Dancing with AI: LuminAI and the Future of Co-Creative Tech
In an ordinary dance studio, improvisation is a conversation of movement between human partners. However, at Georgia Tech’s Expressive Machinery Lab, located in TSRB, one of the dance partners might be an AI. The lab’s director, Brian Magerko, Regents Professor and Director of Graduate Studies in Digital Media, has built a career exploring the intersection of artificial intelligence, creativity, and human-computer collaboration. His research focuses on how studying human and machine cognition can inform the development of new interactive creative experiences. Magerko’s projects range from interactive art installations to educational technology, all aimed at blurring the line between human expressiveness and machine intelligence. One of the clearest examples of this vision is LuminAI, an AI system that quite literally dances with humans.
An AI Dance Partner Named LuminAI
Three different variations of LuminAI were projected onto the Hologauze scrim, a highly reflective and transparent projection net. LuminAI can collaboratively improvise movement with a human partner using computational formalizations of improvisational movement theories from theater and dance.
LuminAI is an interactive art installation that allows participants to improvise dance movements with an artificially intelligent virtual dance partner. The experience feels like a duet: the human and the AI take turns mirroring, riffing, and responding to each other’s moves. “The line between human and non-human is blurred,” Magerko explains, prompting participants to rethink their relationship with AI technology and showing that it can be expressive, social, and playful. The installation places humans and machines on equal footing as co-creators, asking how we can co-create experiences together as partners, not just as users and tools. Participants get to explore movement creatively with the AI, often forgetting that their improvisational partner isn’t human. Earlier this year, Magerko’s team even staged the world’s first improvised human-AI dance performance, featuring LuminAI alongside dancers from Kennesaw State University. After years of research, LuminAI has evolved from a playful museum exhibit into a serious exploration of how AI and humans can collaborate to create art.
Movement Theories Meet Machine Learning
How does one teach an AI to dance? LuminAI’s virtual dancer observes and analyzes the human participant’s motions using formal frameworks from the performing arts. Initially, the system was built around the Viewpoints movement theory from theater, which breaks down improvisational movement into basic elements, such as spatial relationships, shape, and tempo. The AI represents the participant’s poses and gestures in these terms and draws on a memory bank of past interactions to improvise its responses. In essence, the agent “learns how to dance by dancing with us,” continuously adapting and expanding its repertoire of moves based on the actions of human dancers.
More recently, Magerko’s lab has begun incorporating Laban Movement Analysis, a dance theory focused on movement qualities such as effort and shape, as another way for the AI to understand and reason about what it sees. By combining Viewpoints’ structural approach with Laban’s qualitative insights, LuminAI can interpret a dancer’s performance on multiple levels, from the geometry of their motion to the expressive dynamics behind it. These movement theories serve as a kind of vocabulary that bridges human creativity and machine computation. Under the hood, LuminAI utilizes machine learning algorithms to cluster similar gestures and recognize patterns, enabling it to respond in kind with a twist of its own. The result is an AI that doesn’t just replay recorded moves but improvises new variations in real time like a true dance partner rather than a pre-programmed robot.
Human–AI Co-Creation in Action
LuminAI is more than a novel interface or a tech demo; it’s a research testbed for co-creativity, the idea that humans and AI can engage in collaborative creativity as partners. Co-creativity in an “embodied” domain, such as dance, is essentially a real-time conversation: each partner inspires and adapts to the other. While this dynamic is familiar in human-to-human settings (from jam sessions to improv theater), it is rarely how we interact with computers today. Magerko and colleagues want to change that paradigm. A recent study examined how human dancers collaborate to inform the design of LuminAI. Through focus groups with experienced dancers, the team identified key factors for satisfying collaborative improvisation. They found that a dancer’s choices emerge from an interplay between multiple influences: their creative intent, their partner’s movements, and even the surrounding environment, all guided by specific strategies and unspoken heuristics for keeping the duet flowing smoothly. In other words, successful improv partners listen and respond to each other, introduce novel ideas, and find coherence, adjusting to the mood and space around them.
Using these insights, the researchers formulated design recommendations to help LuminAI behave more like an artistic collaborator. LuminAI exemplifies a human-centered approach to AI design: the system is built to participate in the art form, not just generate outputs. The payoff is an experience where dancers report feeling truly “in sync” with the virtual partner, as if the AI understands the creative conversation.
Promoting AI Literacy Through Dance
One unexpected outcome of the LuminAI project has been its potential as a learning tool. Interacting with a dancing AI in a public exhibit can spark curiosity: How is the AI seeing my movement? How do you decide what to do next? Realizing this, Magerko’s lab recently redesigned LuminAI to explicitly foster AI literacy, helping everyday people understand AI through a hands-on, embodied experience. In another recent paper, the team describes a new setup where the installation is split into multiple panels, each revealing a different piece of LuminAI’s “thought process” to the user file. For instance, one panel might visualize the motion-capture input, showing the stick-figure or silhouette that the AI perceives. In contrast, another panel displays how the AI clusters or remixes the moves in response to them.
A Broader Vision of Co-Creative Technology
Dr. Brian Magerko’s work blurs the line between art and AI. Projects like LuminAI show how his lab fuses human creativity with technology.
Brian Magerko’s work on LuminAI sits at the edge of a larger trend: using AI to augment human creativity rather than replace it. This human-centric ethos stands out in a tech landscape often focused on automation and efficiency. Projects like LuminAI ask: What new forms of art and expression become possible when AI is a creative partner? The answers so far have been inspiring; public museum visitors have danced with LuminAI in playful exhibits, and professional dancers have treated it as an equal counterpart in live performances. “Co-creativity is a fundamentally social, human experience that computers historically haven’t engaged in,” Magerko says. “Interdisciplinary work that bridges cognitive science, the performing arts, and computer science helps usher in a future where our machines can improvise, imagine, and collaborate with us.” LuminAI’s success showcases the potential of this vision: an AI that doesn’t just do a task for you but does something new with you. It is a powerful reminder that technology can inspire and even learn from us in the most creative ways.
References
John Gunerli et al. (2024). Video Segmentation Pipeline For Co-Creative AI Dance Application.