Sarah Fdili Alaoui

Executive Summary

Sarah Fdili Alaoui bridges artistic practice and scientific research through a practice-based methodology where technology functions not as body augmentation but as provocateur. AI disrupts embodied practice in ways that engage artistic craft rather than replace it. Her work marks a fundamental shift from viewing AI as collaborative partner to recognising it as a generative probe that provokes movement without claiming creative agency.

I was very much looking at technology as a way to extend the body, augment the body. Now I'm looking at technology as a provocateur — something that would disrupt it in a way that will engage me as an artist into doing my craft.

Her breakthrough work *For Patricia* (2024) embodies this philosophy through a radical performance structure: AI trained on Trisha Brown's choreographic patterns directs live dancers and musicians in real time, generating ongoing surprise that requires acute listening and zero-latency physical response. Critically, AI does not generate movement in *For Patricia*. Sarah maintains creative authorship over all choreographic material and uses AI as an algorithmic director that shuffles her choreography according to the postmodern dance principles it has learned from Brown's archive.

For Patricia: Material Versus Structure

The conceptual move at the heart of *For Patricia* is precise. Sarah retains creative authority over all movement material while ceding structural authority to the algorithm. She and her composer create every choreographic and musical element themselves. The AI reorders it.

I wasn't interested in generating material because I was interested in doing that myself as I wanted to behold the creative power. But I wanted something that would direct us on stage, that would give us a certain structure. It's like a very elaborated dice that shuffles things around.

The work is an homage to postmodernism. Choreographers like Trisha Brown and Merce Cunningham used dice, chance procedures, and systematic probes to generate and structure material. Sarah's AI is the contemporary equivalent: a systematic probe for provoking movement rather than a creative agent in its own right.

The work emerged through serendipity. Initial plans involved finding explicit rules to restructure material. During a residency in Bucharest, a librarian offered a hard drive containing Brown and Cunningham video data from the National Centre of Dance archive. Experimenting with that material, the team realised rules could be implicit, already existing within the vocabulary itself.

All of these things in a creative process is very normal. You spend two months working on something and then there are all these coincidences that arrive and give you an idea and then it becomes the central core idea of your piece.

Acute Listening as Symbiotic State

The performance structure creates a cognitively demanding real-time experience. Dancers and musicians learn all sequences perfectly, thoroughly rehearsed, with no improvisation in the vocabulary itself. The AI directs which sequences to perform, in what order, including reversals and variations, displaying instructions on screen using colour codes. One hour of constant acute listening.

You have to be in an acute state of listening, understanding where you're going at every moment, every second. You're never stopping and thinking what is the machine doing. You're constantly going with the flow of what the algorithm is doing. There is very little distance between us and the machine in that sense. We kind of just perceived it, put it in the body.

Sarah and her composer built the system themselves, knowing exactly what it was doing based on what features it was using and how it was considering input data. Trust here is grounded in technical comprehension. Even so, the generative system resists full cognitive transparency.

Even if cognitively it's always difficult to make sense of these generative AI systems because there's very little understandability behind it, you can perceive what it's trying to do. But that's not the point. The point is I wanted something to create with me. I create the material, it creates the structure, and then we play together.

Surprise is structural rather than incidental.

Surprise is so present in the piece that it's actually difficult to flush it out. We go on stage and all we have is an AI that's surprising us at every single moment. That's basically what we're playing with.

The Embodiment Challenge

Dance presents AI with a challenge unique among art forms. Film, video, and image-making operate through screen representation, AI's native medium. Dance requires bodies.

AI doesn't have a body and it doesn't understand bodily interactions. So it's the most challenging of all. Even movement generation systems — there always needs to be a body to interpret that. It could give you the dance movement but it's on a screen, and that's not what dance is.

This shapes her research with PhD students, which develops small-scale, vocabulary-specific generative models. Training on voguing or dancehall, with motion capture captured directly from practitioners, preserves choreographic identity and creates archival potential. Generic models trained on massive datasets produce fluid, physiologically plausible movement but lose the distinctive characteristics that define particular dance forms.

The more you train it with generic data, the better models are. But the more you do that, the more you lose specificity. You gain in terms of realism, but you lose specificity.

Neither direction fully resolves the problem. The choice of which to optimise reveals what the practitioner believes dance is for.

Specific Deployment Versus General-Purpose Tools

Sarah's practice distinguishes carefully between artistic deployment and general-purpose tools. In *For Patricia*, the team spent extensive time training the AI, understanding its behaviours, integrating it into the creative vision alongside costumes, sound, and lighting. The AI became part of the set, contributing to the overall artistic project within a defined role inside a bounded system.

This differs fundamentally from aspirations for AI as choreographic partner in the way large language models function for writing. Those systems are general, public, and asked to operate across any context a user brings. The AI in *For Patricia* operates inside one work, with one company, against one vocabulary.

I understood that for me dance remains a matter of bodies and what the machine is there to do for me is to provoke me into movement. That's why it's very postmodern, because that's what postmodernism did — find all kinds of probes to provoke movement.

Agency Over Explainability

When Sarah describes her ideal future AI system, the priority is clear. Practitioner control matters more than algorithmic transparency.

I would want it to be very malleable and adaptable to people's work and vocabulary and choreographic thinking. I wouldn't want a blackbox. I would want something trainable, where people have lots of control. Not that much explainability — I don't think anyone wants to understand what these algorithms are doing — but control and agency that you're able to say to an algorithm: this isn't what I want, I want it to do it this way.

The distinction matters. Explainability asks too much of most practitioners. Controllability asks the right amount. The algorithm should have interesting things to propose while remaining faithful to what practitioners programme or direct.

All four collaborators on *For Patricia*, two dancers and two musicians, maintained autoethnographic journals throughout the creation process. The resulting analysis revealed six distinct phases of development, shaped heavily by serendipitous encounters that became central organising principles. The creative process, Sarah argues, is always like this. Months of structured work, then a chance conversation in a Romanian dance library, and the whole piece pivots. The probe is real. So is the dice.