[Reading Science] 'There Is a Musical Instinct'

KAIST Discovers Principle Behind Brain's Music Instinct Without Learning Using Artificial Neural Network Model

The notion that "music is a universal language" has been scientifically validated through artificial intelligence.

KAIST (President Kwang Hyung Lee) announced on the 16th that a research team led by Professor Ha Woong Jeong from the Department of Physics identified the principle by which musical instincts can emerge in the human brain without special learning, using an artificial neural network model.

Distinguishing Music and Non-Music in the Latent Space of an Artificial Neural Network Trained on Natural Sounds Without Music.

The research team utilized a large-scale sound dataset (AudioSet) provided by Google and discovered that neurons (units of the nervous system) selectively responding to music emerged spontaneously while the artificial neural network was being trained to recognize various sound data.

These neurons did not respond to human conversations, animal cries, sounds of trees swaying in the wind, rain sounds, or mechanical noises, but showed high responsiveness to various types of music, including instrumental and vocal music, and were formed spontaneously.

Musicality of the Brain and Artificial Neural Networks Illustration (Generated by DALL·E3 AI based on the paper). Provided by KAIST

These artificial neural network neurons exhibited response properties similar to neurons in the brain's music information processing areas. These properties were not limited to specific music genres. They were commonly observed across 25 different genres, including classical, pop, rock, jazz, and electronic music.

Professor Ha Woong Jeong stated, “Through these results, it is expected that artificial implementation of human-like musicality can be used as a foundational model for music generation AI, music therapy, and music cognition research.”

However, there are clear limitations to this study. Professor Jeong added, “This research does not consider the developmental process through music learning and should be noted as a discussion on the basic music information processing in the early stages of development.”

Dr. Kwang Soo Kim from KAIST’s Department of Physics (MIT Department of Brain and Cognitive Sciences) served as the first author, and the study was conducted together with Dr. Dong Kyum Kim (IBS). The research was published in the international journal Nature Communications (paper title: ‘Spontaneous emergence of rudimentary music detectors in deep neural networks’). This study was supported by the National Research Foundation of Korea.