Why can't modern musicians make great music?

“I don't see a dystopian future” A conversation with Stephan Baumann

Stephan Baumann could be the coolest music professor you've ever met. The researcher and musician works full-time at the German Research Center for Artificial Intelligence (DFKI) and makes electronic music in his free time under the artist name MODISCH.

At the beginning of the millennium, Baumann and others founded a company called SONICSON specializing in music recommendations. The music recommendation systems area has grown rapidly over the past two decades. Baumann gave lectures on the technology of AI-supported music at the Popakademie Baden-Württemberg and now works for the DFKI in Berlin and Kaiserslautern.

His dual identity as a researcher and active musician gives him unique insights into the future of music. In an interview with the Goethe-Institut, Baumann talks about current trends in AI-supported music, the influence of the tech industry and the new environment in which the musicians move.

The streaming service Spotify has been accused of paying non-existent bands to produce music for its playlists (which the company denies), and Shazam now operates a record label that signs new bands based on its user data. Are we being manipulated by the big tech companies, or is the increased use of user data opening up new opportunities for music lovers?

Stephan Baumann | © Marie Gouil It's difficult to stay up to date and not be overwhelmed by this type of manipulation, I agree. It's a bit risky because now we not only have access to recommendations of content that has been produced for others, but we also get recommended content that has been created for us and tailored precisely to our tastes.

We know that AI machines are trained on certain styles of music. At some point the music industry will give us exactly what we personally want, for example the perfect house melody 2020 with a disco influence from the seventies and a pleasing singing by the former singer from Moloko. The potential for this is already there.

Is this a vision for the future or is it already happening today?

In music I already see today what the future will bring. This is of course due to my job, and not because I could foresee the future. Right now we are working with the most modern cutting edge technology. Take the jukebox project, for example. I was really speechless when I saw what can be generated automatically there. Without any human interaction, a complete song including vocals and imaginative lyrics is created. However, the complex calculations still produce a few sound artifacts. I think it takes ten hours to produce about a minute, but it's really breathtaking.

A sound artifact is a minor disturbance. To put it simply, scientists go on and on into tiny windows of time where they can find all the elements they need for a particular syllable or note, and then put it all back and forth until it finally sounds like one of the Beatles singers would sing a made-up song. But in some places it's not perfect. I suspect that this technology will be optimized in ten years and can be used to have the perfect song for the individual listener produced by a machine instead of a musician.

The question is: why? The machine does not understand the meaning of life and death. As musicians, we make music to compensate for our feelings or just to have fun, but the machine doesn't care. So we have to ask why: what is the point of machines doing the composition for us? The idea is crazy. The only benefit might be that a professional composer could stop writing boring music for a mass audience. Instead, he could make sophisticated avant-garde music. So I don't see a dystopian future coming, but very, very progressive musical tools. A drum computer, a sampler, AI-generated music. No problem. I will continue to have fun with my music machines as I write my own songs.The robot band Compressorhead can play an acceptable song, but do they really feel it? | © dpa picture alliance / Alamy Stock Photo Could this technology threaten musicians' livelihoods in the future?

I think no. If you talk to musicians in the avant-garde scene or to young musicians, you will find that they are already using AI-generated material and enjoy working with it. They give it their own character. Holly Herndon is one of the figureheads of the scene - extremely cranky and an independent star. She has a PhD in computer science. She has created a whole universe of AI-generated sounds and voices that she mixes with her own music, which she releases in albums and with which she goes on tour. In my opinion, AI is just another facet in the artist's tool palette.

A few years ago, no one could have foreseen the rise of the Bandcamp website. We talked about Spotify and whether manually created playlists would prevail against the recommendation algorithms - and then came Bandcamp. For the young, it's like folk music back in the sixties. You can be closer to the artist and the movement. I find the origin of these counter-movements very interesting. Every time we think, "Oh God, we are lost," there is a countermovement. Is that maybe human nature? Fortunately, we continue to deny the technological singularity.

Will the extensive use of AI-based music recommendation systems by consumers lead to the emergence of more monotonous, more homogeneous modern music?

Unfortunately yes. When I see Spotify's Top 10 or listen to what my kids are listening to, there's a lot of futuristic hip-hop power pop with autotuning vocals. I don't know what to call this - and I don't like it. I don't like that I can't tell the songs apart; I can't even tell if it's Beyoncé or whoever it is. It all sounds the same to me, and I believe that the filter bubbles in the recommendation systems contribute to this.

But I also see some counter-movements in niches - I see people who want to do things differently. So I wouldn't say this is a slowly sloping curve that we'll all slide down on until we end up with just one song for all of the world's population!

But I still wonder where the real creative act is to be found. In the AI ​​world, we strive for forms of real creativity, but that is very difficult. It's difficult even for established artists, who are often asked about their musical influences and authenticity. Where's the point where a great artistic idea flashes - something that no one has done before? We can have bespoke music, AI-powered music that is exactly what we are socialized on. But where does the new factor come into play? The creation of something new? So far I can't see that in AI-generated music or in algorithms.Custom-made music may be possible with AI, says Baumann, but that doesn't mean it's creative | Photo: sgcdesignco / Unsplash I've been looking at the board game software from Google's DeepMind recently. These systems not only defeat world champions, but they also train each other without human intervention. In the past, these programs knew the perfect traits of their human opponents. Now they play against each other and make moves that a person would not choose, but which lead to success. It's a little scary. Garry Kasparov has analyzed the new AlphaZero software and explained that it is a kind of synthetic creativity: purposeful moves that a human would not make. I wonder if we can apply this algorithmic technology to music. Could new music movements then perhaps arise?

But there is one thing we must not forget: with these board games, it is easy to describe the goal. The machine is told: "You are perfect when you reach a certain state - you have won or lost." In music and art, however, the result is open. It is much more difficult to describe the desired end state. You can't say, “This is what the perfect song looks like. Try to make one yourself. ”That is impossible.

In other interviews you talked about AI technology, which measures the listener's mood and other reactions to music and reacts accordingly. Are we ready yet?

We already have the hardware to measure that. Here at DFKI I am working on a project in which we measure the reactions of listeners based on heart rate, skin temperature and galvanic skin reaction. We know these factors, and now there is stable hardware in place so that we can measure what is happening. But interpreting the data is still very, very difficult.

We can try applying machine learning and algorithmic AI processes to it, but it's still difficult to say exactly what's going on. If heart rate and skin temperature change when people hear a certain chord, it can mean different things to different people. So you really need excellent data, which you can only collect if you conduct in-depth interviews with the test subjects whose reactions are being measured. You have to find out what was actually going on inside them. And you have to be able to rely on them reporting precisely and objectively about their subjective mood and what the music means for them personally. I would say this is really the holy grail in my field.

More keywords from Stephan Baumann about the future of creative AI can be found here.