The Benefits of Embodied AI

The present, gathered

What the present has delivered

Artificial intelligence arrived with enormous promises about productivity, creativity, access, and the amount of work people would no longer have to do. Some of those promises are real. The tools are extraordinary. But by 2026, we are also beginning to see the strain created by the way this technology has entered everyday life.

And subscription costs are only part of the story. The infrastructure required to support large-scale AI, including chips, data centers, electricity, cooling, and new power capacity, is becoming part of the wider cost of technological life.7

So when people say they are tired of AI, I do not think they are simply rejecting technology. They are responding to what this particular form of technology is asking of them.

Three concerns

Fatigue, cost, and suspicion

Three concerns keep appearing.

The third is the more difficult problem, and this is where my interest in embodied intelligence begins.

The turn

What happens if we begin somewhere else?

Instead of asking only how much more artificial intelligence can produce, I have been asking what kind of intelligence we actually want to build and where that intelligence should begin.

That means beginning with a body that moves, senses, remembers, adapts, learns through repetition, responds to its environment, carries culture, and develops knowledge through relation with other people.

For SEE Lab, this distinction changes the entire architecture. An embodied system cannot simply gather everything first and decide what is permissible later. The person whose movement produces the knowledge has to remain present in the structure of the research. Consent, refusal, attribution, cultural authority, provenance, and conditions of use have to travel with the data.

It also changes what we expect the machine to do. The goal is not to have the machine take over the act of being human. I am interested in systems that can help a person perceive more of what they are already doing.

A dancer may know a movement intimately and still be unable to see a recurring change in weight distribution across hundreds of repetitions. Two practitioners may sense that they are moving together without being able to detect a 200-millisecond difference in coordination. A person may repeatedly avoid one region of their kinesphere without consciously recognizing the pattern. A movement tradition may contain a structural relationship to another tradition that becomes visible only when thousands of movement units are compared.

Those are useful computational problems because the machine is extending perception rather than replacing the person who gives the movement meaning.

Inside the lab

What embodied AI can reveal

This is the kind of work we are beginning to do inside SEE Lab.

It can also compare movement across time, practitioners, traditions, environments, sound structures, or other conditions held in the corpus.

None of these findings tells the person what the movement means.

The system can produce measurements, relationships, comparisons, visualizations, and evidence. Meaning develops through the person, the practice, the culture, the history, and the relationships surrounding the movement.

The dancer should therefore leave the system knowing more about their movement, not having been told by a machine what their movement means.

Permission as infrastructure

Consent changes the technology

This is also why I do not think consent can remain a form someone signs before capture and then disappears from the technical system.

If embodied AI is going to work with intimate human knowledge, permission has to become part of the infrastructure. A movement record should carry information about who authorized it, what can be done with it, what cannot be done with it, whether it can be compared, whether it can train a model, whether it can circulate publicly, and whether it can later be withdrawn.

This is one place where I think we have an opportunity to build differently from the beginning.

The proving ground

Why dance is a difficult and useful proving ground

We have been developing this architecture through dance because dance is an unusually difficult form of embodied knowledge to simplify.

Movement may involve the body, sound, effort, geography, social relation, material objects, facial movement, cultural memory, and environmental conditions at the same time. A system that reads only skeletal geometry will capture something real, but it will also miss a great deal.

African and African diasporic movement traditions have been especially important to this work because many of them exposed the limits of the analytical tools I was using very quickly. Timing may be organized differently. Sound may function as instruction rather than accompaniment. Weight and gravity may be related to particular environmental conditions. Movement structures that appear visually similar may carry completely different functions.

Those difficulties forced us to build more carefully.

SEE Lab now works with several computational primitives for different registers of an embodied event: kinetic movement, sound, felt effort, geography, material culture, facial movement, and bodily trace. These can be studied separately and eventually brought together without assuming that one representation explains the whole body.

If the system can learn to work responsibly with that degree of complexity, then the methods developed through dance become useful for much wider forms of human movement.

A different product philosophy

What kind of product comes from this?

Much of the present AI economy is organized around producing something on a person’s behalf: writing the email, generating the image, summarizing the meeting, making the presentation, producing the video.

Embodied systems create another possibility. The product can help people understand, extend, or work with capacities that remain fundamentally theirs.

The category of intelligent systems can be larger than what we have so far built.

The frontier, and the future it points at

A different relationship with intelligence

What interests me most about this work is the possibility of changing the relationship between the person and the computational system.

The person does not become raw material for the machine. The machine does not become the authority over the person.

Instead, the system becomes an instrument through which aspects of embodied knowledge can be made visible, compared, preserved, and returned to the people working with them.

This is why SEE Lab begins with movement, but I do not think movement is where the implications end. Human beings know through bodies. We learn through repetition. We adapt to environments. We remember physically. We carry culture through gesture, posture, timing, sound, proximity, and relation. Much of that knowledge has remained difficult for computational systems to work with because our machines have been better at reading representations of human life than human embodied experience itself.

Embodied Computation and Intelligence is an attempt to work on that problem.

The technology is still developing. There are many things the system cannot yet read, many things it should never be allowed to infer, and many questions about governance, ownership, culture, and consent that will remain active for as long as the work continues.

But I think the starting point matters. If we begin with people, with bodies, with consent, with culture, and with the knowledge already present in human experience, we may build a very different kind of intelligence.

Not one whose success depends on doing more of our lives for us, but one that helps us understand more of what we are already capable of doing ourselves.