Artificial Intelligence in Music Composition for Media

Written for the University of Hertfordshire. (2022)

Abstract

Artificial Intelligence is a powerful tool designed to enhance human efficiency, and it has augmented productivity and accuracy in many data-related industries as it developed. However, this research paper explores the effect Artificial Intelligence will have on the media industry, which is assumed to rely heavily on human collaboration and creativity. More specifically, the paper explores the effect Artificial Intelligence Visual Artist (AIVA), the first recognised AI music composer, will have on the demand for music composers for media. The paper analyses the differences between composers using AIVA as their ally and AIVA as their competitor to determine how it will affect their employability in the future. The conclusions proposed by the paper allow media composers to better prepare for the disruption AIVA and other AI will create in the industry through an educated prediction on the subject. This makes the research results relevant to music composers, the media industry and everyone involved as creators or consumers.

---

1. Introduction

As the twentieth century had aeroplanes, mobile phones and computers as its groundbreaking technology, Artificial Intelligence (AI) is the pioneering technology of the twenty-first century. Through recent years, we have seen its constant evolution as it has spread around all industries, helping humans be more efficient at their jobs. Some of these examples include smart personal assistants such as Apple’s Siri or Amazon’s Alexa, Covera Health’s ability to recognise patterns more accurately to diagnose patients in healthcare, and Betterment in the finance industry, which uses a robo-advisor fueled by AI that finds the best investment plan for its customers. Humans expect AI to outperform them in data-related jobs and roles where finding trends and patterns is vital to success. However, throughout recent years, AI has started to disrupt an industry which was believed to be faithful to us human beings: the media industry.

The media industry is defined as a group comprising journalism, photography, television, and any form of communication for either entertainment or persuasive purposes (Media, 1995). Hence, as AI developed, the idea of it displacing humans in this industry seemed highly unlikely. AI used to be seen as a technology that followed human instructions through the source code installed in it and could not be creative on its own. However, in recent years, scientists and engineers have evolved AI’s thought process and ability to comprehend abstract concepts by developing different learning methods, making it capable of designing illustrations, editing photographs and composing music.

This research paper will explore the effect AI will have on the media industry. More specifically, AIVA’s effect on the demand for music composers for media. Only a little research has been conducted regarding the subject, as AI, like AIVA, is in the infancy stage. Nevertheless, it is crucial to understand the disruption it will cause in the media industry, as it is an industry assumed to be faithful to human creativity and collaboration.

2. Background

2.1. The Development of Artificial Intelligence

AI was conceived to augment the human ability to perform everyday tasks more efficiently. Jeff Dean, the leader of Google’s AI division, explains how AI has a variety of methods to absorb, understand and learn information (Dean, 2017). Initially, humans provided computers with specific instructions determined by a set of rules written by a scientist. This first step allowed machines to function with only the information provided by the scientist in charge. Consequently, scientists wanted to teach computers how to learn by themselves, and to accomplish this, humans created a new process of understanding named Machine Learning (Dean, 2017). This new method allowed them to learn how to learn by enabling computers to understand and find patterns independently, using the information given to them to make more accurate predictions. Machine Learning allowed AI to reduce the dependency on scientists as it grew its knowledge pool and started functioning more independently.

Eventually, Deep Learning, a subcategory of Machine Learning, was developed. Jeff Dean, Google’s AI division leader, explains how Deep Learning, similar to Machine Learning, permits AI to build layers of abstraction, constructing a deep neural network through time. Hence, AI can grow its knowledge foundation by comprehending patterns in the information scientists feed (Dean, 2017). However, Machine Learning and its subcategories face limitations. Most importantly, the effect of inputting uncategorised information. A lack of organisation creates a pool of knowledge that lacks specificity, making the machine suffer from a lack of interpretability and reproducibility (Girnyak, 2022). Moreover, the need for a scientist to feed new information can create a lack of data as Deep Learning only permits AI to build a neural network with the data it has, preventing it from expanding content on its own (Girnyak, 2022).

