Why You Have 600 Muscles to Do a Job That Needs Far Fewer
The "extra" muscles in your body aren't redundant engineering. They're the reason movement can have quality at all — and AI is finally learning to work at their level.
Hold your arm out in front of you and slowly raise it. Simple movement. Now do it again, but this time imagine you are lifting something precious and fragile. Then again, as if pushing through deep water. Then again, as if you couldn't care less.
Same arm. Same path through space. Same start and end. And yet four completely different movements — anyone watching would see four different things, and you would feel four different things doing them.
Where does that difference live? Not in the path your hand traced; that was identical every time. The difference lives in your muscles — in how they organized themselves to produce a movement that, from the outside skeleton's point of view, was exactly the same each time.
This is one of the deepest and least-appreciated facts about the human body, and this year, AI is finally beginning to work at the level where it matters.
The "Problem" That Is Actually the Point
Here is a fact that puzzled scientists for most of the twentieth century. The human body has roughly 600 muscles. But the number of independent ways your skeleton can actually move — the joints, the directions they bend — is much smaller, a few hundred at most. You have far more muscles than you strictly need to move your bones around.
For an engineer, this looks like bad design. Why build a machine with hundreds of redundant motors when fewer would do the job? If you only need to bend the elbow, why have so many muscles influencing it?
The Soviet physiologist Nikolai Bernstein, working in the mid-1900s, realized this "redundancy" was not a flaw. It was the source of something essential. Because you have more muscles than you need, there are always many different ways to make the same movement. You can bend your elbow by contracting the bicep and letting the tricep relax — smooth, efficient, easy. Or you can bend it while also tensing the tricep, so the two muscles fight each other and the joint becomes stiff and controlled. Same bend. Totally different movement.
The "extra" muscles are what make movement quality possible. Without redundancy, every movement could only be done one way. With it, every movement can be done in countless ways — effortful or easy, held or released, anxious or settled. The redundancy is not a problem to be solved. It is the room in which quality lives.
What Somatic Practice Knew All Along
If you have ever taken a class in something like the Alexander Technique, Feldenkrais, yoga with an attentive teacher, or any serious dance training, you have encountered this fact from the inside — even if no one mentioned muscles or Bernstein.
A good movement teacher spends much of their attention not on where you move but on how you move. They will watch you do a simple action and see that you are using far more effort than you need — tensing muscles that don't need to be involved, holding your shoulders, gripping where you could release. They will guide you to do the same movement with less, with different organization, with a changed quality. And when it works, both you and the teacher can feel and see the difference, even though the movement looks, from the outside skeletal path, exactly the same.
What the teacher is working with, whether they name it this way or not, is your muscular organization — which of the countless possible muscular solutions you are using to perform a given movement. Somatic practice is, in large part, the art of becoming sensitive to your own muscular organization and learning to change it.
So here is a striking convergence. Biomechanics, working from physics and measurement, discovered that movement quality lives in the redundant muscular solutions to a movement. Somatic practice, working from felt experience and centuries of embodied pedagogy, discovered the same thing from the other side — that the quality of a movement, independent of its path, is real, perceptible, and trainable. Two completely different traditions, one conclusion: the meaning of a movement is not in where it goes, but in how the body produces it.
Where AI Has Been Stuck
Until very recently, AI systems that generate or analyze human movement worked entirely at the level of the skeleton. They represented movement as a sequence of joint positions — where the elbow is, where the knee is, frame by frame. This is the level that cameras can see and that motion capture records.
But we have just established that the skeleton's path is precisely the level at which movement quality is invisible. Remember the four arm-raises: identical skeletal paths, four different qualities. An AI working only with skeletal data literally cannot tell them apart. The information that distinguishes them — the muscular organization — was never in the data it received.
This is why AI-generated movement has often felt subtly lifeless even when it is technically accurate. The joints go to the right places. But the quality — the dimension that the muscles carry and the skeleton does not — was missing from the start.
What Changed This Year
This month, several research teams released AI systems that work at the muscle level rather than the skeleton level. The most significant, an open-source framework called MuscleMimic (from teams at EPFL and McGill), builds detailed models of the human body with hundreds of individual muscles — 416 in the full-body version — and trains AI to produce movement by activating those muscles, the way a real nervous system does.
The breakthrough that made this possible was computational. Simulating hundreds of muscles accurately is enormously expensive, far more so than tracking a skeleton. For years it was simply too slow to use for training AI. New techniques using graphics processors (the same chips that power video games and, increasingly, all of AI) made muscle simulation fast enough to train on hundreds of movements in a matter of days.
And here is the result that matters most: when the researchers checked the AI's muscle activations against real measurements from human bodies — using EMG, the technology that reads electrical signals from muscles through the skin — they matched. The AI was not just producing movements that looked right. It was activating muscles the way real bodies do.
Why This Matters Beyond the Lab
An AI that works at the muscle level is, for the first time, working at the level where movement quality actually lives. It can represent the difference between a held gesture and a released one, between effortful and easy, between the four arm-raises — because those differences are muscular, and the AI is now muscular too.
For anyone interested in movement as an expressive, felt, meaningful human activity — dancers, somatic practitioners, movement therapists, athletes — this is the development that finally points AI toward the dimension they care about. Not movement as a path through space, but movement as a quality of being.
A caution, as always in this series: working at the muscle level closes part of the gap between AI and lived movement, but not all of it. Measuring a muscle's electrical activity is still not the same as feeling your own movement from the inside. The felt sense of moving — the first-person experience that somatic practice cultivates — remains beyond what any external measurement can capture.
But the muscle level is much closer to that felt experience than the skeleton ever was. The "extra" 600 muscles, it turns out, were never extra. They were where the meaning was all along. And AI has finally started to listen at their level.
Further Reading
Bernstein, N. A. (1967). The co-ordination and regulation of movements. Pergamon Press.
Li, C., Wang, C., Ziliotto, B., et al. (2026). Towards embodied AI with MuscleMimic: Unlocking full-body musculoskeletal motor learning at scale. arXiv:2603.25544. https://arxiv.org/abs/2603.25544