these fragments, much like a visual artist creating a collage, to build up dense, complex, and often chaotic sound masses. This technique is particularly effective for creating indeterminate and aleatoric textures that defy traditional analysis. It represents an artistic embrace of generative entropy, finding beauty and expressive potential in the randomness and unpredictability of the algorithmic process, yielding sonic results that would be nearly impossible to conceive or meticulously notate through traditional compositional methods alone (Laidlow 2022). These strategies, taken together, signal a significant maturation in the artistic engagement with AI, one that is defined by critical inquiry, strategic intervention, and the architecting of novel human-machine creative systems.
2. Algorithms, Techniques, and Their Deep Integration into the Compositional Process A robust and meaningful discussion of AI’s role in music necessitates a move from abstract philosophical debates to a concrete, technically-grounded examination of the specific tools being utilized. The unique architectures and operational logics of different algorithms directly shape, enable, and constrain artistic outcomes. This section provides a detailed overview of the key AI algorithms and systems at the forefront of contemporary composition, followed by an in-depth, analytical exploration of their application in the works of Robert Laidlow. This analysis will forge explicit links between specific technologies and the compositional methodologies of interlocking, hidden layers, and collaging, demonstrating how artistic strategy and technological capability are deeply intertwined. 2.1 Symbolic and Generative AI Tools: The Architects of Abstract Musical Notation Symbolic AI represents a paradigm that operates not on the physics of sound waves, but on the abstract, structured language of music itself. Its domain is symbolic data, most commonly MIDI (Musical Instrument Digital Interface), which encodes discrete musical events such as pitch, onset time, duration, and velocity. This approach treats music composition as a problem of sequence generation, akin to natural language processing. • MuseNet and the Transformer Revolution: Developed by OpenAI, MuseNet stands as a landmark example of a large-scale generative model for music, built upon the revolutionary “transformer” architecture that has come to dominate the field of AI. Trained on a massive and diverse dataset of MIDI files spanning centuries of Western musical practice, it learns deep, multi-layered stylistic connections. The key innovation of the transformer architecture lies in its “self-attention mechanism.” Unlike previous recurrent neural network (RNN) models that process data sequentially, one step at a time, the attention mechanism allows the model to weigh the importance of all previous notes in the sequence simultaneously when predicting the next note. This ability to create direct connections between distant points in a musical piece overcomes the “long-term dependency” problem that plagued earlier models, resulting in a significantly higher degree of long-term structural coherence and stylistic consistency. Despite this technical leap, MuseNet’s output often reflects the statistical mean of its vast training data. It excels at producing stylistically plausible and often charming melodies and conventional harmonic progressions, but can struggle to generate the kind of deep, developmental, and structurally surprising logic that characterizes masterworks of the classical canon (Laidlow 2022). Its strength lies in stylistic emulation, its weakness in true structural innovation. • Clara and the Legacy of Recurrent Neural Networks (RNNs): An earlier LSTM-RNN (Long Short-Term Memory Recurrent Neural Network) model used by Laidlow, Clara serves as a perfect foil to MuseNet, exemplifying both the initial promise and the ultimate limitations of this older architectural paradigm. LSTMs and other RNNs are inherently sequential; they process a musical piece one note at a time, maintaining an internal “memory state” that is passed along the chain. While theoretically capable of learning long-term patterns, in practice they suffer from the “vanishing gradient” problem, where the signal of influence from distant past events becomes too weak to affect current predictions. This technical limitation has direct and audible aesthetic consequences: a tendency for the model to get “stuck” in short, repetitive loops of notes or phrases, and to produce passages of music that are harmonically and rhythmically static, lacking any sense of forward momentum or development. Furthermore, these models showed a high propensity for “overfitting”—essentially memorizing specific patterns from the training data rather than learning generalizable stylistic principles—which further contributed to their lack of creative flexibility (Laidlow 2022). • GPT-2 and Its Successors as Conceptual Engines: While primarily renowned as a natural language processing model, OpenAI’s Generative Pre-trained Transformer 2 (GPT-2) and its more powerful successors share the same fundamental transformer architecture as MuseNet. Their unprecedented power lies in their ability to understand context and generate remarkably coherent and stylistically flexible sequences based on a given textual prompt. Laidlow’s use of GPT-2 in his work Alter is particularly insightful because it represents a sophisticated, lateral application of the technology. He employed it not for the direct generation of musical notes, but as a conceptual sparring partner and an engine for ideation. By feeding the model thematic prompts related to the piece’s subject matter (such as authenticity and artificiality) and analyzing the generated text, he used the AI to uncover unexpected poetic imagery, to forge novel conceptual juxtapositions, and to develop a rich literary and philosophical substrate that would then be translated into purely musical terms by the human composer (Laidlow 2022). This represents a highly abstract and artistically mature use of AI, not as a content generator, but as a tool for augmenting conceptual creativity itself. 