Communication in Latent Space Requires Constraints

I sometimes think that what we call willpower is not simply the power to resist an addiction. Maybe it is the ability to harmonize and coordinate planning–and–action.

Imagine that there is only a plan. We think, imagine, worry, and plan again inside the brain, but never act. We become trapped in an intention-action gap, something often seen in executive dysfunction: the plan exists, but initiating and organizing the action becomes difficult.

Now imagine the opposite: action without enough planning or stopping. We bite our nails, move repeatedly, eat too fast, speak too quickly, or reach for our phone before realizing that we have made a choice. Tics, compulsions, habits, and addiction are medically different phenomena, so they should not be explained by one circuit. Still, they make me wonder about the same broad question: what happens when planning and action are no longer well coordinated?

Dopamine-related circuits are often introduced as the circuits of reward and addiction. But dopamine does much more than make us want pleasure. It is involved in learning, motivation, vigor, and deciding when an action is worth taking. I wonder whether part of its larger role is to help coordinate the distance between a plan and an action.

The anterior midcingulate cortex, or aMCC, may also be relevant. Many neuroscientists and popular science writers like to connect this region with willpower. Calling it a “willpower center” is too simple, but it does participate in effort, conflict monitoring, and action selection as part of wider networks. Maybe what we experience as willpower is not one brain region becoming stronger. Maybe it is the plan-action system being coordinated precisely enough that we can act, adjust, or stop.

But what does this have to do with BCI?

Is the body an obstacle or a buffer?

As I mentioned in Philosophy of Brain-computer Interface, BCI is fundamentally an act of bypassing a neuronal pathway. This is exactly why it can be so valuable. When injury or disease blocks the usual pathway, BCI may reinforce, rehabilitate, restore, or enhance a biological function.

But what do you think the body is?

Elon Musk often says that the body has a biological bandwidth limit. We cannot speak or type as fast as we think. From this perspective, the mouth, hands, and muscles are obstacles between the brain and the world.

On the other hand, Jieyu Zheng and Markus Meister argue in Why do we live at 10 bits/s? that although our sensory systems receive enormous amounts of information, deliberate human behavior operates at around 10 bits per second. This is a theoretical estimate, not a literal measurement of the prefrontal cortex. Still, their argument points toward a strange bottleneck: the brain processes many things in parallel, but much of our deliberate behavior proceeds serially.

The body can therefore be viewed in two ways:

1. an obstacle;
2. a buffer.

When we speak too much, the mouth gets tired. When we write for too long, the hand hurts. More importantly, the slow process of turning thought into language forces us to gather our thoughts. We reduce their entropy, encode them, and create a form that another person can understand. Writing with a pencil is not simply a slow output method. The slowness itself can be part of the creative process.

Communication in latent space

Imagine that many chips and electrodes are implanted in the brain, allowing internal or covert speech to be decoded before it becomes spoken language. Current imagined-speech BCIs are still limited but do have a potential to transmit one person’s private latent space directly to another. And suppose one day they can.

Communication might no longer require the mouth, hand, or even a finished sentence. Rich internal representations could be decoded and transmitted with much higher bandwidth. That sounds efficient. But communication in latent space may require constraints precisely because a latent thought is not yet a deliberate expression.

The slow process of language does not only transmit information. It selects, compresses, and transforms it. If BCI skips this encoding process, it may also skip part of the plan-action process. We could communicate faster but become more reactive and automatic, releasing an impulse before it becomes something we have chosen to say.

Think about it. People addicted to their phones already watch TikTok and other digital media until their eyes are tired and hands hurt. Sometimes physical exhaustion is what finally makes them stop.

If the body that provides this buffer disappears, can you stop your brain?

There is one more asymmetry that concerns me. When the hand, voice, or muscle is injured, we at least have established forms of rehabilitation. When repeated BCI use changes a neuronal circuit through plasticity, our tools for detecting and reconnecting that circuit are still very limited. The brain can adapt to an interface, but adaptation is not automatically healthy simply because performance improves.

This is not why I oppose BCI. It is why I study closed-loop stimulation.

I see enormous potential for BCI to reinforce, rehabilitate, restore, and enhance biological function. Next-generation biohybrid interfaces will require closed-loop stimulation alogrithm and intelligent system that can treat the body more safely, help it recover, strengthen its abilities, and connect it with new tools. But such an interface should not simply decode more or stimulate faster. It should listen to the biological system–physiologically grounded, observe how it is changing, and respond to that change.

If we build the accelerator, we should also understand the brake.

Perhaps the body is a bandwidth limit. But it is also where planning and action are harmonized, where an impulse meets a constraint, and where a thought becomes something we choose to communicate.

Further reading

• Zheng and Meister, The unbearable slowness of being: Why do we live at 10 bits/s?, Neuron (2025).