Dear Algorithm: Letters on Music and Machines
Artificial Intelligence (AI) is one of the technologies whose recent boom not only practitioners of classical music observe with a keen eye. The rapid advancement of AI and GAI technologies in the last years has accumulated much attention and criticism from various fields and professions - but as a matter of fact, AI has been around for longer, hidden and largely invisible in infrastructures that we frequently encounter in daily life, be it in using grammar and spelling checks to write emails, the use of navigational systems to determine the most efficient travel routes, or in encountering personalised advertising on our favourite social media. As for any emerging technology, however, the effects of AI on society are still largely unclear. This technology also poses new questions for various artistic practices, including classical music.
What is AI, and how does it work?
AI is a technology that performs specific tasks that attempt to mimic human intelligence, such as reasoning, problem-solving, decision-making, or language understanding (Simons & Wils, 2024). Based on algorithms that are designed through statistical methods, much of the recent focus of AI development has revolved around machine-learning algorithms (Laidlow, 2024). Machine-learning algorithms, as composer-researcher Robert Laidlow (2024) explains, learn based on a given data-set and encode specific rules into a digital model, which is then used to carry out the task in question. Generative AI, or GAI, more specifically, is able to produce supposedly ‘new’ output with help of the digital model and based on the variety of material that this model has been exposed to or pre-trained with (Simons & Wils, 2024).
What you can find here are letters that the three main authors of the website exchanged via email over a few weeks. These letters are personal reflections on the roles, possibilities, and boundaries of Artificial Intelligence in our society and musical practice. And this exchange is far from concluded: we invite you to respond with your own letter. Send it to mcicm-fasos@maastrichtuniversity.nl and we will publish it on this page.
Maastricht, June 6th, 2025
Dear Denise, dear Jorge,
I hope you are both doing well. It is a pleasure to work with you on our research project on classical music and digitality that we hope to present through a website. Today, there are no trains due to a strike of our national railway company NS and I am sort of stuck in my apartment in Maastricht, ironically looking out over empty railways behind the building where I am. This letter is meant as the start of a correspondence between us on the topic of artificial intelligence at its many relations to what we call classical music. This term has become increasingly problematic recently for all sorts of reasons, as you know, and I suggest we go with the rather broad definition of music that uses notations to be performed, borrowing the approach of Matthew Aucoin in The Atlantic. We share an interest in how this music exists, and consequently how AI will affect this existence.
To be honest, the topic of artificial intelligence was not at the top of the list of things that fascinate me. Even though I am aware of the huge implications machine learning will have in our lives, the hyped discourse on the threats and promises of this technology remind me of the reception of the personal computer in the late 1980s. At the time, I worked as a freelance journalist and with a friend we decided to write about what the PC would do to the writing of novels and poetry. We interviewed writers and asked them if they used a computer and if so, if it made a difference. One of them, Harry Mulisch, told us that he refused to write on a computer. He used a fountain pen because its ink would flow onto the paper "as warm as my blood". The obvious romantic overtones in some of the responses led us to reflect on whether this focus on the difference between handwriting, and computer writing made any sense. Was it not the case that the PC would also bring new opportunities that would redefine what writing actually is?
I had to think of this article when I watched a short YouTube video called Bach vs. AI: Spot the difference. It starts with a challenge: can you tell the difference between two versions of a harmonized Lutheran chorale melody, one of them by AI, the other by Johann Sebastian Bach. I liked the challenge. To me, the first harmonisation sounded odd and at some points completely out of place, whereas the second felt immediately familiar. But I was not sure. Postponing the correct answer, the presenter explained why telling the difference is so important: it is all about human creativity! If a machine can make music as good as Bach, what does that do to our self-understanding as creative beings endowed with the imagination to bring the new into the world? Should we think about AI as a threat or an opportunity? The presenter, Professor Marcus du Sautoy, a mathematician who has a chair in Public Understanding of Science at Oxford University, gives an interesting answer. As humans we tend get stuck into our past experiences and we start behaving a bit like machines, reproducing what is already there. AI kicks us out of our comfort zone, sort of.
