Its primary purpose the experiment was not reached, but some things are still taught the AI.
Mathematics, University of California, Irvine wrote artificial intelligence RandomForest, which analyzed more than 500 thousand songs of the last 30 years — from 1985 to 2015. The main objective of the study is to teach the artificial intelligence to write hit music, and has not been achieved.
Despite this, the neural network was able to predict what track will be a huge hit and starts to sell well. Scientists have identified the traits that unite the various well-known songs, ranging from the name of the artist, ending with the genre, the recording Studio, as well as timbre, tonality and rhythm.
RandomForest could only sounding tracks with a precision of 74% to determine its popularity, and when the algorithm gets to data with the artist name and record label on which he recorded — the accuracy of his analysis grows to 84%.
According to the study, over the past 30 years in the music became less light melodies, but much more sad tracks in which the lyrical “I” is opposed to the more positive “we”. It takes in the charts that is fun and exciting tracks, which is not very much.
Over the past 20 years, the world fell sharply, the popularity of rock music in its place came electronic with elements ambient, added the researchers.