Photo: Manuel Nieberle
AI in Music: “Much More Creative Than Us in Some Aspects”
By Martin Haase
Artificial intelligence can compose. And that is a benefit for musicians, says Dr. Esther Fee Feichtner. She is a musician and computer scientist, two professions that now have an interface thanks to artificial intelligence: AI-generated music. She is now convinced that the technology has enormous potential for the media landscape.
Dr. Esther Fee Feichtner completed her doctorate at the International Audio Laboratories Erlangen and is now the head of the “Artificial Intelligence in Culture and Arts” (AICA) digitalization college. Students at the Munich University of Music and Performing Arts and the Munich University of Applied Sciences learn how they can use AI for themselves.
Dr. Feichtner, what is the aim of AICA and are there comparable colleges?
Dr. Esther Fee Feichtner: It's about giving artists the opportunity to work with artificial intelligence. We want to clarify the question of what one's own contribution to AI-generated content is. At the same time, we want to reduce the fear of the technology so that they discover artificial intelligence as a tool for themselves. With AICA in Munich, we occupy a special niche with a focus on the cultural and creative industries, while other colleges deal with digitalization in the agricultural sector, for example.
What fascinates you most about artificial intelligence at the moment?
Feichtner: Personally, I find it exciting that we have to redefine terms that we thought we knew exactly what we meant. For example: How does a machine understand what an apple is? And is this a different understanding to that of us humans? We think apple and perhaps have the taste in mind, the colors red and green, the apple as a symbol of fertility, the fall of man and so on. We may think that humans can understand this better than the machine, but AIs have been trained with this incredible variety of information and know more than a human individual. Now the interesting aspect is: how important are emotions and sensations associated with an object in human understanding?
If AI has much more information, can it be more creative than a human?
Feichtner: Most people answer “no” to this question, but I would like to point out that: Our understanding of creativity is basically how AI works. When musicians listen to music all their lives and deal with different styles, all these experiences flow into the creative process. If the composed piece then sounds new, this only means that the composer has broken down all the impressions as small as possible and reassembled them. The larger these parts are and the more of them are taken over unchanged when they are put together, the more the listener has the feeling that it is a plagiarism. In other words, the question we ask ourselves in terms of creativity is: how big do you choose these parts that you have heard before and put them together again? And AI works no differently, but has a much greater wealth of experience thanks to the amount of training data. On top of this, AI can experiment with this larger data basis without fear of its own or external demands. Humans can only do this when they are in the flow. In these aspects, AI is much more creative than we are. And we need to define more clearly what distinguishes human creativity from AI.
»Our understanding of creativity is basically how AI works.«
Dr. Esther Fee Feichtner
Photo: Manuel Nieberle
AI-generated music could bring back real creativity
As a composer, can you tell the difference between AI-generated music and a piece composed by a human?
Feichtner: That depends on what percentage is AI-generated. MusicGen, the program from Meta, generates 30-second pieces. That sounds really good. And if you now generate lots of these half-minute tracks and perhaps add human help with the transitions, I probably wouldn't be able to tell the difference without conscious listening. The fact that this works so well is also due to the fact that even without the use of AI, popular music is written according to average patterns and we are inundated with it. Most hits follow the same patterns. These songs are catchy because they sound familiar. AI can copy this very well. For these reasons, it is becoming easier and easier to reproduce what has always been on the music market. That's why I hope that radio stations, for example, will increasingly go in search of very individual music. Because if uninspired average music is so easy to produce using AI, this could be an opportunity for creative people to bring real innovation to the music market and for radio stations to gain a competitive advantage by selecting particularly creative styles.
So on the one hand, music made by humans can even be enhanced if it stands out from what is produced by AI. On the other hand, you want to give musicians AI as a tool. How does that work together?
Feichtner: It starts with the fact that I can easily play around with different AI-generated chord accompaniments for my melody. Meanwhile, I could even have a whole song generated from a melody and listen to different variations based on that melody. Then I can choose how best to emphasize my melody and the message I want to get across to the world.
There are currently countless AI tools springing up every day. What advice do you have for artists in this phase?
Feichtner: They should simply choose a tool and learn to prompt with it. Because no matter which program you end up using in two years' time: The basic principle remains the same. It's like learning a programming language. It won't make much difference which tool you add to your artistic toolbox.
Photos: Manuel Nieberle
Feichtner: The goal is “full immersion”
You have already touched on the subject of radio. What do you think of a completely AI-generated radio station?
Feichtner: That depends on how the stations use AI. On the one hand, there is the option of programming this station in the same way that Spotify works, for example. Assumptions are made and the AI is fed with information such as: The target group is 30 to 38 years old, lives in Bavaria and prefers to listen to a certain set of artists. But for me, that's just a gimmick, because it removes the added value of radio: curating music, giving listeners a choice and always introducing something new. But you could also create this AI radio station in such a way that it challenges the listeners a little. In this case, the AI first plays the station's standard repertoire. Then you program an innovation factor: for example, you could say that the AI should record three new songs a day. This then increases from day to day. This has added value for the listeners, because this is the only way to create a taste. This results from what you know plus a little innovation. In this way, a radio station can contribute even more than it does today to making the public more receptive to a wider range of musical styles.
Why do we need AI for this?
Feichtner: On the one hand, the advantage is that the AI can use the patterns within the standard repertoire to better recognize what goes down well with the listeners and then confront them with songs that only deviate a small step from this and don't cause complete irritation. We don't want people to switch off. On the other hand, you don't need a presenter who has to have the courage to justify it and take responsibility if it doesn't work.
Do you have a vision of how we could use AI in the future?
Feichtner: On the one hand, I hope that we use AI for recurring tasks that we can hand over so that we can concentrate on important tasks. So many mistakes happen in every profession imaginable just because people don't have enough time. I hope that we have less stress and more time to be human. On the other hand, with regard to the technical development of AI, I hope that we will eventually achieve full immersion for artists. At the moment, we sit in front of a computer and type in words. But the type of input doesn't really matter to the AI, it always has to convert it into a different output. This input can also be voice, movements or neural signals. So it would theoretically be possible at some point for me as an artist to simply close my eyes and fully engage with my thoughts and feelings without having to actively operate anything. I might only need a few sensors on my body. AI could then convert the signals sent to a computer via the sensors into music. This prospect really inspires me.
XPLR: MEDIA in Bavaria
Part of Medien.Bayern GmbH
Balanstraße 73 / Haus 11
81541 München
info@xplr-media.de
XPLR: MEDIA in Bavaria is an initiative of Medien.Bayern GmbH and is funded by Bayerische Staatskanzlei and Bayerische Landeszentrale für neue Medien.