2.2 Assistive and Co-Creative AI Systems: The Improvising and Sound-Sculpting Partners In contrast to the “offline” generation of symbolic models, this class of AI systems is explicitly designed for dynamic, real-time interaction, positioning the AI as a responsive, improvising agent or a sophisticated sound-sculpting tool within a live performance or studio environment. • Somax 2: The Co-improvising Agent: Developed and refined at IRCAM, the world-renowned French institute for music and acoustic research, Somax 2 represents the state-of-the-art in interactive systems for human-machine co-improvisation (Somax2 2023). Its operational paradigm is fundamentally different from that of large, pre-trained models like MuseNet. Somax 2 operates on a “corpus-based” model of learning. The human musician “teaches” the system in a bespoke manner, by providing it with a small, carefully curated selection of their own musical material—be it recorded improvisations, composed phrases, or even fragments of existing pieces. The AI analyzes this personal corpus and builds a unique generative model based exclusively on its specific stylistic features and syntax. In a performance setting, it then actively listens to the human musician’s live playing via a microphone, analyzes the incoming audio in real-time, and generates stylistically congruent musical responses that can either mimic, accompany, or counterpoint the human’s contribution. This architecture fosters a genuine improvisational dialogue and perfectly exemplifies the principle of distributed agency, where creative control flows fluidly and dynamically between the human and machine performers in a tight, responsive loop (Somax2 2023). • DDSP (Differentiable Digital Signal Processing): The Neural Sound Sculptor: A truly groundbreaking project from Google’s Magenta research team, DDSP forges an unprecedented and powerful bridge between the abstract, high-level control of neural networks and the concrete, low-level reality of digital audio synthesis. The term “differentiable” is the key innovation. It means that the entire audio synthesis pipeline is constructed in a way that its parameters (such as the fundamental frequency of an oscillator, the amplitude of its harmonics, and the characteristics of a noise filter) can be directly controlled and optimized by a neural network using the same backpropagation techniques used to train the network itself. This gives the composer direct, intuitive, and learnable control over the most fundamental components of timbre. In Laidlow’s Silicon Body, DDSP was used to perform radical, real-time style transfer on audio signals. The system was trained to extract a simplified representation of pitch and loudness from one sound source (the “Source” layer) and use it to control the DDSP synthesizer, which was itself trained to recreate the detailed timbral characteristics of a different sound (the “Target” layer) (Laidlow 2022). This allowed for a form of sonic puppetry, imprinting the timbral DNA of one sound onto the performance gestures of another. • Raw Audio Generation (WaveNet, SampleRNN, RAVE): The Direct-to-Waveform Synthesizers: This family of deep learning models represents the most direct and computationally intensive approach to sound generation. Rather than creating a symbolic representation like MIDI, these algorithms generate the audio waveform itself, sample by sample, typically at a rate of 16,000 or 44,100 samples per second. Models like WaveNet (from DeepMind) and SampleRNN use autoregressive, recurrent architectures to predict the value of each individual audio sample based on all the previous samples. RAVE (Real-time Audio Variational autoEncoder) uses a different, more efficient autoencoder architecture to learn a compressed “latent representation” of a sound, which can then be decoded to reconstruct the audio. In Alter, Laidlow experimented with these models in an attempt to mimic the human voice, but found their true power lay in their ability to generate novel and uncannily realistic orchestral textures and complex, evolving electronic soundscapes (Laidlow 2022). Because they operate at the most fundamental level of sound, these tools excel at creating sounds with a rich, organic, and minutely detailed quality that often eludes traditional synthesis methods. 