Why is music everywhere?

We often found ourselves seeking to perceive the myriad sounds in the world as if they were notes on a piano. Whether it be the chirping of birds, the whispers of the wind, or even the hum of a car engine, each sound, at some moment, could resemble a brief excerpt of music to someone. While defining music proves challenging, the art of composing harmonies that resonate with sounds of nature feels instinctively human. Likewise, exploring why music abounds in our world seems an inherent quest for us.

Our work, recently published in Nature Communications, seeks to shed light on these questions. We report that an ability to process music can emerge spontaneously in an artificial neural network as a byproduct of adapting to natural sound processing.

Brief overview

Music, often referred to as the universal language, is a cultural universal. Researchers have found that music exists in almost every society and has common rhythmic and melodic patterns worldwide. Neuroscientists have identified specific regions in the auditory cortex responsible for processing musical information in our brains, even in individuals without explicit musical training. This raises the question of how such universality can emerge, despite the spectacular diversity of sensory input from different cultural environments. 

We used artificial neural network models to show that the ability to process music can emerge spontaneously, simply by training the network with natural sounds. Using Google's extensive AudioSet dataset, the artificial neural network learned to recognize various soundscapes. Interestingly, we found neurons within the network model that responded selectively to music. These music-selective neurons exhibited high responsiveness to a variety of music genres, such as classical, pop, rock, jazz, and electronic music, while showing minimal response to other sounds such as speech, animal sounds, and machinery sounds.

Further investigation suggested that music selectivity may work as a functional basis for the generalization of natural sounds. First, we found that music-selective neurons in the network mimicked the response properties of neural populations in the music-processing regions of the human brain. For example, the artificial neurons encoded the temporal structure of music, showing a minimized response to temporally shuffled music clips. Furthermore, we found that the process of generalization during training is critical for the development of music selectivity. In line with this, inhibiting the activity of music-selective neurons in the network significantly reduced the ability to process other natural sounds. 

In a nutshell, our proposal centers around the notion of music as an 'instinct', suggesting that an evolutionary adaptation to process natural sounds may have contributed to the development of a universal template for music. Nonetheless, it should also be noted that the current study primarily focuses on the rudimentary stages of music processing and does not consider the complete developmental aspects of music learning.

We invite readers to delve into our original paper for a more detailed exploration and potential implications of our findings. Finally, we express our sincere gratitude to the collaborators and reviewers for their invaluable comments.

This research was supported by the National Research Foundation of Korea.

The images was generated by DALL·E AI based on the contents of our paper.