Like Deep Learning, various learning systems started to stem from Machine Learning. Reinforcement Learning gives AI the liberty to decide what to do next and learns based on trying to maximise its cumulative rewards (Sutton and Barto, 2018). So, unlike Deep Learning, Reinforcement Learning permits AI to try new things and know if they are correct or wrong. Still, even though machines could learn through all these methods and, through time, deepen their ability to comprehend and respond to more complex and abstract concepts, having a machine produce creative tasks was never seen as viable.

2.2. The Creativity of Artificial Intelligence

Margaret Boden defines creativity as “the ability to come up with ideas or artefacts that are new, surprising and valuable” (Boden, 2004). Oxford University Professor Marcus du Satoy explores how machines will always be able to produce something new. Nevertheless, to surprise us and create something of value, they need to learn what we, as a species, find surprising and valuable (du Satoy, 2020). Additionally, Carl Rogers defines creativity as “the emergence in action of a novel relational product, growing out of the uniqueness of the individual on the one hand, and the materials, events, people, or circumstances on the other” (Rogers, 1954) which du Satoy summarised as creativity being the tool humans use to explore our emotional worlds (du Satoy, 2020). In both scenarios, we can infer that creativity is defined as something native to human beings. However, Professor du Satoy defies this by questioning if AI can be creative.

As mentioned, AI was developed to enhance human efficiency in everyday tasks. However, as AI developed, a second prompt emerged: can AI be creative? If so, can it enhance human creativity? Professor du Satoy explains how AI can be creative through an example of the Chinese game “Go.” When the world’s best played against AI, they only managed to win one game out of five, and what was highlighted from the matches was the machine making a move categorised as very weak at the beginning of the game. However, the move permitted it to win the match, hence doing something new, surprising and of value, following Bowden’s definition of creativity (du Satoy, 2020). This experience pressed whether machines can be creative by breaking the patterns they learn to create new solutions and ideas.

2.3. The Birth of AIVA

Amid the discussion of whether AI can be creative, in 2016, a start-up in Luxembourg founded AIVA. As founder Pierre Barreau explains, AIVA’s purpose is to “give everyone access to a personalised life soundtrack based on their story and personality” (Barreau, 2018).

AIVA uses Deep Learning and Reinforcement Learning to understand how to create a music composition following specific parameters provided by the user (Barreau, 2018). Since its birth, AIVA has read over 30,000 music compositions ranging from Vivaldi and Mozart to Tchaikovsky and Beethoven. Hence, building a neural network through Deep Learning allows AIVA to find patterns in music, and Reinforcement Learning helps her understand how to compose music as she predicts the following best note or chord (Barreau, 2018). As she reads these 30,000 music compositions, each one is labelled with over 30 categories, including mood, composer style and tempo, helping her to fight the limitations of Machine Learning as she systematically organises her knowledge and provides the most precise output after the user selects the parameters (Barreau 2018).

Since its inception, AIVA can now compose in twenty-two genres, including modern cinematic, jazz and tango. AIVA can also compose based on influences the user uploads as audio files to the interface or by using built presets by AIVA and its users. The parameters the user selects include genre, tempo, key signature, ensemble and length of the piece. Via the pool of knowledge AIVA already has, she now relies on Reinforcement Learning to strengthen her ability to compose music. So, how will such an accessible, mathematically creative and efficient machine affect the demand for media composers?

3. Analysis

The paper’s analysis explores AIVA as a composer’s ally and AIVA as a composer’s competitor to understand the effect AIVA will have on the demand for music composers. Within each argument, the paper explores AIVA’s accessibility, creativity and efficiency as the central ideas.

3.1. AIVA as a Composer’s Ally

AIVA allows more people to become music composers due to its accessibility. As the AIVA team describes, the user interface and pricing system create an accessible and affordable approach to music composition. An example is the Pro Annual membership, which makes composers owners of any music they create with AIVA, giving them the right to monetise it for $33 a month (Aiva.ai, 2016). Technology has always been the primary resource defining how composers approach music for media, and AI, such as AIVA, has simplified and made the job more accessible (Haiming, 2021). Having accessible technology redefines the sophistication of the process as it creates a new tool that facilitates access to music creation, bringing more equal opportunities to those who have not received appropriate training due to a lack of resources or other reasons. Thus, improving the quality of tools at the composer’s reach allows more people to join the career path and express their emotional worlds more easily.