2.3 An In-Depth, Analytical Examination of Applied Techniques in Laidlow’s Works • Interlocking Applications—The Weaving of Human and Machine Agency: The second movement of Three Entistatios serves as a powerful case study in the interlocking methodology, not merely as a technical workaround but as the central formal and aesthetic principle of the piece. The movement is constructed as an explicit, audible dialogue. One can almost trace the process: Laidlow composes a lyrical, expressive phrase for the ensemble; this phrase is then fed into the AI, which responds with a cascade of continuations, perhaps denser, more chromatic, or rhythmically fragmented. Laidlow then acts as a discerning editor, selecting the most compelling 2-3 seconds of the AI’s output, excising the rest, and seamlessly grafting it onto his original phrase. He then composes his own musical response to the AI’s fragment, and the cycle begins anew. The result is a seamless and coherent musical tapestry where the very notion of a single author is deliberately problematized. The seams between human intention and machine-generated probability are artfully blurred, creating a final work that is qualitatively different from what either the human or the machine could have produced in isolation (Laidlow 2022). In Alter, this same technique is brilliantly re-imagined along a vertical, rather than horizontal, axis. A live mezzo-soprano sings a phrase. This audio is captured by a microphone, fed into a real-time AI system which analyzes its pitch and contour, and then generates an immediate electronic response that either imitates, harmonizes, or provides a strange, distorted echo of the vocal line. This creates a dynamic, unpredictable, and theatrical game of real-time call-and-response, a live duet between a human singer and her algorithmic shadow (Laidlow 2022). • Hidden Layers Applications—The AI as Structural Architect: The conceptual and structural core of the large-scale work Alter is built upon a profound and elegant application of the hidden layers methodology. An AI model was prompted to generate a series of simple, unadorned melodies. However, instead of using any of these melodies directly in the piece, Laidlow subjected them to a rigorous harmonic analysis. Through this process, he discovered a recurring statistical anomaly in the AI’s output: a persistent and structurally significant tonal ambiguity centered on an oscillation between the pitches A-flat and A-natural. This seemingly tiny fragment of AI-derived information—this “harmonic seed”—was then extracted and elevated by the human composer to become the central structural and harmonic principle for the entire 30-minute composition. The tension between these two microtonally adjacent pitches permeates every aspect of the work, from its moment-to-moment harmonic language to its large-scale formal design, acting as a secret, unifying, and generative force. In this instance, the AI’s contribution is not on the audible surface at all; it is entirely conceptual. The AI has functioned as an unwitting structural architect, providing the foundational blueprint upon which the human composer built the entire edifice (Laidlow 2022). • Collaging Applications—The Aestheticization of Algorithmic Chaos: The first movement of Three Entistatios provides a vivid demonstration of collaging as a deliberate aesthetic strategy. For this piece, Laidlow used generations from the early, highly unpredictable ‘Clara’ model, collecting dozens of short, disjointed, and often musically nonsensical fragments. In the final composition, these fragments are not presented in isolation but are layered atop a more conventionally composed, coherent musical texture performed by the live ensemble. The aesthetic effect is one of sonic overload, of controlled chaos. The AI-generated layers function as a kind of musical “glitch,” a chaotic and irrational force that constantly disrupts and destabilizes the logic of the underlying human-composed material. This technique directly mirrors 20th-century avant-garde practices like the visual collages of Kurt Schwitters or the musical quotations in Luciano Berio’s Sinfonia. It is a strategy that artistically re-purposes the technical shortcomings of the AI—its lack of coherence, its tendency for non-sequitur—and transforms them from a bug into a feature, from a technical flaw into a source of profound aesthetic interest and expressive power (Laidlow 2022).
3. Authenticity, Aesthetic Conflicts, and the Pervasive Conservative Bias of AI The integration of artificial intelligence into the deeply humanistic tradition of musical composition forces a direct and often uncomfortable confrontation with our most deeply ingrained aesthetic values. This section critically interrogates the complex web of ethical and artistic judgments that surround AI’s creative output, focusing with particular intensity on the culturally charged perception of authenticity and the powerful, pervasive, and structurally inherent tendency of current AI models toward stylistic conservatism and mimetic fidelity. 