Make AI compete in boosting/maintaining a person's mood by playing them the right music.
It is a solution to this challenge.
Create a super personalized music recommendation algorithm by making AI get better at playing a "game" with a given set of rules.
Instead of tweaking music recommendation algorithms from time to time to improve them, use an alternative way that lets maximally automatize and personalize the process.
Similarly how AI improves in playing games like chess or goa against itself or a different AI, it can improve in games with more practical aims as long as the rules are clear.
In this case, define the aim for AI to raise the person's mood through music as much as it can, when it can't raise it, it should try to maintain it so that it at least doesn't drop further.
The mood level can be measured by the AI through different means:
Subjective evaluation by the person him/herself. This is the simplest approach where the user simply presses feedback buttons while the music is playing, once the feedback is below a certain threshold, the system changes the song/melody into something different and takes it from there. If the feedback is rather good, it tries to include more similar songs/melodies.
Physiological parameters. The data about the person's brain activity (especially that of the limbic system), heartbeat, breathing, electrodermal activity, skin temperature, etc. is sent to the AI while the music is playing. Those are continuous and more objective measures if you know how to convert the data into mood scores. Currently, wireless sensor systems are well developed and don't require bulky wires or for the user to sit in one place, devices similar to the Muse headband and Empatica wristband can be used to build the system.
Combination of the two. I think the smartest way to go about this would be to first correlate an individual's physiological parameters with their subjective mood evaluations and then create an "atlas" for AI to convert continuously monitored physiological parameters into mood scores. This could be done in a preparatory phase and then the system would keep refining the connections between physiological data and subjective mood evaluation as the user gives subjective feedback while listening to the music.
An AI could compete either against itself or a different AI - a tweaked algorithm. It would take turns in playing songs just like it takes turns in making moves in a board game, after each turn, the situation would be adjusted according to the person's mood level.
Instead of selecting pre-recorded songs, AI could also mix different melodies together and even create music. Turns as moves in the game could be defined as playing music for limited time intervals, like 1 minute or till the feedback from the user drops under a certain threshold.
First feed the AI all the person's listening history, playlists, etc. Then make it learn further in the process from the user's feedback while listening to its selected/made music. If the result is not good enough after a certain time, tweak the algorithm a bit and make it play against the old version. Repeat until the results are satisfactory.
Povilas S Jul 04, 2022
Please leave the feedback on this idea
Please leave the feedback on this idea
Please leave the feedback on this idea
Please leave the feedback on this idea