Music Information Retrieval from a Multicultural Perspective

For me music information retrieval is completely new domain. I had several thoughts in the past about it such as – when you upload a clip to youtube how do the make sure you don’t violate any copyright? how do you do it fast and scaleable? How do you recommend someone a new music? Can you just design it like any other recommendation system? How to process the signal to differentiate and recognize the different instruments and so on.

As a data scientist and software engineer it is clear that when choosing features, which features to extract, we are skewed. We have a gut feeling and we usually follow that. Even when choosing which algorithms to use or which tool to use we usually turn either to something we already succeeded with and have experience or to a new algorithm \ technology we want to learn. We are sometimes blind to those biases.

Big parts of software development and data science is done on the western world, i.e US, Europe, etc. there are cultural biases to the engineers and the way they think. Our gut feeling is biased. In this case our gut feeling is biased toward how we in the western world understand what music is and how it is built – scores, notes, intonation, etc. Giving the examples in the lecture and stories behind them, it is clear that the way music is interpreted and experienced different in different cultures.

It is quite shocking to think about science that way – although we think we are quite rational and following some scientific \ research \ independent process we are biased to begin with. The implications for me are immediate – how recommendation system work (not only music). Simply think of movies, maybe the categorization to genres is external to other cultures? Maybe they care about different features in products?

I’ll try to bare it in mind next time I design a recommendation system.