3.1 The Perception of Artistic Authenticity: A Confluence of Effort, Intentionality, and the Romantic Ideal The empirically verified “AI composer bias” is far more than a mere statistical curiosity; it serves as a crucial diagnostic tool, offering a window into the core set of values and expectations that audiences bring to the experience of art. The consistent devaluation of music believed to be AI-generated stems from a perceived deficit in what is nebulously termed “authenticity,” a concept that is itself a complex tapestry woven from historical narratives, cultural beliefs, and cognitive biases (Shank et al. 2022). • The Enduring Shadow of the Romantic Ideal: The aesthetic framework of Western art music, particularly since the 19th century, has been overwhelmingly dominated by the Romantic ideal of the artist. This powerful cultural narrative posits the artist as a heroic, quasi-prophetic individual whose work is valuable precisely because it is a direct, unmediated expression of their unique inner world, their personal struggles, and their profound emotional landscape. This narrative frames the creation of significant art as a process of immense effort, personal sacrifice, spiritual torment, and intellectual wrestling. Art is valued not just for its final form, but for the story of its arduous creation. AI, by its very perceived nature—impersonal, unemotional, computational, and, above all, instantaneous—stands in stark and irreconcilable opposition to this deeply embedded narrative. The belief that an AI can produce a complex symphony with little to no “effort” triggers a powerful cognitive heuristic that automatically devalues the final product, as the work is perceived as fundamentally “unearned” (Shank et al. 2022). The music is divorced from the romanticized struggle that, in the minds of many listeners, gives it its value and meaning. Walter Benjamin’s concept of the “aura” of a work of art, which he argued stems from its unique existence in time and space and its history, is highly relevant here; an AI-generated piece, infinitely reproducible and without a human “hand,” is perceived as lacking this aura. • Art as an Act of Intentional Communication: A prevailing and deeply intuitive view in aesthetics is that art is a privileged form of meaningful communication. An artwork is understood as a vessel, a conduit intentionally crafted by one conscious mind to transmit a complex payload of ideas, emotions, perspectives, and experiences to another conscious mind. When a listener is told a work was created by an AI, this entire communicative circuit is perceived to be broken. The listener becomes acutely aware that the “composer” had no subjective feelings to express, no personal story to tell, no philosophical point to make, and no communicative intent whatsoever (number9muses 1y ago). The music is thus immediately re-contextualized as an empty signifier, a syntactically correct but semantically void arrangement of sounds. This reveals a crucial insight: our aesthetic appreciation is not, and perhaps has never been, a pure response to sensory stimuli (the objective qualities of the sounds themselves). It is a complex, heavily mediated cognitive act that is profoundly shaped by our “theory of mind”—our assumptions and beliefs about the mental state and intentions of the entity we believe to be the creator (Shank et al. 2022). 3.2 Mimetic Fidelity and the Structurally Inherent Conservative Bias of AI A critical, and often insufficiently examined, aspect of AI development in the creative arts is the very methodology used to define and measure its success. The dominant paradigm for the validation of AI-assisted music composition (AIMC) has been, and continues to be, one of mimesis—that is, success is measured primarily by the AI’s ability to successfully and convincingly imitate existing human styles. • The Turing Test as a Flawed Artistic Benchmark: The implicit, and often explicit, goal of many academic and commercial AIMC projects is to create a system that can pass a musical variation of the Turing test. The system is deemed successful if its output—a chorale in the style of Bach, a sonata in the style of Mozart—can fool a panel of expert human listeners into believing it is a genuine, human-composed work. While this represents a formidable technical and engineering challenge, the adoption of this test as the primary benchmark for artistic success has profound and deeply problematic aesthetic consequences. It structurally enshrines imitation, stylistic plausibility, and normative conformity as the highest possible virtues, while implicitly penalizing novelty, originality, and stylistic deviation, as such traits would immediately reveal the work’s non-human origin. In the context of creativity theory, this approach validates only “exploratory” creativity (generating novel combinations within an existing conceptual space) while being incapable of assessing “transformational” creativity (changing the rules of the space itself). • The Inevitable Result: A Deep-Seated Conservative Bias: This validation methodology creates a powerful and self-perpetuating conservative feedback loop. AI models are trained on historical data, and they are evaluated based on their ability to reproduce the statistical regularities of that data. This process architecturally predisposes them to produce music that aligns with the statistical mean of their training corpus and is readily “human-acceptable.” This inevitably and inexorably pushes the entire field of AI composition toward familiar, tonally consonant, and structurally conventional styles. It incentivizes the creation of music that is aesthetically safe, predictable, and fundamentally unchallenging, effectively transforming AI into a high-technology engine for producing generic, pastiche, or what might be termed “easy listening” classical music (Valkenburg 2024). • The Irreconcilable Conflict with the Avant-Garde: This structurally inherent conservative tendency places the current paradigm of AIMC in direct and irreconcilable philosophical opposition to the entire historical project of the musical avant-garde. The central ethos of the avant-garde tradition—from the atonal revolutions of the Second Viennese School, to the aleatoric experiments of John Cage, to the complex systems of Brian Ferneyhough—has been one of perpetual critique, radical subversion, and the deliberate, often violent, dismantling of established musical conventions and systems of meaning. The avant-garde seeks to systematically challenge and frustrate audience expectations, to question the very definition of what constitutes “music,” and to push the boundaries of sonic expression into entirely new and uncharted territory (Valkenburg 2024). While AIMC, guided by the mimetic logic of the Turing test, strives for assimilation and acceptance within established frameworks, the avant-garde thrives on antagonism and the negation of those very frameworks. This fundamental clash highlights a critical limitation of current AI architectures: while they can achieve a masterful command of existing stylistic languages, they are not, as currently conceived, architecturally or philosophically equipped for the kind of critical, self-reflexive, and norm-violating thought that is the very engine of radical artistic innovation and paradigm shift (Valkenburg 2024). 3.3 Style Transfer and Authenticity as a Deliberate Aesthetic Theme: The Critical Case of Silicon Body The powerful capacity for AI-driven style transfer, far from being a mere technical novelty, can serve as a potent artistic tool for a direct and critical engagement with these very themes of authenticity, control, and technological mediation. Robert Laidlow’s Silicon Body, particularly its second movement, stands as a masterful and deeply unsettling example of this artistic strategy. The piece is constructed upon a clear and powerful conceptual premise: a “Target” layer, consisting of recordings of viscerally familiar and engaging human dance music (styles like jazz, techno, and folk), is audibly and graphically controlled and distorted in real-time by an abstract, inscrutable, and fundamentally inhuman “Source” layer generated via the DDSP algorithm (Laidlow 2022). The listener’s experience is meticulously designed as a journey of gradual estrangement and dawning horror. The piece begins with sounds that are rhythmically engaging and stylistically recognizable. However, as the movement progresses, these familiar sounds are progressively bent into increasingly unnatural and grotesque shapes. Rhythms are warped and stretched in physically impossible ways, instrumental timbres are twisted into alien forms, and stylistic signifiers are deconstructed and reassembled into nonsensical configurations. The music is intentionally engineered to induce a profound sense of the “uncanny valley,” that unsettling perceptual space where something appears simultaneously familiar and deeply, fundamentally alien. This process is designed to make the listener feel, with growing certainty and discomfort, that the music is profoundly inauthentic, a hollow puppet being animated by the unseen strings of a strange, non-human intelligence (Laidlow 2022). Laidlow explicitly connects this carefully crafted aesthetic experience to the pervasive, contemporary anxieties surrounding digital authenticity. The musical process becomes a direct metaphor for the disorienting effects of deepfake videos, the invisible hand of social media algorithms shaping public discourse, or the feeling of being controlled by systems we cannot see or understand. The piece thus brilliantly weaponizes the concept of inauthenticity, transforming it from a perceived flaw of AI music into the central expressive and critical theme of the work itself, using the technology to offer a powerful commentary on the very condition of living in a technologically saturated age (Laidlow 2022).
4. Algorithmic Structuralism, Repetitive Logics, and a New Conception of Musical Time Prolonged, deep engagement with the tools and processes of artificial intelligence does more than just provide a composer with new musical material; it begins to suggest, and in some cases demand, entirely new ways of conceiving of musical structure, form, and the fundamental experience of time. The internal logic of algorithms—their statistical nature, their iterative processes, their relationship to data—can serve as a powerful source of inspiration for compositional paradigms that lie far outside the traditions of Western music. This section explores how this engagement can lead to a new form of data-driven “musical structuralism,” a radical re-contextualization of repetition, and the emergence of a non-linear, “algorithmic” conception of musical time. 4.1 Musical Structuralism in the Age of AI: From Serialism to Data-Driven Arrays The way AI models, particularly those employing data augmentation techniques, process and understand music—not as a linear narrative of expressive gestures, but as a high-dimensional space of interrelated statistical parameters—finds a fascinating historical precedent in the mid-20th-century movement of musical structuralism and serialism. Thinkers and composers like Milton Babbitt and Pierre Boulez sought to create music where every element—pitch, rhythm, dynamics, timbre—was organized according to a