Superhumans: The Dream of Opening the Mind

Imagine a young woman in Nairobi placing her hands on a piano she has never played.

Outside, a motorcycle climbs the road. Somewhere a kettle clicks off. The room contains all the ordinary sounds of a morning, and one extraordinary possibility: the distance between wanting to make music and knowing how might become smaller than any previous generation could have imagined.

She is wearing a non-invasive interface. An intelligent system is learning how she attends, anticipates, remembers, and corrects herself. Perhaps it helps her acquire the skill faster. In the most ambitious version of the dream, it helps establish capacities that would otherwise require years of practice. Her hands begin to move.

That scene is an imagined future. No research cited here demonstrates that a wearable can install piano mastery, surgical expertise, mathematics, or another person's memories. I begin with it because it expresses the destination I want to investigate: a world in which the boundaries of human ability are more generous.

I call that ambition Superhumans. The word should describe an expansion of opportunity, rather than a new hierarchy of human worth.

The prototype and the promise

We have an early internal prototype connected to this vision. That is a founder-reported claim, not an independently validated scientific result. I am not claiming that it has demonstrated skill downloading, produced AGI-level human intelligence, or established a clinical benefit. Its architecture, implementation, and other technical details remain private.

What I can share is the question: could advanced AI help us understand and influence the conditions under which the brain develops extraordinary abilities, using an interface that does not require surgery?

My strongest hypothesis goes further. I believe we should investigate a future in which people can acquire complex capabilities without the conventional years of learning, and an intimate partnership with AGI lets human beings operate at its effective intellectual level. That is the destination I am betting on, not an outcome I can promise. Neither proposition is established. Biological intelligence, a person assisted by software, and a machine's own reasoning are different things; parity would need a definition and evidence.

The ambition deserves a clear boundary precisely because I want it taken seriously. An early prototype is a beginning. A working theory is a beginning. The demonstration must still arrive.

The strange invitation of savant abilities

Accounts of acquired savant abilities are among the reasons this question refuses to leave me alone. Some people reportedly develop striking, narrow abilities after neurological change. The stories suggest that our ordinary picture of access to our own capacities may be incomplete.

Darold Treffert's review of savant syndrome distinguishes exceptional abilities associated with developmental conditions and other central nervous system conditions. These are unusual, heterogeneous phenomena. They do not establish that every person carries a complete, hidden library of expert skills waiting for an activation signal.

A 2021 case series on sudden savant syndrome describes eleven self-referred people reporting unexpected new abilities without an apparent precipitating injury. It is a case series using reports and surveys, not a randomized experiment demonstrating how to reproduce the outcome. Sudden and acquired savant accounts should also not be treated as interchangeable categories.

The inspiration I take is therefore a question, not a mechanism. What changes when a capacity becomes accessible? Attention? Inhibition? Motivation? Perception? The way existing knowledge is organized? The willingness to spend enormous amounts of time on one activity? Different cases may have different explanations.

We should investigate without romanticizing injury or disability. A person is more than the astonishing thing they can do, and an acquired ability can coexist with profound losses. The responsible dream is to understand useful mechanisms without reproducing the harm.

An extraordinary case is a doorway to a question. It is not yet a recipe for a capability.

We are learning to listen

Non-invasive interfaces already provide real reasons to believe that the conversation between brain and machine can improve.

The 2026 Nature Neuroscience Brain2Qwerty study decoded typed sentence production in healthy volunteers using EEG and MEG. Across the study's 35 participants, the reported mean character error rate was about 29% for MEG and 65% for EEG. Those figures describe a constrained experimental task. They do not mean arbitrary private thoughts can be read, nor that an ordinary consumer headset achieves the MEG result.

The Brain2Qwerty v2 project subsequently describes continuous MEG sentence decoding, reporting mean word accuracy of 61% and 78% for the best participant. Word accuracy and character error rate are different metrics. The authors themselves identify excessive errors and bulky equipment as obstacles before clinical transfer.

This is still an extraordinary scientific direction. Brain activity can contain task-related information that useful models can learn to recover. Better sensors, larger datasets, and improved decoders can change what is possible.

But listening and teaching are different verbs. A decoder estimates what a signal means. It does not thereby know how to cause a person to acquire the corresponding skill. Reading a musical score from a camera does not tell you how to build a pianist.

The harder direction: from estimate to intervention

Non-invasive stimulation is a separate research field with evidence that must be read task by task.

A 2026 sham-controlled laparoscopic-training trial involving 33 medical students reported a selective advantage for bimanual coordination when stimulation accompanied training. It did not find an independent benefit on the other two assessed task categories. The participants practiced; expertise was not downloaded.

Another 2026 study of interleaved premotor stimulation found no evidence of improved motor-sequence learning in its tested setting. A null result is part of the map. It helps prevent the field from turning every plausible neural story into a claim of enhancement.

Together these examples support a careful research agenda: investigate whether particular interventions improve particular learning outcomes for particular people. They do not support a universal claim that electrical or magnetic stimulation unlocks genius.

The far-future possibility remains open as a hypothesis. AGI might help identify more effective personalized interventions, or discover that some imagined shortcuts are impossible with a given interface. Either answer would be knowledge. A serious program must be able to learn the second answer as well as the first.

A little mathematics changes the dream

Think of the brain as an evolving system. Its internal state changes over time; sensors provide partial observations; an intervention changes some of the conditions under which the next state develops. The useful questions are what can be inferred, what can be influenced, and whether the change produces a lasting capability.

This is a conceptual model, not a complete mathematical description of a brain. It is valuable because it separates three problems that impressive demonstrations can blur.

