Jennifer Walshe, a renowned composer and professor of composition at the University of Oxford, had been exploring artificial intelligence (AI) long before it became a contemporary fascination within the media and access to song-creating software like Udio and Suno became widespread. She has been operating within this sphere for around a decade.
In 2018, her project Ultrachunk involved improvising with an AI replica of herself. By 2020, she engaged an AI system to conceive a fictional interpretation of 20th-century music in a piece dubbed Late Anthology of Early Music. During that same year, she formulated an AI Enya for a project titled Ireland: A Dataset. More recently, she penned an insightful essay spanning 8,000 words on AI for the Polish Unsound festival, named 13 Ways of Looking at AI.
Walshe argues that the discourse around AI tends to lack depth and adequate research. She believes people fail to perceive the nuances present in this field and suggests that media outlets prefer to simplify complex issues.
AI platforms get their responses from vast data sets, frequently sourced from the internet. Walshe’s early AI experiments were based on data sets she had personally compiled. Her initial ventures included the Text Score Dataset, where text scores, a type of unconventional musical score popularised by composers such as Cornelius Cardew and John Cage during the 1960s, provided instructions to the performers through poetic and ambiguous language instead of traditional musical notation. Throughout this time, Walshe was utilising the precursors of the AI platforms widely recognised today.
Walshe recounts her experience using Microsoft Azure, which allowed her access to the platform as it was formerly known. She had reached out to Open AI too, garnering early access to one of the GPT versions. The data entry task fell upon her assistant, Ragnar Árni Ólafsson, whose cognitive process, she believes, underwent a reconfiguration due to the job. The idea of adding each text instruction spotted along the streets into the data set was something that intrigued the duo. The project, initiated in 2021, she feels would be redundant now as the same information could easily be fed into AI systems like ChatGPT or Claude.
The uncertainty inherent in her music is reflective of the Cagean artistic approach, making AI a suitable medium for her. Walshe talks about how John Cage’s use of contingency was a means to breach his own artistic and stylistic barriers to delve into fresh inspirations. This, she explains, was perceived as a chance to dabble in innovative experiments by herself and other early adopting artists.
Walshe shares her thoughts with regard to accessible platforms that churn out music of varying genres as instructed. The notion of it being the end of music is something people seem to relish, she believes. The advent of Udio was a moment of revelation for her, prompting a humorous foreboding about the industry’s future.
Experimenting with these platforms still keeps her engaged, as seen when she created various renditions of Kurt Schwitter’s sound poem titled Ursonate using Udio.
Referencing her essay, she expresses her amusement over the enjoyable pastiche found in a lot of AI developed work available to the public. She cites the example of a Sinatra AI version where an adept human performer has interpreted Gangsta’s Paradise by Coolio in an ingenious jazz arrangement, imitating Frank Sinatra’s voice.
The voice model of Frank Sinatra was used in a remarkably skilful manner to train and convert human performance, rather than simply typing ‘Frank Sinatra singing Gangster’s Paradise’ for the output. Walshe classifies this effort as a form of ‘fan fiction’. This mirrors the creativity seen when one takes a Kendrick Lamar dish track and reshapes it into a 1970s funk song before posting it on YouTube. The concept is familiar to Generation X thanks to the remix and mash-up culture. The innovative use of technology has always been a tool people use to engage and interact with each other.
Walshe also appreciates a YouTube channel that portrays AI-created songs as historical pieces of work. Consider ‘I Glued My Balls to My Butthole Again’, styled as a 1950s skiffle or ‘It’s Time You Took a Shit on the Company’s Dime’, an 1980s retro hit, both created with ease using Suno or Udio. The concept and follow-through altogether might take mere 20 minutes. Despite this, some argue that the true art lies in the birth of the idea.
There are fans of AI music who view the craftmanship and skill that come with creating music as inherently elitist. They argue that there has always been a structural inequality preventing people from creating music. According to this view, platforms like these would eventually allow everyone to create their own pieces, such as ‘I Glued My Balls to My Butthole Again’, thereby serving as a real testament of human creativity.
Walshe remains sceptical. Despite coming from impoverished backgrounds, people have been crafting music for many centuries, as she points out. Music might require effort, but it’s a process that brings happiness. “Should you wish to stimulate everyone’s creative spark, you could spend the [billions in] capital funding on purchasing musical instruments for kids. My grandparents were workers and they owned a piano, a highly complex musical apparatus. Back then, the notion that music was out of reach didn’t exist.” If an individual is “able to articulate a song they’d like produced, I don’t believe that having said song delivered constitutes ‘releasing their creativity’.”
She further emphasises that the mundane “cues” users must utilise to command AI systems to produce music deviate significantly from traditional ways musicians or listeners have considered, discussed, sensed and written about the art form. “It’s entirely reductive. You’re not expressing your feelings of listening to the song; you’re not explaining what the music did to you physically or its cultural implications.”