pre-compositional, abstract system or series. AI offers a new, and perhaps more powerful, path to a similar goal. It inherently treats music as an emergent phenomenon, a surface that materializes from the complex relationships between its underlying parameters. This perspective gives rise to a potent concept of infinite transposability: the idea that if the proportional relationships between all the musical parameters remain constant, the fundamental identity and integrity of the musical idea are preserved, regardless of how its surface features are transformed or translated (Laidlow 2022). • Arrays as a Concrete Compositional Method: In his works Gravity and Warp, Laidlow gives this neo-structuralist impulse a tangible and systematic form through the compositional method of “Arrays.” He begins by mapping the core elements of his musical language—pitch sets, timbral profiles, textural densities, and temporal flow—onto a series of independent, controllable data arrays. The resulting musical composition is not “written” or “improvised” in the traditional sense; it is generated as the necessary and inevitable outcome of the programmed interactions and trajectories of these predefined data structures. This radically shifts the locus of the creative act. The composer’s focus moves away from crafting individual notes and phrases and toward the high-level task of designing the underlying system, its constituent parts, and its behavioral rules. In his piece Gravity, this structuralist concept is pushed to its logical and theoretical extreme. The arrays governing the music are programmed to move inexorably towards their absolute endpoints—either maximum intensity and complexity, or total nullity. This process is intended as a direct musical analogy for the theoretical challenges faced by physics when dealing with the extreme scales of singularities and the Big Bang. The point at which all four of the work’s governing arrays converge on their zero value is compositionally and experientially defined as absolute silence, a structural void, the heat death of the musical universe (Laidlow 2022). 4.2 Algorithmic Time and the Radical Redefinition of Repetition The internal processes and temporal logics of AI algorithms offer compelling and fertile new models for musical time that break decisively from the linear, teleological (goal-oriented), and narrative-driven models of temporality that have dominated Western music since the Enlightenment. • Epochs, Training Loops, and Cyclical Acceleration: The training of a deep learning model is an iterative process that occurs in cycles called “epochs.” In each epoch, the model repeatedly processes the data to refine its parameters. The third movement of Three Entistatios is a direct musical translation of this process. The piece is built entirely upon a single, twenty-note musical cell. This cell is then subjected to a relentless iterative loop. With each repetition, it is played faster and at a lower dynamic, musically representing an algorithm honing in on a solution over successive cycles. The form of the piece is the process itself (Laidlow 2022). • Multiple, Co-existing Reference Frames: The concept of a cyclical, algorithmic time is expanded and made more complex in the work Chromodynamics. Here, the same core musical material is presented to the listener simultaneously in several different “reference frames.” Imagine three instrumental groups, each playing the exact same melodic loop, but with each group playing it at a different, unrelated tempo. This creates a complex, disorienting, and richly layered texture where the listener perceives the same musical “event” from multiple temporal perspectives at once, a musical realization of a kind of relativistic time inspired by the parallel processing capabilities of computational systems (Laidlow 2022). • The Aesthetics of No-Foresight: A key architectural feature of the now-dominant transformer models like GPT-2 is that they are purely retrospective in their operation. When generating a new token (be it a word or a musical note), they possess a powerful “attention mechanism” that allows them to “look back” at the entire preceding context, but they have absolutely no ability to plan ahead or aim for a future goal. This profound technical limitation was brilliantly embraced by Laidlow as a guiding formal principle in his piece Disc Fragments. The piece involves a live performer setting a text that is itself being generated in real-time by an AI text generator. The music proceeds, developing its own logic and momentum, and then, at the arbitrary and unpredictable moment that the text generator reaches its pre-set character limit, both the spoken text and the instrumental music halt instantly and brutally, often mid-word or mid-phrase. This creates a jarring, unsettling, and profoundly anti-narrative conclusion that perfectly and powerfully mirrors the non-teleological, goal-less, and purely process-driven nature of the underlying algorithm (Laidlow 2022). 