Quantum reinforcement learning shapes musical behaviour in virtual agents

Eduardo Reck Miranda and Moein Fahmideh Vatandoost of the University of Plymouth have developed an “agentic orchestra” that uses parameterised quantum circuits to create musical agents. These agents evolve unique compositional behaviours, termed “mindsets,” through trial and error within a musical environment, diverging from systems reliant on large datasets of existing music. The researchers demonstrate that these quantum computing agents function as adaptive musical partners, developing individual “musical personalities through interaction rather than imitation.” Experiments on IBM quantum hardware confirm musically meaningful learning is achievable even with current technology.

Quantum Reinforcement Learning for Musical Agent Development

Each agent within the “agentic orchestra” is constructed as a trainable quantum circuit, allowing for the evolution of distinct “mindsets”, learned predispositions toward specific compositional behaviours like rhythmic variation or melodic balance. This approach prioritises learning through self-generated musical actions, rather than replication, and enables the exploration of musical decision-making spaces using quantum superposition and entanglement within a hybrid quantum-classical loop.

The researchers implemented two distinct agent architectures to investigate how internal representation impacts musical output; agents of one type process pitch and rhythm independently, while agents of the other type treat pitch and duration as unified “note events.” This architectural divergence allows for a subtle examination of how different information processing strategies shape the resulting musical character. At each learning step, the agent embeds its current musical context into a quantum register, representing multiple potential interpretations simultaneously through superposition.

Agentic Orchestra Framework: Beyond Corpus-Driven Approaches

The agentic orchestra framework departs from conventional computer music composition by prioritizing the evolution of “musical mindsets” over replication of existing styles. Rather than relying on extensive musical datasets, the system encourages compositional behaviours through trial-and-error interaction within a simulated musical environment. This approach, detailed in a forthcoming paper in Computer Music Journal, uses parameterised circuits to implement musical agents, allowing them to explore musical possibilities through quantum superposition and entanglement.

The framework’s design specifically avoids the need for large-scale quantum computing infrastructure, demonstrating functionality on current Noisy Intermediate-Scale Quantum (NISQ) devices and validated through experiments on IBM quantum hardware. The researchers articulated this distinction through the concept of an agentic orchestra, where each agent develops a unique compositional approach shaped by musically interpretable reward functions. These functions, which can be combined and weighted, guide the agents towards specific aesthetic goals, encouraging a diversity of “mindsets” within the ensemble.

The system’s scalability is a key design consideration, with compact encodings, modular agent definitions and hybrid quantum-classical learning techniques enabling operation within current hardware limitations. “We demonstrated this across two complementary architectures,” the authors state, highlighting the system’s flexibility and potential for diverse musical applications. The framework’s emphasis on process-oriented interaction and emergent behaviour offers a compelling alternative to corpus-driven imitation.

Mindsets: Evolving Distinct Compositional Behaviours

The framework establishes a clear distinction between how musical agents process information, building agents that process pitch and rhythm independently and agents that treat pitch and duration as unified “note events” to explore the impact of internal representation on compositional output. This architectural divergence allows for the creation of varied musical textures and styles, shaping the resulting musical mindsets without relying on pre-existing musical datasets.

These reward functions guide the agents toward aesthetic goals like rhythmic variety, melodic balance, or textural smoothness. If multiple rewards are employed, the total reward is calculated as a weighted sum, with user-defined weights allowing for customisation and multi-objective optimisation in the evolution of distinct mindsets; the equation RΣ = sumxλxRx defines this process, where λx represents the weights for different reward measures.

Musically Grounded Reward Functions Shape Agent Learning

The agentic orchestra diverges from traditional approaches by evolving reusable musical mindsets through quantum reinforcement learning, rather than relying on extensive datasets of existing music. This framework introduces a method for shaping compositional behaviours through musically interpretable reward functions, allowing for the development of distinct “mindsets” across multiple agents. These mindsets are not pre-programmed but emerge through trial-and-error engagement with a musical environment, a departure from corpus-driven imitation. This flexibility allows for the creation of agents with varied aesthetic goals and compositional strategies.

Quantum Circuits as Parameterised Musical Agents

Parameterised circuits underpin a novel approach to musical composition, moving beyond reliance on extensive datasets of pre-existing music. This system, termed the “agentic orchestra,” prioritises active engagement with music over passive imitation, a departure from traditional corpus-driven methods.

The compact expressivity of these circuits is particularly valuable in performance contexts, allowing for real-time responsiveness and improvisation. The use of quantum circuits isn’t merely a technological novelty. It supports interactive music where structure emerges organically through adaptation between human performers and the quantum agents.

Demonstrations show how agents generalise their learned behaviour to previously unseen material, operating in interactive and offline batch modes using MIDI-encoded music. The goal is to support creating a collaborative environment where agents contribute unique perspectives and respond to the evolving musical landscape.

The research establishes an experimental foundation for quantum computer music, demonstrating that musically meaningful agentic systems can be realised even under current noisy intermediate-scale quantum (NISQ) conditions. By contributing both a working methodology and a conceptual framework, the work opens avenues for future developments in the intersection of quantum computing and creative expression.

Agentic Orchestra as Adaptive Musical Partner, Not Composer

Each agent’s learned musical behaviour, termed a “mindset,” is determined by trainable parameters within these circuits, allowing for the evolution of distinct compositional approaches. This approach prioritises the development of unique musical personalities through interaction, rather than simply reproducing established styles. The resulting system positions quantum computing agents less as autonomous composers and more as adaptive musical partners, each developing its own musical personality through interaction rather than imitation.

The agentic orchestra was initially conceived for interactive performance with human musicians, but the system also functions in batch processing mode, generating music from existing recordings. “Like the interactive improvisation systems and generative environments that have shaped the field of computer music,” the researchers write, the agentic orchestra is designed as an interactive partner. Exploiting quantum phenomena such as superposition and entanglement enables agents to explore musical decision spaces in ways that traditional algorithms cannot.

They are interested in systems that behave as instruments, environments, or collaborators, rather than replacements for human authors. The team’s work represents what they believe to be the first application of quantum reinforcement learning to computer-aided music composition and performance, building on existing research in grid-world tasks, robotic controls and function approximation scenarios. The system’s potential extends beyond current hardware limitations, as early studies suggest hybrid quantum-classical models may outperform classical deep reinforcement learning as quantum hardware scales up.

The focus on learning from self-generated musical consequences, rather than pre-existing data, is a key differentiator. This approach allows for the exploration of musical possibilities beyond those present in training corpora, encouraging genuinely novel compositional outcomes.

👉 More information
🗞 Beyond Corpus-Driven Imitation: Evolving Musical Mindsets in the Agentic Orchestra via Quantum Reinforcement Learning
✍️ Eduardo Reck Miranda and Moein Fahmideh Vatandoost (Interdisciplinary Centre for Computer Music Research (ICCMR), University of Plymouth, Plymouth, UK)
🧠 ArXiv: https://arxiv.org/abs/2611.24290v1

Stay current

See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.

Avatar of The Quant

The Quant

The Quant possesses over two decades of experience in start-up ventures and financial arenas, brings a unique and insightful perspective to the quantum computing sector. This extensive background combines the agility and innovation typical of start-up environments with the rigor and analytical depth required in finance. Such a blend of skills is particularly valuable in understanding and navigating the complex, rapidly evolving landscape of quantum computing and quantum technology marketplaces. The quantum technology marketplace is burgeoning, with immense growth potential. This expansion is not just limited to the technology itself but extends to a wide array of applications in different industries, including finance, healthcare, logistics, and more.

Latest Posts by The Quant: