AI Music: How Artificial Intelligence is Changing Music Composition

Artificial intelligence is making significant strides in the realm of music, once considered a uniquely human domain. Neural networks are now capable of composing melodies, and their abilities are improving daily. Numerous examples of AI-generated music exist, including the album 'Hello World,' created entirely by AI in collaboration with human musicians. Singer Taryn Southern partnered with the AI algorithm Amper to record the song 'Break Free,' and the Flow Machines project, in collaboration with composer Benoît Carré, created 'Daddy's Car,' a song stylistically similar to The Beatles. AI has even ventured into classical music, with the AIVA network completing Antonín Dvořák's unfinished piece 'From the Future World' and Flow Machines imitating Bach. A neural network from Yandex, along with composer Kuzma Bodrov, composed a piece performed by the New Russia Symphony Orchestra under Yuri Bashmet. This introduction explores the growing influence of AI in music and sets the stage for a deeper dive into its capabilities and implications.

The process of AI music composition mirrors how AI creates other forms of art. First, the neural network learns by being fed vast amounts of musical data. The more data it receives, the better it becomes. Based on this learning, the AI attempts to recreate harmonies similar to those it has heard. While the results may not always be Grammy-worthy, AI can often generate pleasing sound combinations. These successful instances are often highlighted in news reports. According to Alexander Krainov, head of the Machine Intelligence Laboratory at Yandex, generating a piece in MIDI format is relatively straightforward. The algorithm writes numerous musical pieces in a compressed digital format, memorizes common patterns, and records the unique characteristics of the composition in a compact form. It then uses its knowledge of harmonies to reconstruct the piece. While most of the melodies produced may sound like cacophony, a small percentage can be quite acceptable and used as a basis for further development by human musicians.

Stanislav Butovsky, a composer, songwriter, and sound producer, explains that each musical genre has a characteristic set of instruments, arrangement techniques, melodic features, rhythms, and harmonic sequences. These elements can be represented mathematically as a set of parameters that, when specified, can generate a piece in the desired genre. Butovsky emphasizes the close relationship between music and mathematics, stating that the form and stylistic characteristics of a musical piece can be clearly described and represented as an algorithm. For example, if a machine processes data from the music of Steve Reich, Terry Riley, Vladimir Martynov, and Philip Glass, it can generate a piece consisting of repeating patterns with slight variations, characteristic of minimalist music. Similarly, if the machine is fed the entire catalog of a band like Любэ, it can produce songs with a similar set of themes and instrumentation. The machine is indifferent to the type of data it processes, as it has no personal preferences.

The ultimate goal is to create AI that can generate complete musical pieces indistinguishable from those written by humans. Currently, AI can generate short segments of music that sound authentic, but achieving consistency throughout an entire piece remains a challenge. Alexander Krainov believes that this is an open problem with a clear path to resolution, and he anticipates significant progress within the next year. The question of whether someone without musical training can write music using AI is debated. Programmers tend to believe that musical knowledge is not essential, while musicians argue that only those with musical education can fully leverage the technology. Ivan Yamshchikov, an AI evangelist at ABBYY, points out that anyone can already create music, regardless of formal training, and AI simply adds new colors to the musical palette. However, Butovsky argues that while AI can make music creation more accessible, a lack of musical knowledge can lead to mediocre results. He suggests that for AI-generated music to be truly successful, users need a good understanding of musical elements and how they combine.

The issue of copyright and ownership in AI-generated music is complex. Alexander Krainov believes that no rights are infringed when training a neural network, comparing it to a musician listening to numerous pieces of music before composing their own. He suggests that the rights to AI-generated music could belong to the person who selects and uses the music or to no one at all, allowing anyone to use it freely. However, Danil Zhdanov argues that the rights could belong to the owners of the neural network, as legal concepts related to intangible assets like music and software can be applied to AI-generated music.

The most likely future application of AI music is in generating background music for environments where music is not the primary focus, such as lounges, restaurants, and gyms. It can also assist musicians in finding new ideas and overcoming writer's block. Danil Zhdanov notes that some composers are using neural networks experimentally to generate initial material for their themes. AI tools can also be used to create "music constructors," allowing users without musical knowledge to generate music with specific tempos, styles, and transitions. Companies like Ampermusic offer such systems, although they often involve human input to some extent. Alexey Kochetkov, founder of Mubert Inc., believes that generative music has significant commercial potential in areas like hospitality, gaming, and applications that require background music but do not prioritize its source.

Market experts generally agree that AI will not replace the best human composers. Alexander Krainov suggests that AI-generated music can serve as a starting point or foundation for human creativity. He compares it to providing a writer with a basic plot outline to stimulate their imagination. Ivan Yamshchikov draws a parallel to the introduction of MIDI keyboards and digital synthesizers, which did not eliminate the need for pianos and analog synthesizers. He emphasizes that AI can successfully recreate or imitate existing styles, but it is more about creative application than replacement. AI can be another tool in the composer's arsenal, as demonstrated at the Gamma festival in St. Petersburg.

One challenge is that AI-generated music often does not sound like it was created by a human, although this is expected to improve as algorithms are refined. Alexey Kochetkov believes that public conservatism and prejudice hinder the widespread adoption of AI music. Danil Zhdanov points out that human brains are not accustomed to the structures and sequences generated by neural networks. Even when the notes are played with natural samples, AI lacks an understanding of genre classification and the nuances of musical performance. He argues that AI-generated music can be unemotional and unable to trigger moods, suggesting that AI needs teachers to cultivate its musical taste and, more importantly, to instill emotions.

Lera Resser emphasizes the importance of integrating the human element into the concept of AI-generated music. She questions the sense of responsibility in AI and the possibility of a genuine dialogue between an AI creator and its audience. She also wonders how people will interpret music whose development patterns they cannot predict. Resser argues that it is not just about our readiness for AI music but also about what we teach AI. She highlights the differences in decision-making processes between humans and AI, emphasizing the importance of empathy and the social factor. Stanislav Butovsky believes that the ability to think creatively, fantasize, deviate from patterns, and evoke emotions are the key skills that distinguish human musicians from AI. He argues that machines, even those capable of harmonically combining notes and arranging compositions, operate according to pre-programmed algorithms and lack the capacity for imagination and creativity.

While AI continues to evolve in its ability to generate music, the consensus among experts suggests a future where AI and human composers work in synergy. AI can serve as a powerful tool for inspiration, idea generation, and background music creation, while human composers retain their unique ability to infuse music with emotion, creativity, and cultural context. The future of music likely involves a symbiotic relationship between human artistry and artificial intelligence, pushing the boundaries of musical expression in new and exciting ways. The key will be to harness the power of AI while preserving the essential human elements that make music a profound and meaningful art form.