4.3 The Warping of Historical Form and the Logic of Branching Narratives AI’s influence is not limited to the creation of entirely new forms; it can also be used as a powerful tool for the deconstruction, critique, and radical reimagining of the most established and canonical forms in Western musical history. • The Palindromic Sonata: A Form on Twin Axes: In his composition Silicon Mind, Laidlow takes on one of the most storied, narrative-driven, and teleological forms in all of Western music: the sonata form. The process begins with using MuseNet to generate a body of musical material in the high-classical style of Mozart. This material is then carefully arranged by the human composer into the conventional sections of a sonata structure: exposition, development, and recapitulation. However, at this point, he applies a radical, AI-inspired temporal logic. The primary structural pillars of the form—the exposition and the recapitulation—are presented to the listener simultaneously in both their original forward direction and in a perfect, note-for-note retrograde (backwards) version. This superimposition of temporal directions creates a kind of musical palindrome, a structure where the music’s past and future are collapsed into a single, static present. This act of temporal sabotage utterly subverts the sonata’s inherent reliance on linear progression, dramatic tension, memory, and anticipation, replacing narrative with a crystalline, mirrored, and fundamentally non-directional structure (Laidlow 2022). • Diverging Axes and the Aesthetics of the Multiverse: Silicon Mind also draws deep inspiration from a key behavioral feature of generative AI: its capacity to produce a multitude of different, unique, and equally valid outputs from a single, identical prompt. This computational behavior is translated by Laidlow into the musical concept of “branching narratives” or “diverging axes.” The piece will proceed along a particular musical path, presenting one complete musical idea (representing one possible AI generation). Then, a sharp, non-pitched percussive cue (a ratchet and a kick drum) acts as a signal for a “reset” or a “branch point.” The music then jumps back in time to an earlier point in the piece and proceeds down a completely different developmental path, presenting an alternative musical continuation (representing a different AI generation from the same prompt). This creates a fractured, disorienting, and multiverse-like listening experience, where the audience is presented not with a single, definitive musical narrative, but with a series of alternative “what if” scenarios (Laidlow 2022). This formal structure, born directly from the one-to-many architecture of generative technology, finds parallels in the aesthetics of hypertext fiction and the branching narrative structures of modern video games.
5. Conclusion: Synthesizing the Role of AI and Charting Speculative Future Directions This exhaustive and multi-pronged inquiry has sought to demonstrate that the role of artificial intelligence within the sophisticated domain of contemporary musical composition is not a simple or singular phenomenon. It is a profoundly complex and multifaceted reality that resists easy categorization. AI simultaneously and fluidly occupies the roles of a pragmatic tool, an agent-like creative partner, and, most potently and revolutionarily, a source of transformative, paradigm-shifting ideas. Its accelerating integration into the creative process does not merely offer new efficiencies or sonic palettes; it necessitates a fundamental and often uncomfortable re-evaluation of our most cherished and long-standing assumptions about art, the nature of creativity, and the very definition of the human-technology relationship in the 21st century. 5.1 The Tripartite Role of AI: A Synthesis of Tool, Agent, and Conceptual Catalyst It is a categorical error to attempt to define AI’s creative function with a single, static term. Its identity is fluid, context-dependent, and defined by the intention of the artist who wields it. At its most basic level, AI can be wielded as an incredibly powerful tool, a computational assistant for automating laborious tasks or generating a vast body of raw material. This view, while accurate, captures only its most superficial application. Interactive systems like Somax 2 push the technology firmly into the realm of an agent-like collaborator, an interactive partner in a creative dialogue (Somax2 2023). Yet, as the nuanced reflections of composer Robert Laidlow himself suggest, the label of a true “collaborator” may still be premature and anthropomorphically misleading. True collaboration implies a shared understanding, a capacity for critical judgment, debate, and the mutual development of conceptual goals—faculties that current AI architectures, which lack consciousness and genuine comprehension, do not possess (Laidlow 2022). Therefore, this inquiry posits that the most artistically significant and historically enduring role for AI in composition may be that of a conceptual catalyst or an idea generator in the most abstract sense. The very processes, architectures, and inherent limitations of the algorithms—their non-linear relationship to time, their probabilistic rather than deterministic logic, their constitutional lack of foresight—have become an incredibly fertile ground for the germination of entirely new compositional theories and aesthetic philosophies. Concepts like algorithmic time, data-driven structuralism, and branching narratives are not merely sounds generated by AI; they are abstract, high-level ideas about the fundamental organization of music that are directly inspired by the technology’s inner workings (Laidlow 2022). In this profound sense, AI’s greatest