AI can now generate movement fast enough to respond within the blink of an eye. That number turns out to be exactly the threshold where a lagging tool becomes a present partner — and where a deeper question begins.

Try a small experiment in imagination. You reach out to shake someone's hand, and they meet your hand at exactly the moment yours arrives. Now imagine the same handshake, but their hand moves to meet yours a third of a second late. The first is a greeting between two people. The second is faintly unsettling — a beat off, a fraction behind, somehow not quite with you.

That gap — the difference between meeting-together and meeting-late — comes down to a remarkably small amount of time. And this month, AI movement generation crossed it. A system from NVIDIA called ARDY can now generate movement with a delay of just 33 milliseconds — about a third of the time it takes to blink. Understanding why that specific number matters reveals something surprising about how human beings experience time, togetherness, and the difference between a tool and a partner.

The Window of "Now"

Here is a fact about your nervous system that you have never noticed, because noticing it is precisely what it prevents. Your experience of "now" is not an instant. It is a window — a small stretch of time, a few tens of milliseconds wide, within which your brain bundles everything that happens into a single felt moment.

Events that fall within this window are experienced as happening together, simultaneously, now. Events separated by more than the window begin to be experienced as one-after-the-other, in sequence. The window is the grain of your temporal experience — the resolution at which your sense of "the present moment" is constructed.

This is why the handshake matters. When someone meets your hand within that window, your nervous system binds the two movements — yours and theirs — into a single shared moment. You experience it as together. When they meet it outside the window, the two movements fall into separate moments, and you experience their movement as coming after yours. Same handshake; completely different felt reality, decided by a few tens of milliseconds.

Why 33 Milliseconds Is the Magic Number

Now bring this back to machines. When an AI system responds to your movement, there is always a delay — the time it takes to sense what you did and generate its response. Where that delay falls, relative to your nervous system's window of "now," decides everything about how the interaction feels.

If the delay is long — say, a fifth of a second or more — the AI's response lands clearly after your movement, in a separate moment. However accurate the response, it feels like a lagging reflection. You move; then it responds. It is a mirror with a delay, and a delayed mirror can never feel like a companion.

But if the delay drops below the window — into the range of a few tens of milliseconds — something changes in kind. The response now falls inside the same moment as your movement. Your nervous system binds them together. The machine's movement is experienced not as coming after yours but as happening with yours. It crosses from reflection to presence.

This is what ARDY's 33 milliseconds achieves. It is not merely "fast." It is fast enough to fall inside the window where human experience stops registering a delay and starts registering togetherness. The number matters because it is a threshold in human experience, not just an engineering benchmark. Below it, a generated movement can — for the first time — feel present rather than delayed.

The Catch: Fast Is Not the Same as Anticipatory

So has the problem of the responsive machine partner been solved? Not quite — and the reason takes us somewhere deep.

Think again about a truly skilled movement partner — a dance partner, a Contact Improvisation partner, someone you have moved with for years. What is remarkable about them is not just that they respond quickly. It is that they seem to respond before you move — to be already meeting you as your movement begins, not after it arrives. They don't catch up to you fast. They arrive with you, as if they felt your movement coming.

And in a sense, they did. An experienced partner reads the tiny preparations that precede your visible movement — the subtle shift of weight, the gathering, the almost-invisible organising of your body that happens in the moment before you actually move. They are responding not to your movement but to your movement forming. That is why it feels like anticipation rather than reaction: they are meeting the movement as it comes into being, not after it has arrived.

Here is the subtle point. A machine that responds to your visible movement in 33 milliseconds is still, technically, reacting — reacting very fast, inside the window, but reacting to a movement that has already begun. It is a superb fast mirror. What it is not, yet, is anticipatory. It meets you inside the moment, but it does not meet you as you form — because it is watching what you do, not sensing what you are about to do.

Where This Points

This is exactly where the sensing technologies this series has followed come back in. Signals like EMG — which read muscle activity, including the preparatory activation that fires 50 to 200 milliseconds before visible movement — are windows onto the movement as it forms, before it appears. A machine that could sense your forming movement, rather than only your visible one, could respond not just fast but early — meeting you as an anticipating partner does, arriving with you rather than catching up.

So the 33-millisecond breakthrough is best understood as the completion of one layer, not the whole building. It solves the problem of delay: a response can now fall inside the felt moment. What it does not yet solve is the problem of anticipation: responding to the movement as it forms rather than after it appears. The first is about speed, and speed has now arrived. The second is about what the machine senses — visible movement, or forming movement — and that is where somatic knowledge, and the sensing that reads the forming body, still has everything to offer.

Thirty-three milliseconds gets a machine into the same moment as you. Feeling you about to move is what would let it truly move with you. The first is a triumph of speed. The second is a question of what we let the machine feel — and it is, still, the open one.

Continue through the archive

For connected context, read This Week in Motion AI: Real-Time Crosses the Threshold and Community Digest: Week of 7–13 July 2026.

References

Zhao, K., et al. (2026). ARDY: Autoregressive diffusion with hybrid representation for interactive human motion generation. arXiv:2607.08741. https://arxiv.org/abs/2607.08741

Wittmann, M. (2011). Moments in time. Frontiers in Integrative Neuroscience, 5, 66. https://doi.org/10.3389/fnint.2011.00066