AIVA augments a media composer’s creativity through innovative ideas. As Professor du Satoy explores, French pianist Bernard Lubat could see his style through a new perspective after improvising with AI: The Jazz Continuator, which scientists trained on his way of playing (du Satoy, 2020). As Lubat played alongside The Continuator, he recognised the AI’s playing style as his own, yet he found the AI was creating new patterns and a fresh approach to his style. So, The Continuator helped Lubat free himself of how he was approaching his sound world and discover new sounds within it (du Satoy, 2020). Similarly, AIVA can be fed with audio recordings of music. Hence, a composer can provide the machine with his style and see what AIVA creates. This pushes composers to discover new ways to tackle their sound world and augment their creativity through AIVA’s innovative thinking process. So, AIVA is not only an accessible tool but can also boost a composer’s creativity, making her a great and innovative tool for media composers.

AIVA simplifies the process of developing music composition due to its efficiency. As new technology has emerged, the composition process has evolved. Composers have shifted from a piano and paper to work on platforms such as Logic Pro X and Cubase, providing extensive sounds and instrument libraries. Technology has facilitated the process of creating music and reaching their piece’s purpose, thus making the exploration of sounds and composition of a piece more efficient. As Bernard Marr illustrates, “according to a McKinsey report, 70 per cent of companies will have adopted at least one AI technology by 2030” (Marr, 2019). This emphasises Scott Cohen’s argument: “Every ten years, something kills the music industry. If you want to know what is next, look at the tech world" (Marr, 2019). Cohen sarcastically reflects on how technology kills the music industry every ten years by revolutionising it. Similarly, AIVA allows composers to be more efficient when creating music, as it can generate original ideas based on the parameters provided by the user in seconds. Hence, giving the composer various options to alter them in whatever way seems suitable. Therefore, AIVA can raise a composer’s efficiency by being an accessible tool which augments creativity.

3.2. AIVA as a Composer’s Competitor

AIVA undermines the importance of specialising in a subject due to its accessibility. Music composition has been an art form studied and practised since history can record, and to become a music composer for media has always required specific studying, training and experience. Hence, the development of a tool which undermines the time a composer takes to study music theory, instrumentation, harmony and more is not an ally but a competitor. An example is when The Luxembourg Federation of Authors and Composers issued an open letter to Minister Xavier Bettel with their discontent as AIVA was commissioned to compose for Luxembourg’s National Day. In response, the Minister of Culture said:

“Luxembourg is a country of artistic creation, and at the same time an innovative country and a world reference for new technologies, and it is clear that these two characteristics can perfectly coexist" (Chronicle.lu, 2017)

Therefore, the accessibility to AIVA can create a shortcut in creating music, undermining the education of a trained composer, as it simplifies the sophisticated process of developing original music for media.

AIVA matches the originality of a media composer by relying on a pool of knowledge. As stated previously, AIVA’s collection of knowledge consists of more than 30,000 compositions, meaning AIVA studies thousands more compositions than any human composer. An experiment at the University of Skovde tested whether a selected audience could tell the difference between AI and human-composed music. Their methodology consisted of playing both pieces without revealing one was AI made to eliminate participant’s bias and ask which one resonated the most emotionally with them. After the audience answered, half selected the AI music and the other half the human music. When revealed to them that one was composed by AIVA, they were all sure they had chosen the human-composed as they did not believe AIVA could create music that resonated emotionally with them. However, half of them felt a greater connection with AIVA’s composition. Audience members even described how they “did not think that an AI would be able to create something that feels like a pretty strong human experience” (Lima and Blixt, 2020). So, as the experiment proved, AIVA can match human originality. AIVA can replicate the human emotions believed to belong to only us.