The first is observability. Do the measurements contain enough information to distinguish the states we care about? EEG source reconstruction is an inverse problem: different internal configurations can produce similar scalp measurements. The EEG source-imaging literature explains why assumptions and constraints are needed to estimate sources. More powerful inference can make better use of data, but confidence produced by a model is not the same thing as information measured by a sensor.

The second is controllability. Even if a system estimates a relevant state, can its available intervention move the brain toward a useful alternative? A person may hear an orchestra clearly while possessing only one very blunt control over its performance. Research on brain modeling for control discusses why understanding the interaction between stimulation and neural dynamics matters for closed-loop design.

The third is system identification. People differ, and a person's brain changes with fatigue, attention, context, and learning. A useful controller needs a sufficiently accurate account of how this particular system responds. A population-average effect may be an inadequate guide for an individual.

I expect mathematical breakthroughs developed for advanced AI to spread into neuroscience: better inverse methods, representations, and tools for identifying and controlling complex systems. That transfer is part of my thesis, not a result our prototype has established. Advanced AI could contribute to all three problems: better decoding, better models of individual responses, and better experiment selection. That is a proposal for research. It does not remove the limits of the measurements or actuators.

A skill is more than a file

“Downloading” is an irresistible metaphor because it compresses the desire perfectly. We want the capability to arrive already usable. But a skill involves more than a sequence of facts. It can depend on perception, motor coordination, memory, judgment, and the body's interaction with its environment.

For that reason, three outcomes must remain distinct. One is assisted performance: a person does something impressive while an external system supplies substantial guidance. Another is accelerated learning: the person acquires an ability faster and retains it. The most radical is direct acquisition: the capability appears with little or none of the practice normally required.

All three could matter. A person who regains communication through an interface need not prove unaided mastery for the technology to be valuable. But an assisted result cannot be advertised as permanent skill acquisition without evidence that it remains after assistance is removed.

There is also an information problem. A low-bandwidth channel can provide a useful cue because the brain already has enormous structure and prior experience. That does not show the channel can transmit arbitrary expertise. If the claim is that an intervention unlocks existing capacity, investigators must test that explanation rather than merely assume it.

The analogy with AGI can mislead too. A person connected to an excellent research agent might complete sophisticated work through a shared workflow. That would be a powerful human-machine system. It would not automatically show that the person's biological brain now contains the agent's knowledge or reasoning capacity.

The distinctions give the dream sharper edges. They show us what to measure.

The experiment worthy of the ambition

I would want a research program that declares its target before seeing the result. Define the skill, baseline ability, comparison condition, assistance level, and meaningful benefit. Use independent assessment and appropriate blinding where feasible. Measure retention, transfer to unfamiliar tasks, and what happens when the interface is absent.

Separate genuine improvement from practice effects, expectation, selective reporting, and hidden software assistance. A narrow gain can be important, but it should retain its narrow name. Reproduce it across people and independent laboratories before making broader claims.

These are research principles, not instructions for self-experimentation or a stimulation protocol. Human studies require qualified clinical and research oversight. A program whose central promise is human flourishing must treat participants as people whose welfare matters beyond the result.

I would also want ordinary failure to remain publishable. If a method improves performance while connected but produces no retained learning, that is an informative finding. If it works for one task and fails for another, the difference may reveal more than a sweeping success story.

Our private prototype does not substitute for this process. Independent validation is what could eventually turn a founder's conviction into a contribution other people can rely on.

Why I want this work rooted in Africa

The deepest appeal is not a competition to produce the most enhanced individual. It is the possibility of changing access to capability.

Imagine better rehabilitation for someone whose voice has been interrupted. Imagine a more effective training pathway for a technician maintaining a rural energy system. Imagine a scientist who can collaborate with powerful intelligence without surrendering authorship or agency. These possibilities vary enormously in difficulty; they are a set of aspirations, not a list of promised products.

A Nairobi program should build local scientific capacity alongside any device. Neuroscience, engineering, mathematics, clinical research, and the humanities belong in the same conversation. So do people who might use the interface. They should help define success rather than arrive only when a demonstration needs an audience.

If enhancement ever becomes powerful, access will shape its social meaning. Systems priced beyond most people's reach could deepen inequality. Employers could pressure people to use them. Neural data could become another resource extracted without meaningful control. Those outcomes are design and institutional questions we can address before the capability becomes commonplace.

I want an ambition large enough to include the person who cannot pay, the person who refuses an interface, and the person whose worth has never depended on exceptional performance.

The hands on the keys

Return to the imagined room. The woman at the piano is still there. We have not established that her morning is technologically possible. We have established something more useful than a premature claim: a set of questions through which we could investigate it.

What is already present? What can be measured? What can be changed? What persists? What transfers? What remains hers when the machine is switched off?

Those questions do not drain the dream of wonder. They give wonder somewhere to go.

I believe non-invasive brain interfaces and advanced AI could open a remarkable new chapter of human capability. I want us to pursue that chapter with the courage to imagine more, the honesty to name uncertainty, and the patience to find out.

Perhaps the first great achievement will be a voice restored. Perhaps it will be a lesson made easier. Perhaps, much later, it will be a form of learning for which we do not yet have an adequate word.

Whatever arrives, I want us to remember why we began: so that more people can bring what is inside them into the world.

Editorial note: This essay expresses the founder’s ideas, developed with AI-assisted research and writing and reviewed against the linked sources on 29 October 2026. Convictions, proposals, and forecasts are identified in the text; scenes framed as imagined are illustrative.

Cover photograph / illustration: Hans Braxmeier. CC0. Original source. Illustrative image: Piano keys by Hans Braxmeier.jpg. Credit: Hans Braxmeier; CC0. This image does not document the research or proposed facility described.