Magenta was created by Google in order to explore the possibilities when it comes to artificial intelligence and creative pursuits like music, and has now been taken a step further with the AI Duet experiment, made by musician and coder Yotam Mann, alongside the Google Magenta team. The experiment uses Magenta as a backend for a bot that tries to harmonize with and complement a user’s input on a piano. Rather than feeding a bot specific directions for how to respond to the massive number of possibilities that it could encounter in processing users’ input from a virtual piano, coders decided to use Magenta to allow it to learn on its own, feeding it only a few key pieces of data to start it off. The result is a bot that is actually pretty decent at what it’s for, and can handle any user from the most experienced virtuoso to somebody simply mashing keys.
While feeding a bot your favorite piano melody and seeing what it comes up with in response is pretty novel, it’s not the central point of the experiment; the whole thing is open-source, and can be built out as a developer wishes to be made into a variety of different musical curiosities. The point is to show off what Magenta is capable of, and drum up interest in using it for musical projects. It is the first AI of its kind, tailor-made to explore music, with a main mission being to bridge the creative gap between humanity and AI, when it comes to the language of music.
This is not the first time that an AI has exercised musical creativity, of course; an AI bot wrote a simple song on the piano last year, which came out sounding a bit disjointed, but certainly passable once human accompaniment was added. Magenta’s potential far exceeds that simple exercise, and with machine learning built in, the project is going to improve its creative chops over time. Magenta’s prowess does go beyond music as the project is also aimed at visual arts. The goal of Magenta is to create a “truly generative” creative AI, and document every step on the way there. You can test out how the AI duet works yourself by heading through the link below.