I find the argument interesting, because it reverses a line of reasoning that highlights how human creativity is fundamentally different from machine learning, that is founded on mimicking existing examples. In the case of AI generated music that mimics Bach, the result is underwhelming to say the least. But if this observation is almost trivial, how do we open up more interesting perspectives? How can we mobilize our backgrounds in science and technology studies, art theory and art ethnography - and in your case as a professional musician, Jorge - to ask other questions than whether we can tell the difference between human and machine, fake and real, or a past rich in meaning and a present of endless repetition?
I am looking forward to reading your thoughts!
All the best,
Peter
Cologne, June 13th, 2025
Dear Peter, dear Jorge,
Thank you, Peter, for your thoughtful letter. I share many of the questions that you raise, yet is impossible me to find definitive answers. When I think about generative artificial intelligence, I – just like you – keep gravitating around the idea that the models draw from things gone by. Sometimes, I need to remind myself that they neither understand, nor think for themselves; the machine (as if there was ever a singular one, ha!) is a sophisticated prediction model, working on probabilities and calculations. In the case of music – a deeply creative and human endeavour – this seems like a strange contradiction. In addition, I wonder… What past are we actually talking about?
Scholars like Kate Crawford or Meredith Broussard have drawn attention to the fact that the data used to build and run these models is fundamentally flawed – just like us humans. They promise wider access and fairer distribution of knowledge but are biased and unreliable, reinforcing and perpetuating existing inequalities. Think, for example, about its production costs, running often on precarious labour, normalising theft and unauthorised data scraping, all of which is devastating to our environment. Who profits from AI? And how does art relate to these issues?
For example, for some time now there has been a huge controversy over the AI-generated images that copy the aesthetics and style of legendary Japanese filmmaker Hayao Miyazaki and his Studio Ghibli. I remember watching Spirited Away (2001) in theatres, when it was just released. I was a kid, barely 11 years old; I had seen plenty of anime on TV, yet those series were not as well and lovingly produced as this piece of art. I remember sitting in my seat mesmerised, the beautiful colours raining down on me like soft rain on a Sunday afternoon, formed by animations that I today can recognise as hand-crafted. Everything about this film spelled ‘love’ and ‘dedication’ to me. To this day, it’s one of my favourites. This is the beautiful thing for me in art: the dedication with which we commit ourselves to giving form to an idea, a feeling, an experience – and making this experienceable for others, no matter what they may take away from it. At its heart, all art is a fundamentally communicative endeavour.
(Ironically, this is also the perverse reasoning for AI developers to shamelessly take it for their models: art is meant to be shared among people.)
At the moment, it seems that at least some of the energy with which artists pursued this dedication is drained by worries about copyright and authorship infringements, by the desire to protect what is precious to them: their art but also, connected to that, their way of life. But let’s not forget that AI is not asking entirely new questions: we faced similar challenges with the advent of printmaking as well as photography. (Shoutout to my students never failing to grapple intensely with Walter Benjamin’s 1935 essay The Work of Art in the Age of Mechanical Reproduction.) We turned out okay.
We are not unequipped to face these problems – at least not as unequipped as it sometimes seems. In the context of music, questions of authorship or copyright might feel less urgent – particularly when looking at the example of Bach you raised – but they are still pertinent. Whose music is fed into these models, and how? We cannot know what Bach would have thought about his art being used to train these models, but in any case, he certainly would not be the only person to ask. What about classical composers who have always been on the margins of history, such as non-white, queer, or female artists? Do they need to be protected from being fed into the machine, or do we need them desperately in the machine to make it more equal?
In a YouTube video by the cult duo TwoSet Violin, Eddy and Brett listen to an AI-created classical music piece. While the discussion is quite technical, at the end of the video the two make the following, poignant observation:
Brett: That’s been my observation in the current set of AI across the board; it’s impressive that it’s able to do things quite well mediocrely but a) it hasn’t reached the top level and b) more importantly you have to remember that it’s always learning from the past. So I don’t know if there will be a time when AI can learn to create something genuine in the future but until then what you heard is a Mozart copy or you heard a Chopin copy but you’re not going to get the next thing.
Eddy: To be honest, I am just kind of sick of hearing AI stuff. Yeah, I’m like dude, you know what AI can’t do? REPLACE YOUR PRACTICE! Right?! AI can’t practice for you. Go practice!
