I have heard, or even listened, like many of you, to some "products" of Artificial Intelligence in terms of music, which have been highly publicized lately.
I mean of course these covers, by the voice of Michael Jackson, of The Weeknd, not necessarily realistic in my opinion, or, more bluffing, of the one of "Tu ne sauras jamais" of the BB, or of the cover of "Thriller" by the deceased Freddie Mercury...
There is in these technological manipulations a lot of curiosity, and it is normal, since we are only at the beginning. We like to be surprised, we like to dream of extraordinary opportunities or, on the contrary, we like to be scared. As for the media, it seems to me that they are more interested in creating a buzz than in teaching.
Some of these covers are truly artificial: given the number of filters applied to The Weeknd's rather monolithic voice, it is difficult to find the "soul" of Michael Jackson's voice, cramped in this imposed framework. Others are a little more convincing, although the typically Jacksonian vocal effects of breathing and vocal inflections applied to the "mold" of Mercury's voice on "Thriller" give the impression - like the monster in the mayor's belly of "Ghosts" - that one has not quite been digested by the other and that he is trying at all costs to extract himself from there. Finally, others are interesting, almost moving, all things considered (I'm thinking of "You'll Never Know").
But before going further on our perception and our feeling, let's first go back a bit and ask ourselves...
Is this phenomenon of imitation / "copying in the style of..." by AI new?
In reality, no. Many musicians, composers, theorists have worked to rationalize music. And for good reason? Wasn't it taught in the Quadrivium in Antiquity and up to the scholastic teaching of the Middle Ages, alongside arithmetic, geometry and astronomy?
The mathematical and physical aspect of music, with its frequencies, its harmonics, its Pythagorean system is no longer to be demonstrated. Is it not said that music is "the number made audible"? It lends itself so well to logical reasoning seen as sequences that the French composer of the baroque period Jean-Philippe Rameau (1683-1764) had imagined the beginning of a model with his " Traité de l'harmonie " (Treatise on harmony) reduced to its natural principles. One could also speak of the mathematical techniques very evident in the work of Johann Sebastian Bach, his contemporary, which already marked the previous eras but were quite systematic with him (counterpoint, canon, fugue, as many sound expressions of equation, symmetry, balance, geometrical deployment etc.)
Therefore, in the 20th century, music was also one of the first fields of application of computer science and it is considered that the "Illiac Suite", in 1956, was the first piece created with an AI: it was then a question of generating a musical work inspired by the style of Jean-Sebastien Bach.
How is this possible? How can AI compose, create new sound works?
Well, this famous "Illiac Suite", for example, used Markov chain techniques: these are statistical techniques used to model temporal processes, or, to put it more simply, to find a model capable of predicting the next event based on past events. All this is possible if we consider music as a sequence of events, composed of notes or chords. We can then predict, make statistics, reproduce loops, patterns, sequences of harmonic grids, lines of notes, principles of arrangements.
Nowadays, AI relies on artificial neural networks* to make music: it is an organized set of interconnected neurons, inspired by the human brain, allowing the resolution of complex problems such as computer vision or natural language processing. Other principles, such as machine learning, are combined. The machine is fed with examples to be inspired or imitated, so that it "digests" them, analyzes them, and is capable of regurgitating them recombined in multiple ways.
These recent neural network technologies have evolved a lot in the last few years, to the point of being very efficient in stylistic imitation or realistic synthesis of the sung voice.
On the function of art and our relationship to the work and the artist
Beyond these pirouettes and experiments, can it really work? What is the interest for the artist and for humanity?
To wonder if, tomorrow, the voice produced by the AI will be able to replace that of the human being, of the singer of flesh and bone, comes back, in fact, to ask the question of the function of art, and of our relation, as human, with it, whatever its mode of expression (music, painting, sculpture, literature...)
Because, to come back to the one we are interested in, are we really looking to hear something new (I think of the hunters of unpublished works) to the point of being seduced by unpublished works and covers that have never been interpreted by Michael Jackson? What are we interested in? Only the voice, a voice that sounds like him, even if it is only the product of powerful logarithms? A voice that is still present, but new, a kind of band-aid on the wooden leg of death that would flatter, in the confines of our reptilian brain, our dream of immortality?
No, I don't think so. At least, not yet. Not our generation.
Beyond a vocal frequency, a voice is an expression. The expression of feelings. The product of a body of flesh, blood and bones. The fruit of a soul, of a life, of an experience, of successes or setbacks, of convictions, of fights. The extension of a body, a spirit, a breath and a respiration, alive and random, capable of failing, like the human condition, at any moment. A shared moment of presence, a parenthesis, a bubble of life.
