A revision of our Fermata.js project in commission of the Knowledge Centre for Artificial Intelligence EhB in partnership with the I Love Science Festival in Brussels. Fermata.js uses AI frameworks to generate music. This version uses Markov Chains to work its probability magic to come up with always evolving melodies, chords, modulations and more.

The first version of Fermata.js used Magenta's Piano Genie model in a Twister-like physical interface, but it actually served as an amazing backup plan to what what Fermata.js originally intended to be, and got to be with Fermata.js 2.0
Our original installation had a big flaw: it was very fun and exploratory to touch buttons all over the place⁠, but there was no real indication on how to actually control this as an interface. Every button was just a button. ⁠
While Piano Genie works really well, and doesn't feel too random, it also feels aimless really quickly, especially when you don't control it in it's intended way, with eight buttons ranging from lowest to hightest pitch. Our Twister-box just had those two spread out over 34 buttons without description.

With the science festival in view, we wanted to shift Fermata.js more towards the direction of a tool, than a plaything. Something you could steer, use as inspiration, maybe even derive actual musical material from. The system that we used was made by Rich Vreeland, better known as the composer Disasterpiece, and used markov chains to navigate harmony, and only harmony.
At any time, the system has a generated tonic or root note, and a harmonic mode. Users then have to the option to play around in that harmonic space by playing individual notes, intervals or chords, all steered by markov chains. They also have the option to alter the harmonic space, by either changing the current harmonic mode, i.e going from aeolian to lydian, or change the mode and the tonic, causing a big harmonic shift.
By setting clear boundaries as to where to system can go, the harmonic space it conjures up is always dazzling.

While it didn't make it into the final project, we were also toying around with implemting a kind of economy into the system, to steer users and gradually introduce them to all the features. For example, one would start out by only playing single notes, unlocking intervals rather quickly, and unlocking chords a bit later, using a point system to allow or disallow users from overutilizing very dense things like chords, hoping it would cause people to play tastefully. After exploring this harmonic space sufficiantly (by building up points), you could unlock the option to move on to different harmonic spaces.
This whole economy didn't get implemented in time, since we were having trouble with the LED rings inside of the buttons. They would have represented this whole economy, and without the visual feedback of this system, it just wouldn't have made any sense to keep it in, but is probably something that will make its way into a next iteration. In the end, it was probably better that we didn't get to use it, as in reality, it wouldn't be able to handle 10 kids collectively smashing the buttons repeatedly.