Taking Eddy seriously, here’s what I think we can do: continue to make music ourselves, grappling with our own creative ideas, and, most importantly… Breaking rules. Because that AI is not (yet) able to do given its dependence on the past; those we will still need to break ourselves. And classical music might have never been before in such a dire need of rule breaking to escape its oftentimes rigid tradition, so it cannot merely remain meaningful to society, but also gain new meanings.
Where there are many ‘remakes’, a friend recently said to me in the context of film, the work of art will not be threatened but become even more important. I can only keep hoping it is so.
Warm wishes and looking forward to reading more from you two,
Denise
Bogota, July 4th, 2025
Dear Denise, dear Peter,
Thank you both for your thought-provoking letters! I just re-read them once more while on a plane over the ocean, which is an unlikely place to be thought-provoked about the relationship(s) between digital technologies and Western classical music. But perhaps an unlikely match calls for an unlikely place to think and write about it.
That digital technologies and classical music are an odd combination would have sounded somewhat obvious to me before working on our project: after all, as a viola player, I could in theory carry out my entire career as a performer in an analogue way, that is, without ever making use of any digital technology. These instruments are generally made by hand (and more valued for it); music scores exist first on paper and are sometimes only physically available; learning an instrument requires an in-person setting that often consists of imitation, repetition, and immediate feedback from a tutor that has not needed technological aids to survive for centuries; a performance typically goes without electronic amplification, and so on.
Yet, the strangeness of the mix with the digital is an idea that I keep coming back to when I read your letters or when I was in reflection mode in any of our project’s activities. If they are indeed such an odd pairing, why do we engage with it, not just the three of us but all the researchers, practitioners, educators and administrators that have done so in recent years? This might sound like a silly question, but I think how we answer it might define how we approach and understand that interaction. I see at least two options: on one side, a kind of we’ve got to catch up, that is, a reaction to so many other fields that are making or have made the jump into the digital, often following market concerns. On the other, I see those who are actually curious to discover and understand what that interaction might actually entail, and not (just) how we can profit from it, that is, how we can incorporate it into current classical music business models.
The trend of catching up with other fields engaged in digital development is to me especially visible in the hype around the topic of AI. Approaches like that of Marcus du Sautoy, as shared by Peter, seem to fall on the second of the groups I proposed above: instead of assuming it is an opportunity, his reflection takes a step back and asks what AI’s implications are for human self-understanding and creativity. Another necessary question in this line could be, what are the implications for the practice and nature of classical music itself?
Here I want to reflect briefly on what seems like a trend in conversations about AI in classical music: the focus (and often problematisation) of the fact that AI draws from or even mimics the past. Beyond the very real problem of the rights of authors whose work is used to train AI models, it seems to me that the charges against AI for depending on the past to generate music ignore the fact that classical music composition does intrinsically the same. Indeed, isn’t that classical music’s current paradigm at its core? Isn’t classical music already in a present of endless repetition? Here I kept thinking of a sentence in Virginia Woolf’s A Room of One’s Own (1929): "Masterpieces are not single, solitary efforts, they are the outcome of years of thinking in common so that the experience of the mass is behind the single voice."
Denise’s example of the YouTube video by TwoSet Violin is especially telling in this light: Eddy and Brett seem to make the (to me) astonishing assumption that that a living great composer would not be drawing from the past in order to make genuine stuff. They don't seem to see a difference between learning from the past and copying it. It's as if they thought the first baroque composer, if there could be such a thing as "first", woke up one day, decided to do away with all the music made until then (that is, from the past) and began to write brand "new" music. Or that a stylistic period like Classicism ended suddenly one night and the next morning the world woke up under Romanticism.
This doesn't speak only of Eddy and Brett, but of the system that educated them. And here is where I believe that we can mobilise our backgrounds and go beyond a “telling the difference”-approach: engaging with AI, for instance in composition or co-creation, as an exercise of rendering visible through disruption – and the uncanny – the tropes that stagnate the practice of classical music.
I look forward to reading your thoughts and continue this collective reflection!
