A loop, not a one-way street
Picture how you reach for a cup of tea. Your hand moves, your eyes watch it, and you make tiny corrections the whole way there. You are not firing off one frozen command — you are constantly checking and adjusting. A brain–computer interface (BCI) works the same way. It is best understood not as a single push of a button, but as a loop that runs over and over, many times a second.
The loop has four steps that flow into each other: sense the brain, decode the intent, act in the world, and feed back the result so the brain can try again. Because the last step loops right back to the first, we call the whole thing a closed loop. The user is part of the circuit, not just a starting point.
Sense
First, the loop has to listen. Your neurons produce tiny electrical signals as they work, and sensors pick those signals up. The gentlest, most common option is an EEG (electroencephalography) cap — a stretchy cap dotted with electrodes that sit on your scalp, a bit like soft buttons resting on your head. Nothing goes inside; it simply listens from the outside.
When clearer, stronger signals are needed — usually for medical reasons — electrodes can instead be placed inside the skull, right on or in the brain. These implanted sensors hear individual neurons far more crisply, but require surgery. For now, just hold one idea: sensing means turning the brain's faint electricity into numbers a computer can read.
Decode
Raw brain signals are messy — a tangle of overlapping ripples, more like the roar of a stadium crowd than a single clear voice. Decoding is the step that listens to that roar and makes a confident guess about what you intend: "they want to move left," or "they are picturing their right hand." It turns a wave of electricity into a meaning.
A decoder is not born knowing your brain. It has to be trained: you try a few example actions while the system watches, and it learns which patterns mean what — a little like a friend gradually learning to read your facial expressions. Once trained, the decoder can run live, guessing your intent moment by moment. That continuous, instant version is called real-time decoding, and it is what lets the next step happen without a noticeable wait.
Act, and feed back
Now the guess becomes something real. The device acts on the decoded intent: a cursor glides across a screen, a robotic arm reaches out, or a wheelchair turns. Crucially, you can see it happen. That sight closes the loop — your eyes report the result straight back to your brain, which gets ready to nudge the next attempt.
This is why speed matters so much. Think of steering a car: if the wheel responded a full second after you turned it, you would weave all over the road. A BCI is the same — when the gap between intending and seeing (the latency) is short, your corrections land smoothly and control feels natural. When it is long, every move turns into an awkward guessing game.
When that feedback is deliberately turned into a learning tool — showing you your own brain activity so you can train it up or calm it down — it has a name of its own: neurofeedback. It is the same loop, used on purpose to help your brain and the machine grow better at understanding each other.