Walter Benjamin, in his work The Work of Art at the time of its technical reproducibility, published in 1936, already worried about the value of the art since this one had become reproducible and, therefore, for the recordable and diffusable music, since this one was dissociable of its transmitter (the artist, the singer, the musician) and this, millions of times (thanks to supports such as the disk and its devices of diffusion).
The philosopher was aware of the fact that "the work of art has always been fundamentally reproducible" and that what one man had done could always be reproduced by another. But the revolution that particularly marked his time was the "technical reproduction" of the work, on a massive and industrial scale, that which no longer involves the hand of man or any other living organ of his expression.
Undoubtedly we are, in 2023, at the threshold of a comparable revolution with the intrusion of the AI in the production or the "creation" of art and music.
For Benjamin, the work of art possessed a unique character because the best of the reproductions would never have known "the here-and-now of the work of art", i.e. "the absolutely unique character of its existence". Question of conception? Because, it is necessary to note that beyond its apprehension with regard to these new technical forms of creation, the human spirit has since evolved well and that we are henceforth inclined to consider, since many decades, that the recording of a voice gives us in spite of everything access to this "present moment", to this "here-and-now" certainly past, but which saw and heard being born, in the microphone, in the intimacy of the studio, such a melodic line at the time of its recording.
Michael Jackson's breath, often clearly and deliberately distinguishable in Bruce Swedien's microphone and in the mix, connects us a lot to this moment and this presence, making them tangible and giving us the feeling of seeing this studio space where things happen and, perhaps, of being hidden in a corner. This has undoubtedly helped to compensate for the loss of aura that Walter Benjamin feared: "what withers in the age of the technical reproducibility of the work of art is its aura" - that aura that he defined as "the unique appearance of a distance", what one feels, for example, when one looks out over "a calm summer afternoon, a mountain range on the horizon". For him, this decline of the aura was explained by the fact that the masses demanded "to make things closer to oneself spatially and humanly". This is what happened with the making of "stars" and this concern to draw the contours of a vocal personality that is well identifiable, tangible, close, corporalized, thanks to recording strategies aiming to restore even the breath, the sound of the mouth, the velvety pinch of the lips. In short, a presence, a proximity, even a virtual inimity, but in which we believe a little and which finishes weaving our relation and our links with the artist.
What would Walter Benjamin say about the relay taken from now on by Artificial Intelligence, able not only to reproduce, but even more to "manufacture" (I voluntarily do not use the verb "to create") songs or unpublished musics, apeing with talent the style of such or such composer or author, or vocal lines which never came out of any living body?
Without talking about the legal and financial issue of copyrights, which I will come back to very quickly, it seems to me that it is the ethical and societal issue that should worry us.
Are we interested in hearing a voice like Michael Jackson's, which is not even produced by an imitator, but produced by a machine without soul or conscience? What do we listen to when we listen to Michael Jackson (or any other artist who lives or has lived)? More than a voice, we listen to his soul, his emotion, his heart, his ability to convince us (whether it is true or not) that what he is saying there, what he is sharing there, is the most important thing in the world for him at that moment and that he has lived it. We feel understood, accompanied in our lives and our trials, we are not alone, we identify and create an attachment with him, even if we will never really know him and he will know us even less.
But can we let ourselves be fooled by a machine that we know for sure, despite its prowess, has neither soul nor conscience, that it has not lived anything and does not even have the will to share lies and artificiality with beings that it cannot even "know" exist?
Should we be afraid of the intrusion of AI in music, or even in society?
Nowadays, specialists such as François Pachet** do not seem to be very worried about this and think that AI is limited because, if it remains (to this day) very efficient in terms of production (generation and combination of sounds), it is not (yet) for composition or melodic creation. This is what explains its greater efficiency in rap (where harmony and melody are relegated to the background) in favor of the scansion of a text by a voice (rarely harmonized by choirs). The model to be imitated is less about creativity and originality than about a technical scheme that can be easily copied by the machine. But what will it be tomorrow?
One can also oppose the fact that digital synthesizers operated a rather similar revolution in the 1980s, closely followed by sampling: who is moved today to hear a sampled violin sound rather than real strings vibrating in a centuries-old maple body? Rare educated and awake ears? A few retrograde purists?
I note that, generally speaking, when some specialists are asked about the legitimacy of being worried about these new features, most of them immediately think "copyright". If this aspect seems important, it does not seem to me to be central. Certainly, money is the key and there is an inadequacy of the principles of traditional copyright to AI techniques. The directive proposed by the European Parliament was not enough and the legislation will necessarily have to evolve. Indeed, the copyright is very complex in music and is declined at several levels: right of edition, composition, recording but also moral right related to the use of the name of an artist. All this will eventually be decided by discussions between experts.
