For decades, brain-computer interfaces have pursued the same objective.
Measure neural activity more accurately.
Build better sensors.
Collect cleaner signals.
Develop stronger decoding algorithms.
The assumption has been remarkably consistent: if we improve our ability to decode the brain, we will eventually build a natural interface between humans and machines.
I believe that assumption is incomplete.
Decoding is necessary.
It is not the destination.
Every Interface Is a Translation
The history of computing can be understood as a history of translation.
A keyboard translates movement into symbols.
A touchscreen translates gestures into commands.
A microphone translates sound into language.
Even today's brain-computer interfaces largely follow the same pattern. Neural activity is treated as another signal to decode—as if thought were simply another input modality waiting for a sufficiently capable algorithm.
This view has produced extraordinary progress. It has restored movement, enabled communication, and demonstrated that intention can be extracted directly from neural activity.
But it also frames the relationship between human and machine in a particular way.
The computer listens.
The brain speaks.
The interaction remains fundamentally one-directional.
A Limb Is Not a Tool
Consider your own hand.
You do not consciously translate neural signals before reaching for a cup.
Nor does your brain wait until movement is complete before receiving information from your hand.
The relationship is continuous.
Sensation and action occur together.
The body is constantly reading and writing.
A limb is not a device that is operated.
It has become part of the self.
This distinction is more than biology.
It suggests a different way to think about interfaces.
The most natural interface is not one that requires less effort to operate.
It is one that eventually stops feeling like an interface at all.
The Wrong Question
Much of today's BCI research asks: how accurately can we decode thought?
It is an important question.
It is also, I believe, the wrong long-term question.
The deeper question is this: what kind of relationship allows a computer to become part of human cognition rather than merely interpreting it?
These are not equivalent.
The first produces better decoders.
The second seeks a different form of communication.
One extracts information.
The other participates in it.
Decoding Is the Beginning
None of this diminishes the importance of decoding.
Without decoding, there is no interface.
Understanding neural activity is the first requirement for understanding how information is represented inside the brain.
Every improvement in decoding reveals something more fundamental than accuracy.
It tells us where information lives.
How it is organized.
Which circuits participate.
When they activate.
How they interact.
In this sense, a decoder is not merely a translator.
It is a mapmaker.
Every successful decoding system contributes to a more complete map of the computational organization of the brain.
That map matters long after the decoder itself is replaced.
Why I Work on EEG
People often ask why I chose EEG.
The answer is not because EEG is perfect.
It is because EEG forces us to understand.
Unlike invasive recordings, non-invasive signals provide limited spatial resolution and significant noise. Success cannot rely solely on better measurements. It requires discovering structure that persists despite those limitations.
The challenge is therefore not simply extracting stronger signals.
It is identifying which structure within the signal actually matters.
In my own work, this led to treating decoding not simply as a machine learning problem, but as a problem of discovering the functional organization of language itself.
The immediate objective was communication.
The deeper objective was understanding.
Representation Before Communication
A brain-computer interface cannot communicate with a system it does not understand.
Before information can flow naturally between biological and artificial intelligence, both must possess compatible representations.
This is why representation precedes interaction.
Before writing becomes possible, we must first understand where meaningful information exists, how it evolves through time, and what computational structure underlies it.
Decoding provides that knowledge.
It identifies the circuits.
The geometry.
The timing.
The organization.
Only then does the possibility of a richer interface emerge.
When the Interface Learns the Human
Modern interfaces ask humans to learn machines.
We memorize shortcuts.
Adapt to operating systems.
Learn gestures.
Configure workflows.
The burden of adaptation almost always belongs to the person.
I believe the future will reverse this relationship.
The interface should learn the human.
It should continuously model how an individual represents language, intention, and interaction.
Rather than asking the brain to conform to the machine, the machine should gradually become fluent in the brain.
The goal is not convenience.
The goal is integration.
The best interface eventually becomes invisible.
Not because it disappears physically, but because it no longer demands conscious effort.
Beyond Decoding
If decoding represents the first generation of brain-computer interfaces, the next generation will not simply decode more accurately.
It will communicate differently.
Reading and writing will no longer be separate stages.
Interaction will become continuous.
The interface will adapt alongside the person who uses it.
Over time, the distinction between user and tool begins to dissolve.
Not because the computer replaces the human.
But because communication becomes as natural as moving a hand or hearing a voice.
This is not simply a better decoder.
It is a different relationship between intelligence—biological and artificial.
Looking Forward
Every mature technology eventually outgrows the assumptions that created it.
The first automobiles resembled horse-drawn carriages.
Early computers resembled mechanical calculators.
Only later did entirely new forms emerge.
I believe brain-computer interfaces remain in that first stage.
We are still building better carriages.
Decoding will continue to improve. It should.
But the long-term opportunity is larger than decoding.
It is to understand how two fundamentally different forms of intelligence can participate in a shared loop of communication—one that neither merely controls nor merely observes, but continuously learns, adapts, and understands.
The destination is not a computer that reads thought.
It is a computer that the brain eventually accepts as part of itself.
When that happens, the interface will disappear.
Only communication will remain.