Best wishes,
Jorge
Maastricht, July 16th, 2025
Dear Jorge, dear Peter,
I could go on with these letters forever, but the summer holiday is almost starting for me, so forgive me if my thoughts aren’t as sharp anymore as they could be. I am sitting in my office. The faculty is deserted – as it usually is this time of year. The students have gone, most colleagues too. I like June and July here (although I’m glad that the desertedness does not last too long): the emptiness of the usually busy and buzzing building is so strange, yet so captivating. As if we’re in an apocalyptic scenario where everyone has vanished but me. At least, the coffee machine is still working.
Thanks, Jorge, for sharing your thoughts. I don’t think that classical music and digital technologies are an odd pairing; in that sense, I hope our research line has fallen into the second category you proposed, the intrinsic curiosity about what actually happens when they interact. Your proposition that AI renders visible those tropes of classical music that help it stagnate is fascinating to me in that regard; what I understand you mean is that it clashes with the aesthetics that are so intrinsic to classical music, such as the autonomy of the composer (which is – historically speaking – reserved for the white, male genius); the authority of musical text and literacy (which is usually learned through years of education and training, not only for professional musicians but also for amateurs); or fidelity (which is of upmost importance in the reproduction and performance of musical works). All of these become exposed and unsteady when faced with AI. If I’d have to look at it normatively, with an eye to the future of this artform, I’d say that that’s a good thing: the technology, in a way, unwillingly asks us revisit the things we have taken for granted, and allows us to critically ask if we want to keep them (and if so, how).
In connection to this, I kept thinking about something that Peter said in our podcast conversation: that while it lends itself well to conservative ideals and aesthetics, classical music is potentially one of the most radically innovative artforms. This made me think of Igor Stravinsky’s premiere of The Rite of Spring in 1913, a work that is said to have deeply shaken what classical music was and could be (at that point of time). I don’t know much about the piece, admittedly – I listened a few times given the countless number of recordings that exist of that piece – but I know that it has been framed as groundbreakingly innovative. Reviews on this radical innovativeness, however, gained traction only years later, as far as I know – before, the piece was infamous for apparently causing audience riots at the Paris premiere.
This could be an intriguing factor about innovations: maybe you don’t always notice when you’re in the midst of them, like sitting in the eye of the tornado. Maybe you first think it’s shit, until you recognise its value. STS scholar Lee Vinsel, in his book The Innovation Delusion, argued: “Unlike actual innovation, which is tangible, measurable, and much less common, innovation-speak is a sales pitch about a future that doesn’t yet exist” (2020, p. 11). AI is much innovation-speak, it seems to me. But could it be a spark helping to light the fire for new artistic ventures in classical music? And when is classical music truly innovative? What is the role of technology in these artistic innovations today?
What is or will be our time’s Rite of Spring?
(Or is the music already playing, and we’re not listening close enough?)
With the best wishes,
Denise
Groningen, July 23rd, 2025
Dear Jorge, dear Denise,
It has been a pleasure to read your letters. I am amazed how many important topics we touched upon while discussing classical music and how its existence is shaped by AI, and more broadly digital technologies. As Denise said in her first letter, the question of creativity that is raised in many of the debates on AI and music or even art is intimately linked to how we draw on the past in music, and what past we are talking about. Or to put it differently, whose past. Do we include the past of those who created on the margins of history?
In your letter, Jorge, you elaborated on the question of how AI reiterates or mimics the past. As you rightly observed, Two Set Violin did – curiously I have to say – not see the difference between learning from the past and copying it. I agree that AI, or any digital technology, can disrupt the tradition of classical music in ways that render visible the stagnation in the practice. As Denise responded to this point, it might well be that we do not see the innovative potential of digital technologies that is already actualized as we speak.
To me, opposing tradition as stagnated practice to innovation that we may not even recognize is helpful, and yet I feel that for me some of the magic of working with music, or art, comes from our experiences in the present. To hopefully explain what I mean, let me share a story with you.
As you know, last year we moved from Maastricht to a small village in the North of the province of Groningen. It has sixty inhabitants and by now we have met most of them, and some have become new friends. One of was a neighbour of 88 years old. He lived down the street and the first time we met him on the street in front of his house, we invited him for coffee. A few days later, the doorbell rang and there was our neighbour. He brought some books he thought would be interesting to us, one of them by George Santanyana. Not only did he share his thoughts, he also played his music for us. Although he never learned to read music, his jazz improvisations were incredibly rhythmic and creative. He also owned a piano and a harpsichord, a small instrument with one manual.