Additionally, the AI-oriented interpretation of creativeness hinges on the belief that artists start off with a predetermined end-goal. This contrasts greatly with the typical workflow of artists. Walshe appreciates it when the system she employs malfunctions and produces surprises. She suggests that as these AI technologies “advance”, thus becoming more predictable, their creative potential becomes less fascinating. “I’m a spontaneous performer. I’m accustomed to freestyling for five minutes before something starts to take shape, and then dissecting those aspects. My interests significantly differ from a youngster looking to create awesome techno music.”
She worked with an AI replica of herself that was meticulously developed in her Ultrachunk project, a joint venture with artist Memo Akten. The public also had the chance to engage with her digital doppelganger as part of an exhibition, which resulted in a bizarre experience for her. “It was rather peculiar observing strangers manipulate you,” she states. “And it’s you without clean hair, or backstage just before a performance… and you have spare teeth and additional eyes and such.”
Walshe is acutely conscious of the potential abuse and misuse of such technology. She expresses concern about young individuals who, as they grow, are creating a digital footprint of themselves and become susceptible to bullying through fabricated content by their peers. She noted instances of indecent imagery abuse in academic settings have already transpired.
Our self-image and voice are personally integral to us and are seen as part of our identity. Walshe mentions an instance when renowned singer, Tom Waits, sued Frito-Lay, a snack manufacturing company, for utilizing a voice very similar to his in a commercial. In a recent case, actress Scarlett Johansson openly objected to the usage of what seemed to be her voice by Open AI for their new digital assistant, ChatGPT. Sam Altman, the CEO of the organization appeared to allude to ‘Her’, a film where Johansson gave her voice to an AI digital assistant, similar to Alexa by tweeting the film’s name. The fact that Altman would hint at a dystopian film about the dangers of AI to promote an AI product appeared to demonstrate a lack of understanding on his part.
Walshe’s concerns extend to instances when this technology is being used to voice-record Frank Sinatra singing songs from the hip-hop genre, which she describes as an interaction of different cultures and signifies that as being amazing. However, when a voice resembling Joe Biden’s is used to misinform and deter senior citizens from voting, Walshe finds it deeply alarming.
The capacity of these emerging systems to generate content with ease also inadvertently leads to an overflow of average and bizarre content on the internet. Walshe refers to this as “gunk.” She says the internet is already overwhelmed with such content. She highlights a detailed article published by 404 Media regarding what they term as the “zombie internet.” The article investigated numerous peculiar Facebook pages showcasing AI-generated images of unique and niche topics such as Jesus or children from developing countries constructing cars from water bottles.
Much of the distasteful online material is sexually explicit, according to Walshe. She mentions her colleague, John Lydecker, who believes that all data is subject to sexualisation. This is visibly demonstrated by an online funding campaign by technology firm Unstable Diffusion, which aimed to generate images incorporating 30% nudity, 30% adult material, and 30% anime. Walshe points out the unrealistic and bizarre representations of the female body within these images. Moreover, she mentions another AI software where users produce offensive and degrading sexual content involving women and Nazi anime, which raises concern about the impact on young minds that are exposed to such content.
Additionally, these inappropriate materials further contaminate the already problematic public data sets that AI systems utilise to generate their outputs. Walshe praises the valuable work of Trevor Paglen and Kate Crawford, who through their ImageNet Roulette project, revealed the deep-seated racism within the image database that forms the basic structure of many AI networks. Following their exposé, the database was removed for necessary improvements. Often, it is artists who are making significant political statements, questioning the issues and engaging with them.
However, Walshe believes that rather than blaming the technology, it is the humans using, regulating, and profiting from AI who should be scrutinised. She insists that decision-making at each juncture is a human affair.
Walshe also envisions certain domains where AI might prosper, notably within the commercial music industry. She relates friends’ experiences of being asked to create music that mimics an admired track, and perceives that AI could be utilised for such purposes. In this context, the AI generated music could serve as background noise or “muzak”, which doesn’t demand our full attention and simply forms part of the everyday sonic landscape.
Walshe firmly holds the idea that human’s intrinsic love for creating music, either with their voices or traditional instruments, won’t ever diminish. This passion stems from the challenges that make music-making rewarding and enjoyable. She mentions working with an enthusiast choir known as Musarc in London, whose members revel in the experience of making music together in a shared space.
The concept of ‘musicking’, as conveyed by musicologist and musician Christopher Small, refers to the social act of engaging in music. Walshe assures that there’s absolutely no risk of this aspect disappearing or people losing interest in composing original music. She equates this to the culinary world where despite ready-to-eat options, many still find delight in preparing meals from scratch.
As for AI’s role in the music industry, Walshe admits being uncertain of its definitive purpose. She even questions whether those developing these platforms have a clear vision. According to her, it seems like they’re handing it over to the public and simply observing the outcomes.
She acknowledges the potential for unforeseeable uses of this technology. Notably, in her AI workshops, participants are often unable to find services worth a €10 monthly subscription among existing AI tools, suggesting that what they actually need is not currently available.
In Walshe’s perspective, there’s a clear lack of necessary dialogue with artists. She criticizes the current trend where solutions are being created for non-existent problems.