A 16-Foot Sculpture That Listens to AI Music
In September 2024, on a stage at the MIT Media Lab, the GRAMMY-winning keyboardist *Jordan Rudess* sat down at a piano to perform a sold-out concert. Next to him on the stage was a 16-foot-tall kinetic sculpture: a 5×5 grid of motorized metal "leaves" that panned and tilted in real time.
Behind both, an AI called a *jam_bot* was generating music in response to what Rudess was playing, and the sculpture was moving in response to the AI. The piece was the result of a year-long collaboration between Rudess and a team of MIT researchers led by Perry Naseck, Lancelot Blanchard, and Professor Joseph Paradiso.
The audience that night watched a human musician and an AI improvising together, with a third entity — the sculpture — translating the AI's invisible decisions into a visible gesture. The piece was not a demonstration.
It was a performance. What this work solves is a real problem in AI performance: *the audience cannot see the AI thinking.* When a human plays, the audience reads the body, the instrument, the breath.
The audience was able to watch an AI's intention in real time. Not its output. Its intention.
When an AI plays, none of that exists. The result is a kind of interpretive gap — the audience hears the music but has no way to attribute the music to a performer. The sculpture fills that gap.
It gives the AI a body. The technical layer is interesting but secondary.
The deeper move is the *authorship move*. Rudess is a virtuoso improviser.
The jam_bot is a generative model trained on his style, capable of extending his lines and proposing futures.
The sculpture is the AI's choreography — a visual proxy for what the AI is "about to play." In one of the most compelling pieces in the set, the team made the sculpture visualize not just the music the AI was playing but the music the AI was considering — the "musical futures" the model had generated but might discard. The audience could see a moment where the AI was about to take the music in one direction, then regenerate, then go somewhere else. This may be the first time a public audience has been able to watch an AI's intention in real time.
Not its output. Its intention.
The implication, in this writer's view, is bigger than the piece. If AI intention can be made legible — to a viewer, to a co-performer, to a critic — then the conversation is no longer "AI replaces humans." The conversation becomes "AI performs alongside humans, and the audience can see what each is doing." That is a different conversation, and Rudess and the MIT team have made it real.
What this work also suggests, beyond the AI context, is the value of making process visible as a design principle. The audience is not consuming a finished product.
The audience is witnessing a process. The sculpture is honest in a way that a closed system is not — the audience sees the AI hesitate, change its mind, choose one path over another. The performance is a record of a process, not just an output. That is a generous thing to offer an audience. It is also, in 2026, increasingly rare.
Until the next transmission.