AIVA reduces the time and cost of developing music due to its efficiency. As CORDIS describes, the efficiency of AIVA can allow indie projects or projects with a deadline to create a music track in seconds and musicalise large projects in a week (Europa.eu, 2022). In comparison, when developing a man-made composition, “composers, arrangers, orchestrators, musicians, studio engineers, music supervisors and publishers are hired to create original soundtracks. This can result in a soundtrack taking up to 6 months and costing up to EUR 500 000” (Europa.eu, 2022). Hence, AIVA’s reduced costs and time consumption create a tool which exceeds human efficiency. Likewise, Pierre Barreau elaborates on the idea of AIVA creating “hundreds of hours of personalised music” (Barreau, 2018) to accompany the exploration of a videogame, building a world that constantly feels new for the player in a matter of days. In comparison, a human can create a finite amount of music in a more extended amount of time, proving the efficiency of AIVA once more. AIVA is not only an accessible and original tool but an efficient one, making her the perfect composer for media projects in search of music.

4. Discussion

The paper discusses the three ideas developed in each argument in the analysis to comprehend the effect AIVA will have on the demand for music composers. Consequently, the conclusions are extrapolated to understand AI’s effect on the demand for music composers.

4.1. AIVA’s Predicted Effect on the Demand for Media Composers

AIVA is accessible to non-trained musicians and industry members as much as to media composers. While AIVA can increase the number of opportunities for all people who want to create music for media and have yet to have the chance to educate themselves, it could negatively affect the demand for composers. Its accessibility undermines their training, meaning access to AI could be enough to create music. An educated guess would predict that its accessibility would ultimately create no need for media composers as the creation of a score is ten clicks away in less than two minutes, rather than years of training and then months of work per project.

AIVA can enhance a composer’s creativity by showing them different ways of approaching their sound world. French pianist Lubat exemplified this when AI allowed him to find new paths within his sound world. Her ability to match human originality via her pool of knowledge could make some projects cut the composer out of the process, as AIVA has proven to be capable of imitating human emotion through her music. Her ability to replicate human originality could stir directors and other industry members to use audio from composers they would like their music to sound like and have AIVA produce as much music as they want within minutes.

AIVA can help composers increase their efficiency as she follows a pattern in the media industry where the latest technology, such as Digital Audio Workstations, helps composers find the right sound more quickly. Nevertheless, AIVA could also encourage directors and production companies to cut the composer from the process as it reduces time and costs. As costs rise in the media industry and projects are demanded to be completed in less time, using AIVA as a shortcut in the process and still matching the same level of originality with higher efficiency could displace composers in the industry.

As the comparison between arguments shows, AIVA can match human originality, is accessible to all industry members and has an extensively higher efficiency level than a human composer. The only dependency AIVA has is on the scientists who feed her new music to push her limits. Therefore, the closest prediction we can make based on the arguments previously stated is that composers will likely not have a choice but to adapt to the changes. AIVA will be a powerful ally, but when it is their competition, composers will have to evolve their role in the industry by expanding horizontally. This means learning new hard skills, such as sound engineering, music editing or sound design. Moreover, composers will also have to grow vertically by innovating new business models and responsibilities for their role in the industry.

The effect of these technological changes on demand for composers in the media industry will also depend on the production companies, directors and governments. These authorities will decide between fostering human collaboration by commissioning composers, the nature of the media industry, or using Artificial Intelligence and exchanging a human composer for the cost, accessibility and efficiency of AI. After all, it is not about AI’s capability but how its abilities will be implemented.

Ultimately, the effect AIVA and Artificial Intelligence will have on the demand for music composers will depend on the composer’s ability to evolve their role in the industry horizontally and vertically and the industry’s decisions when implementing it. Soon, AI’s technological revolution will disrupt the media industry and its demand for music composers. So, as new technology always does, industry members and their roles must adapt to the new changes these new tools will bring, hopefully being aware of the importance of human creativity and collaboration.

From Wix Media (2022)

Aiva.ai. (2016). AIVA - The AI composing emotional soundtrack music. [online] AIVA. Available at: https://www.aiva.ai. [Accessed: 13 Nov. 2022]

Barreau, P. (2018). How AI could compose a personalised soundtrack to your life. [online] YouTube. Available at: https://www.youtube.com/watch?v=wYb3Wimn01s [Accessed: 17 Nov. 2022].

Boden, M.A. (2004). The Creative Mind: Myths and Mechanisms. London: Routledge. [Accessed 13 Nov. 2022].