If AI allows artists to "create" something new or to carry out projects that they would not have been able to do alone, why not. We can't prevent humans from using and evolving technology. But how far should we go? For what purpose? Can we consider that ersatz pop music like the Beatles, or rock'n'roll like Elvis, have the same interest, the same message, the same impact, the same value in terms of cultural heritage, whether they have been produced by an AI or by human brains of musicians?
For me (it must be my conservative or "retrograde" side), I think of the movie "Her" and of this man, who fell in love with a voice, with an evolving AI that became his best friend by dint of its ability to answer him and to interact with him like a real person that one would only have on the phone or on social networks because it is at the other end of the world. Maybe, in a few years, the next generations, used to these AIs and robots, will be able to let themselves be fooled, to confuse everything, to appreciate, to even love, these non-individuals for what they are not. But isn't this the antechamber to the madness that awaits us?
And... is this the society we want our children, grandchildren and great-grandchildren to live in? In this relationship with "reality" - a notion that can evolve faster than expected...?
I consider it as positive and very educational that these decoys are publicized. Beyond the "buzz", the surprise and the people who find (sincerely, poor people...) great that tomorrow we will have nothing to do (long live the monetary helicopter and the metaverse), these amusing deceptions have at least the merit of awakening our conscience and questioning us. Are we going to let this happen and, if so, how far?
Personally, I find it interesting that AI can bring help, acceleration, in useful domains like medicine, where there is an emergency for a good number of still incurable diseases, but I think that it should only be solicited with parsimony, and not without a good part of ethics, in domains where the proper of man (including for his physical and mental health) is to exploit and develop his intelligence and his gifts, to fulfill his mission of "co-creator" on this planet. I think that man, who has an annoying tendency to idleness and destruction, should continue to develop and take care of the intelligence he possesses, which deserves to be worked on rather than replaced. Relying on the machine to repair what he could avoid breaking and "creating" works that will never have the symbolic, cultural and civilizational scope and value that his own have had, seems to me pernicious. At a time when we talk about ecology, sustainable development, return to the earth, preservation of resources, short circuit, isn't the use of AI a bit paradoxical? Isn't the shortest circuit between the mind and the object, between the imagination and the work, the brain/heart/hand path? Why make a detour via an energy-consuming, inevitably polluting and highly random externalization, a tool that we may control less and less?
In fact, to summarize, my very personal opinion is similar to that of Margaret Atwood, who said this:
"When a civilization is nothing but ashes and dust... only art remains". So what will be left if art itself, this supreme means of expression, is no longer within its competence, having entrusted it to algorithms?
I think again of James Joyce's words: "What matters above all in a work of art is the vital depth from which it has sprung." What to say...
Above all, I fear, like Hannah Arendt, that with this possibility of letting the robot do, imagine, shape in its place, society will end up "not wanting art, but only leisure"...
And if tomorrow, art is no longer, as André Malraux said, "the shortest way from man to man", but that of the machine to man, what will be the objective of this machine? Our subjection to a tool which, far from giving us the change, by dint of digesting our knowledge, our words, our reactions, our logical senses, will end up supplanting us?
Will Michael Jackson start a "new" posthumous career thanks to Artificial Intelligence? Will Michael Jackson reconstituted by avatars and voice synthesis and touring the world in 2059 be the perfect double of the one we knew? Whatever the lucrative (rather than artistic) ambition of such projects, I bet that for us, the answer is no. For the next generations, I prefer to keep hope...
*Artificial neural network: In the field of artificial intelligence, an artificial neural network is an organized set of interconnected neurons allowing the resolution of complex problems such as computer vision or natural language processing.
It is a particular type of machine learning algorithm (such as support vector machines (SVM), decision trees, K-nearest neighbors, etc.) characterized by a large number of layers of neurons, whose weighting coefficients are adjusted during a training phase (deep learning).
There are many types of artificial neural networks such as recurrent neural networks, auto-encoders, transformers or generative adversarial networks. Source : CNIL
**François Pachet is a scientist, musician and researcher at the streaming giant Spotify. He is behind Hello World, the first album composed with the help of artificial intelligence, released in 2018.
Want to know more?
🎧Listen to this short podcast by France Musique : "Musique et Intelligence Artificielle, le son du futur est déjà là !"
📖Read this article by France 24, Musique et intelligence artificielle : "l'idée d'une substitution de l'artiste est un fantasme"
🎧Listen to the first music made by a computer, the "Suite Illiac"
And... don't forget to help me finance the translation in English of my latest book, from my doctoral thesis on Michael Jackson, by clicking on the following link:
I need you!