impact may not be on the notes we ultimately hear, but on the very way composers think about the elemental building blocks of their art. It is a technology that, like the printing press or the internet, has the power to reconfigure the entire intellectual and creative ecosystem in which art is made. 5.2 AI as an Ethical and Social Hyperobject: Navigating the Unseen Challenges The profound implications of artificial intelligence extend far beyond the composer’s studio. It can be productively understood through philosopher Timothy Morton’s concept of a “hyperobject”—a phenomenon so massively distributed in time and space that it defies easy comprehension, yet has tangible and profound local effects. AI, as an abstract, globally-scaled system, fits this description perfectly. Its integration into our culture brings with it a host of formidable and urgent ethical challenges. • Unresolved and Urgent Ethical Quandaries: The proliferation of high-fidelity AI-generated content has created a legal and ethical minefield. The fundamental question of copyright for AI-generated art remains largely unresolved, with landmark legal cases still being fiercely debated (Davis 2023). Who is the legal “author” of a work created in partnership with an AI? The user who provided the prompts (DerIntrigant 1y ago)? The corporation that developed the AI? Or the multitude of original creators whose data was used for training? These questions strike at the heart of our systems of intellectual property and artistic accountability (Davis 2023). • The Precarious Future of Creative Labor: The potential for AI to automate creative tasks poses a significant threat to the livelihoods of many working musicians. While it is unlikely to replace top-tier composers, its ability to generate high-quality “functional” music at scale could decimate fields like production music libraries, which provide scores for film trailers, advertisements, and video games. This necessitates a broader societal conversation about the value of human artistry and the economic structures needed to support it in an age of intelligent automation. 5.3 The Future of Creativity: A Vision of Structured Uncertainty and New Artistic Horizons The path forward for a meaningful artistic engagement with AI lies neither in a Luddite rejection nor a naive, utopian embrace. Instead, it demands a posture of critically informed, technically fluent, and creatively adventurous exploration. This involves a sustained investigation of the productive tensions and dualities it brings to light: the Real and the Fake, the Future and the Past, the System and the Secret. The most promising paradigm for future development appears to be that which can be termed “structured uncertainty.” This concept describes a co-creative system that judiciously balances the randomness and indeterminacy of generative processes with the constraints, patterns, and guidance provided by a human user (The Shape of Surprise 2025). Interactive, real-time systems that operate on this principle, like the co-improvising agent Somax 2, offer a compelling glimpse into a future of enhanced, symbiotic creativity that empowers rather than replaces the artist (Somax2 2023). Future research and artistic practice must continue to push the boundaries of AI’s integration into large-scale forms like orchestral music. Yet, it is equally crucial to explore what this technology might mean at a more intimate scale—to ask what “chamber music in an age of high technology” might look and sound like. By fostering these smaller, more responsive, and conceptually-driven interactions between human minds and algorithmic processes, composers can harness the full potential of artificial intelligence, not as a mere novelty, but as a genuine tool for forging the vital and unheard musical languages of the 21st century and beyond.

Bibliography Atanacković, David. 2024. “Artificial Intelligence: Duality in Applications of Generative AI and Assistive AI in Music.” INSAM Journal of Contemporary Music, Art and Technology 12: 12–31. Davis, Ben. 2023. “U.S. Copyright Officials Just Issued Their First Definitive Ruling on A.I. Art—and It’s Not Good News for Artists.” Artnet News, February 23, 2023. DerIntrigant. 1y ago. Comment on “AI is Composing Classical Music Now—Genius Tool or Total Disaster?” Reddit. EntrepreneurOld7792. 6mo ago. Comment on “AI is Composing Classical Music Now—Genius Tool or Total Disaster?” Reddit. Laidlow, Robert. 2022. “Artificial Intelligence Within the Creative Process of Contemporary Classical Music.” PhD diss., Royal Northern College of Music and Manchester Metropolitan University. number9muses. 1y ago. Comment on “AI is Composing Classical Music Now—Genius Tool or Total Disaster?” Reddit. Shank, Daniel B., Courtney Stefanik, Cassidy Stuhlsatz, Kaelyn Kacirek, and Amy M. Belfi. 2022. “AI Composer Bias: Listeners Like Music Less When They Think It Was Composed by an AI.” Journal of Experimental Psychology: Applied 28, no. 4: 760–73. Somax2 – A Distributed Co-Creative System for Human-Machine Co-Improvisation. 2023. Paper presented at the HHAI 2023 Conference. The Shape of Surprise: Structured Uncertainty and Co-Creativity in AI Music Tools. 2025. Paper to be presented at the AIMC Conference. Valkenburg, Govert. 2024. “Validation of AI and avantgarde music composition.” Paper presented at the EASST-4S 2024 Amsterdam Conference.