Cambridge Dictionary. (1995). In: Cambridge Advanced Learner’s Dictionary. [online] Cambridge Dictionary. Available at: https://dictionary.cambridge.org/dictionary/english/media [Accessed: 22 Nov. 2022].

Chronicle.lu. (2017). Ministry of Culture Rejects Criticism of Musical Choice. [online] Chronicle. Available at: https://chronicle.lu/category/politics/21802-21802-ministry-of-culture-rejects-criticism-of-musical-choice [Accessed: 23 Nov. 2022].

Dean, J. (2017). How Will Artificial Intelligence Affect Your Life [online] YouTube. Available at: https://www.youtube.com/watch?v=BfDQNrVphLQ [Accessed: 17 Nov. 2022].

du Satoy, M. (2020). Can Machines And Artificial Intelligence Be Creative? [online] YouTube. 28 Feb. Available at: https://www.youtube.com/watch?v=7tHIHCQccjg [Accessed: 17 Nov. 2022].

Europa.eu. (2022). CORDIS | European Commission. [online] Europea. Available at: https://cordis.europa.eu/project/id/876982 [Accessed: 20 Nov. 2022].

Frey, B. (2019). Introduction to Composing with AIVA. [online] YouTube. Available at: https://www.youtube.com/watch?v=SR-UWkSTmAQ [Accessed: 14 Nov. 2022].

Futurism. (n.d.). A New AI Can Write Music as Well as a Human Composer. [online] Futurism. Available at: https://futurism.com/a-new-ai-can-write-music-as-well-as-a-human-composer. [Accessed: 18 Nov. 2022].

Girnyak, M. (2022). What are the Major Limitations of Machine Learning Algorithms? [online] Postindustria. Available at: https://postindustria.com/what-are-the-major-limitations-of-machine-learning-algorithms/ [Accessed 21: Nov. 2022].

Haas, P. (2017). The Real Reason to be Afraid of Artificial Intelligence | Peter Haas | TEDxDirigo. [online] YouTube. Available at: https://www.youtube.com/watch?v=TRzBk_KuIaM. [Accessed: 14 Nov. 2022].

Haiming, L. (2021). The Advantages of Using Computer Programs and Artificial Intelligence in The Creation of Film Music. International Journal of Mechanical Engineering, [online] 6(3), pp.1731–1735. Available at: https://kalaharijournals.com/resources/DEC_225-44.pdf [Accessed: 17 Nov. 2022].

HeadStuff. (2018). Will Artificial Intelligence Penetrate the Music Industry? [online] Head Stuff. Available at: https://headstuff.org/entertainment/music/will-artificial-intelligence-penetrate-the-music-industry/ [Accessed: 16 Nov. 2022].

Lima, A. and Blixt, A. (2020). Investigating the Possibility of Bias Against AI-Computer-Composed Music. [Bachelor Degree Project] pp.1–31. Available at: https://www.diva-portal.org/smash/get/diva2:1572294/FULLTEXT01.pdf [Accessed: 23 Nov. 2022].

Marr, B. (2019). The Amazing Ways Artificial Intelligence Is Transforming The Music Industry. [online] Forbes. Available at: https://www.forbes.com/sites/bernardmarr/2019/07/05/the-amazing-ways-artificial-intelligence-is-transforming-the-music-industry/?sh=6d721d805072 [Accessed: 23 Nov. 2022].

Mason, P. (2018). Paul Mason: Resisting machine control. [online] YouTube. Available at: https://www.youtube.com/watch?v=JFNCwPWzKao&t=778s [Accessed: 16 Nov. 2022].

Moura, P.D.F.T. (2017). AIVA Technology: An Extraordinary AI Music Start-Up. [online] Live Innovation. Available at: https://liveinnovation.org/aiva-technology-extraordinary-ai-music-start-up/. [Accessed: 15 Nov. 2022].

Rogers, C.R. (1954). Toward a Theory of Creativity. ETC: A Review of General Semantics, [online] 11(4), pp.249–260. Available at: www.jstor.org/stable/42581167 [Accessed: 19 Nov. 2022].

Sutton, R.S. and Barto, A. (2018). Reinforcement learning: an introduction. London: The MIT Press.‌ [Accessed: 19 Nov. 2022].