When Your Body Becomes a Point Cloud
Somatic movement practice meets generative AI — and the gap between them tells us something important about both
Stand still for a moment and notice how you're breathing. Feel the weight of your feet on the floor, the slight tension around your eyes from reading this screen, the micro-adjustments your torso makes to keep you upright. That continuous, self-updating, felt sense of being in a body — what somatic practitioners call interoception — is the central material of practices from Body-Mind Centering to Somatic Experiencing.
Now consider what a motion-capture system sees when you stand still: 23 points in three-dimensional space, updating 120 times per second. Your spine becomes a vertical array of vectors. Your breath shows up as a subtle oscillation in thoracic joint coordinates. Your weight distribution appears as a ratio of pressure values. The richness is mathematical; the felt quality is absent.
This gap — between movement as lived and movement as data — sits at the center of a genuinely interesting collision happening right now between somatic body practices and generative AI.
What AI motion systems actually do
Generative AI motion models — the kind that can produce realistic walking animations, choreograph movement from a text prompt, or predict where an athlete will move next — work by learning statistical patterns from large datasets of recorded movement. These datasets are almost always built from motion capture: that system of reflective markers and infrared cameras that reduces the human body to a skeleton of numbered points.
The AI learns, in essence, the biomechanics of human movement. It gets quite good at the shape of motion: the arc of an arm swing, the timing of a step, the posture shifts that accompany emotional states. When you ask one of these models to generate "a person walking anxiously," it draws on correlations between body carriage, pace, and chest compression that appear reliably across many recordings.
What it doesn't learn — because the data never contained it — is what that anxious walk feels like from inside. The activation of the nervous system. The slight pressure behind the sternum. The way attention narrows to a tunnel. The continuous feedback loop between sensation and movement that somatic practitioners work with directly.
This isn't a criticism. AI systems are doing something genuinely impressive with the data they have. But the omission matters.
The felt sense problem
Eugene Gendlin, the philosopher-psychologist who developed Focusing, coined the term "felt sense" in the 1970s to describe the pre-verbal, body-level knowing that underlies our most significant insights about ourselves. A felt sense isn't an emotion exactly — it's the whole-body response to a situation before it gets named or analyzed. It's the thing that somatic practices of all kinds try to cultivate awareness of.
Felt sense is, almost definitionally, first-person. It can be pointed at, worked with therapeutically, and even inferred by a skilled outside observer — but it cannot be measured by sensors placed on the outside of a body.
This creates a structural challenge for any AI system trained purely on external movement data. The model can learn that people carrying grief tend to have collapsed chests and slower gait. What it can't learn from that data alone is that the collapse is an expression of an interior state — and that the interior state is the thing that actually changes.
A concrete meeting point
Here's where it gets generative rather than just limiting.
Consider a somatic therapist working with a client who has chronic shoulder tension. Traditionally, the therapist observes, offers verbal or touch-based guidance, and relies on the client's self-report to understand what's shifting. Now imagine adding a real-time motion analysis tool — not to replace the therapist's attention, but to offer a second register of observation.
The AI system flags a pattern the therapist hadn't consciously tracked: the client's right shoulder consistently elevates during pauses in speech, then drops when they begin a new sentence. The therapist brings this to the client's attention. The client, attending more closely, notices that the elevation corresponds to a held breath — a habitual, unconscious bracing against being heard.
The AI found the pattern. The therapist made it meaningful. The client felt it.
Motion analysis tools of this kind are increasingly available to movement educators and therapists, and practitioners are beginning to explore how machine-generated pattern recognition can extend — not replace — their observational skills. The key move is treating AI output not as the answer but as a prompt for somatic inquiry: What do you notice when you see this? What does it feel like from inside?
What each side learns from the other
Researchers working in AI-generated motion are beginning to recognize that their models can produce movement that is biomechanically plausible but expressively hollow. A generated walk might pass visual inspection but feel, to a trained observer, somehow wrong in a way that's hard to articulate — right shape, wrong quality.
Somatic practitioners have a vocabulary for exactly this. In Laban Movement Analysis, for example, movement is described not just by shape but by effort qualities — the weight, time, space, and flow characteristics that distinguish a punch thrown in rage from one thrown in playfulness, even if the joint trajectories are identical.
Translating that qualitative vocabulary — distinctions that practitioners spend years developing — into features that AI systems can be trained on is one of the more promising research directions at this intersection. It won't give AI a felt sense. But it might give it a better map.
The deepest question
The collision between somatic practice and generative AI surfaces a question that neither field can answer alone: What is movement for?
For AI, movement is currently functional — locomotion, manipulation, communication. For somatic practitioners, movement is also relational, regulatory, expressive, and transformative in ways that resist reduction to function. Sitting with that tension, rather than collapsing it too quickly in either direction, may be where the most interesting thinking happens.
The body that data can't fully capture remains, stubbornly, the body you actually live in. That's not a limitation to be fixed. It might be the whole point.
Further Reading
Gendlin, E. T. (1978). Focusing. Everest House.
Laban, R., & Lawrence, F. C. (1947). Effort. Macdonald & Evans.
Mehling, W. E., Wrubel, J., Daubenmier, J. J., Price, C. J., Kerr, C. E., Silow, T., Gopisetty, V., & Stewart, A. L. (2011). Body awareness: A phenomenological inquiry into the common ground of mind-body therapies. Philosophy, Ethics, and Humanities in Medicine, 6(1), 6. https://doi.org/10.1186/1747-5341-6-6
Tversky, B. (2019). Mind in motion: How action shapes thought. Basic Books.
Varela, F. J., Thompson, E., & Rosch, E. (1991). The embodied mind: Cognitive science and human experience. MIT Press.
Zhang, M., Cai, Z., Pan, L., Hong, F., Guo, X., Yang, L., & Liu, Z. (2022). MotionDiffuse: Text-driven human motion generation with diffusion model. arXiv. https://arxiv.org/abs/2208.15001