A few weeks ago, we met his wife who told us that our neighbour friend was in the hospital. As it turned out, his lung cancer was untreatable. We visited him twice after he received the bad news. The second time, he told us how happy he looked back on his life and pointed at colourful feathers of the pheasant that inhabits their back garden. He also said that he wanted to give me his harpsichord. Two days later, his son and two of his grandsons who live down the street came to bring the small, beautiful instrument. As it happened, our neighbour died on my birthday. When we went to see him a few days later, a pheasant feather lay next to his body.
The harpsichord now stands near our baby grand piano which looks huge and black next to the wooden instrument with its light lime wood sounding board and black keys. Pressing these for the first time immediately made it clear that the instrument was badly out of tune. It was in fact unplayable. I called our piano tuner who said that harpsichords are usually tuned by their owner as this needs to be done quite regularly. We made an appointment, and he would teach me how to tune the harpsichord. Slightly impatient, I started to search for YouTube videos about how to tune a harpsichord. The very first one I found was made by Alice Chauqui Baldwin and it showed how to tune the Vallotti temperament by ear. She mentioned the app ClearTune as her favorite and I downloaded it.
The next thing I needed was a tuning hammer, a device to turn the tuning pins. I quickly found one online, but then I discovered that there was one lying in the harpsichord, which makes sense given how often you need one. Now I had everything to start tuning the harpsichord. As it turned out, ClearTune allows you to tune a variety of historical temperaments and using one of these is part of the fun of playing a harpsichord, because it allows you to play baroque music in the keys that were used at the time. I decided to tune the harpsichord in the Vallotti temperament. This temperament has some very pure intervals and chords, but some are less good. It sounds quite different from the modern equal temperament that has become ubiquitous since the nineteenth century. It took me some time to find out which tuning pin connected to what string and that a single key sounds two strings, tuned an octave apart.
Troubled at first by the little music stand that was in the way, I learned that I could easily remove it and I now had full access to the pins. Using the tuning hammer, I raised or lowered the tone, and the app showed if I moved in the right direction. When a small yellow triangle turns green, you are in the right spot. It was interesting to see how I more and more was able to determine the sweet spot by ear; the app trained me what to listen for. I realised that musicians always had to use their ears, and that harpsichord players were exceptionally well tuned to notice the minute differences in pitch that separate noise from music.
When I finished the tuning, I was amazed by how special the instrument sounded. As predicted, some chords were more pure than I know from the piano, others had a strange tension without sounding out tune. I started to play a Courante by Handel and noticed how little movement the keys need and how precise the timing needs to be. The next day, the Vallotti temperament had deteriorated because of the minute movement of the wooden components and sounding board over which the strings are stretched. In a few days, however, tuning the harpsichord became a routine. I got to know it better; it is a Wittmayer, made by a German harpsichord builder who sold thousands of these instruments back in the 1960s and 1970s, when early music became en vogue and people wanted to play Bach on period instruments in their homes.
Forgive me the long story, but I wrote it down because playing on the well-tempered harpsichord was indeed a magical experience. In this experience several things come together: my eagerness to learn about the sound of early music, but also about the instruments we use to play this music. The vast amount of information and teaching material that is available online, thanks to musicians who share their knowledge and expertise. The ClearTune app that helps me to tune and that trains my ears. And of course the harpsichord itself that, like all musical instruments, responds with great precision to my movements, inviting me to adjust my touch and sharpen my sense of hearing. Compared to the piano, its sound is feeble. One has to listen very attentively to hear the richness of sound and resonance. Some strings are missing and I will need to find someone who can repair small defects.
Playing my small harpsichord, it becomes clear to me how innovation, or even progress, in music has often meant louder sounds produced by bigger instruments in larger spaces. The baby grand standing next to the harpsichord visualizes this trend. Am I going back in time when I with keyboards? Is my experience that of connecting to a past that has vanished and is brough to live again? Or is my enthusiasm entirely in the present, springing from the joy of learning new things and skills, and being able to notice differences I did not know existed?
I think that these are precisely the kind of questions my neighbour friend would have loved to discuss with me.
Dear Denise, dear Jorge, many thanks for all your thoughts and ideas, and I hope enjoy the rest of your holidays!
Best wishes,
Peter