François Pachet, a computer scientist focusing on artificial intelligence and director of the Spotify Creator Technology Research Lab, has dedicated his whole life to researching music. He has been active in research since 1992, including in projects supported by the European Commission, like the Lrn2Cre8 project. We asked him his views on the direction music research is taking.
François, do you have such a thing as a holy grail in your research?
Yes, you see, when it comes to research I am an obsessive guy; for my whole life as a researcher I have been obsessed by the same questions. One of them is: why do we like certain kinds of music and not others? Specifically, why do I like this song? Is my taste specific to who I am as a human being? And when I like something, is it me liking it, or is it the social pressure around me?
For example, I love The Beatles, and I am convinced it is me as an individual who likes them. But then I look at data and I see that all the people born at the same time and place as me, and perhaps so to the parents with the same lifestyle, like them too. So is taste a personal choice or is it a societal phenomenon? Or is there no answer to this question, because humans are social creatures?
How can you answer a question like that using technology, when taste in art is so subjective?
The answer is data. Even vague questions like this can be formulated into concepts that you are able to test in an experiment. Experiments bring you data on what people listen to, which you can examine for patterns in how people react to different things. If you have data, you can answer all questions. That is why a focus on data is the most important trend in music-focused research in computer science.
In the last 20 years there has been a huge shift in what is most valuable in the technology industry. It used to be algorithms, but now it is data. To explain: 20 years ago, computer science was driven by creation of algorithms, to which you used to feed data for it to processed in a specific way. Nowadays it is the opposite; you start with a massive amount of data, for which you create a very general algorithm that looks for patterns. Therefore, there are no more specific algorithms. What is specific now is the data. All significant IT companies, including Google and Facebook, have adopted this trend. As a consequence, algorithms are often open source, but data is not. This paradigm disadvantages university research, as universities do not have data to work with.
How do you imagine this trend will influence the direction of science and the music industry?
I believe in the future of data, as I know that data contains all the answers and has the power to do miracles. All you need is to ask the right questions and be able to extract the information. However, I am concerned for the impact this trend will have on human science and human experience-based industries, which I believe are threatened by data. Let's look at, for example, musicology. People have studied the theory of music for centuries. The experts in this field are musicologists, whose role it is to analyse and advise on how to play and listen to music. They set down rules about what is a good piece of music, a quality melody, the right intonation, or the best rhythm. Until now, these theories are built without using any data about how people listen to music in real life.
Today, however, we have access to this kind of data, as people access music through devices that have the ability to record information and, most importantly, share it. We cannot tell what is going on in the listener's brain, but we know when, how long, how often they listen and so on. Such data enables us to analyse more precisely what aspects of music are more successful than others. Therefore, based on such analysis we often discover that results don't always match what the experts have claimed.
In the same manner, data can defeat experts in other fields, including analytical games such as chess or Go, but also driving a car, and even medicine. Modern computers can be fed data from more cases than a doctor can ever see in person, and they can calculate further then a human brain. So when it comes to analysis based on experience, it is impossible to compete with computers. Certainly, humans will always have something computers won't: human feelings and intuition. But the power of data changes the game. Experts have to learn how to adapt to this new situation. I am not afraid for the future of human experts in these fields, as long as they accommodate to the changes technology brings.
Why is taste in music a question you are particularly interested in?
When I was a child, I listened to Pink Floyd, and I found it very boring. But the more I listened to it, the more interesting it got, and the process gave rise to my big question. Since a young age, music has been part of everything I have done. Given the major role that music plays in my life and work, the question of taste comes naturally, as likeability is a very important aspect of music; when one creates music, all that matters is to make something people will like.
Leaving my obsession apart, listening to music is one of the activities that people enjoy and spend time on the most; probably together with drinking alcohol, although I don't have exact figures. Either way, enjoying music is very human. Moreover, now it is part of the identity of young people. Hence, in my opinion, looking for answers to this kind of questions does matter.
Do you play a musical instrument yourself?
Yes, I play jazz guitar and piano.
Now for a different topic. Do you think that European Commission's Future and Emerging Technologies programme (FET) brings value to researchers in Europe?
Yes, FET offers space for variation in research, unlike other research-supporting platforms. It is the only funding available for high-risk interdisciplinary collaborative research, and it represents a place to be open minded, creative and bold. That is why it has so much value. I believe it has particularly high value for universities, which should serve as a space to explore the unthinkable. Often universities try to compete with private companies in producing marketable results, but they cannot win – they do not have access to enough money and, most importantly, data. Market demand should not be their driving force; they should be independent and visionary. But if they do not take the market into account, FET is their only chance for survival.
Do you think that research FET supports brings value to European markets? Some of the FET projects are very theoretical.
Of course: when a company launches a new product, it is thanks to a very long series of discoveries behind it, and FET makes that happen. Technological innovation is composed of a succession of steps that do not always seem important, when taken out of context; and a breakthrough will not come without previous research. What is important to understand is that some research can be highly oriented towards producing marketable products, while some is more long-term oriented. Both types are crucial, because they complement each other, and one cannot exist without the other. All technology companies understand that research is important.
I know there are great people in the European Commission who are working to make the system of research support as meaningful as possible and make sure the dedicated money is well spent. It will always be a struggle to decide how to use public money efficiently, but FET should definitely be part of the plan.