Elad Liebman, a University of Texas Austin Ph.D. student, along with Peter Stone, UT Computer Science Professor and Professor Maytal Saar-Tsechansky of UT Austin’s McCombs School of Business, developed an intuitive software that can create playlists based on the AI’s perception of the user’s mood. The goal for its development is to bring music streaming ability to the next level, from recognizing music choices based on user preference, to user’s present mood.
Liebman, who also has a degree in music composition, says the experience is akin to having a personalized deejay who plays songs based not only what a listener personally enjoys. The AI also has the ability to intelligently organize the music selections into a playlist that can respond to how the listener feels at the moment.
Saar-Tsechansky, who is a professor of Information, Risk, and Operations Management said that their goal is to expand music streaming services by providing a software that can create playlists that adopt to a person’s frame of mind when shifts in emotions occur. The McCombs School of Business professor, elaborated by citing instances of mood changes, such as when an individual gets into his or her car after a day of long meetings; or when waking up on a weekend morning.
Saar-Tsechansky even goes further, by saying that the AI can also work in any business setting engaged in recommending things in a specific sequence, to humans.
How the UT of Austin’s Personalized Deejay Works
The AI which works on a mechanism called Monte Carlo Search can generate 10 songs as a playlist. While one is playing, the engine will look into tens of thousands of possible sequences that can predict the nearest type of music that the listener will enjoy. The predictions for the playlist sequencing continues with every new song that a person plays. In deference to its function and the Monte Carlo name of the search engine, programers Liebman and Stone call their AI, Deejay MC.
Liebman said that the program can be adapted to other types of media, from news stories to various kinds of videos. According to Liebman,
“Learning algorithms do not deal with taste, they just deal with data,” “A programmer can simply change the dataset with anything, provided people are using the program in a similar manner.”