The flashiest version of neurotechnology usually begins with a dramatic promise: a thought turned into text, a brain signal converted into an image, a future where machines understand people before they speak. Yonathan “Yoni” Swechinsky has spent enough time with real brain data to be careful with that kind of excitement.
He is not cynical about the field. He is building in it. But his optimism has been shaped by systems that had to work under pressure, EEG signals that refused to behave cleanly, and years spent learning how much of neurotechnology depends on work the public rarely sees.
“The work that moves neurotechnology forward is often the work people do not know how to talk about,” Swechinsky says. “Cleaning signals, building pipelines, proving reliability. That is where the field becomes real.”
That view separates him from the more theatrical side of brain-computer interfaces. Swechinsky has worked in neurotechnology for six years, including five years at brain.space, the company formerly known as EEG-Sense. There, he helped build the software powering a state-of-the-art EEG headset and led the research team analyzing the resulting data.
The system also flew to the International Space Station, where astronauts recorded their brain activity using software he wrote. The point, for Swechinsky, is not the novelty of brain data in space. It is what that setting demanded from the system. In orbit, software cannot rely on the forgiving conditions of a lab. It has to survive the process of getting there. It has to work without a support team standing nearby. If it fails, the experiment may simply be over.
“That kind of environment removes a lot of fantasy,” Swechinsky says. “You learn what it means for something to be a real system and not just a promising demo.”
His skepticism is most visible when he talks about thought-to-text, one of the most attention-grabbing areas in neurotechnology. Research teams have explored ways to convert brain activity into words or images, using methods that range from EEG to fMRI. Swechinsky sees value in some of the underlying research, but he is unconvinced by the naive consumer framing around thought-to-text.
The reason is practical. Communication already has fast channels. People can speak. They can type. Silent-speech technologies may also provide efficient routes for input. If a brain interface produces less information per unit of time than those existing methods, the result may be scientifically impressive without being useful enough to become a product people rely on.
“Thought-to-text sounds exciting because the phrase feels futuristic,” he says. “But I always come back to value per unit of time. If a person can get more information out by speaking, the product case becomes much weaker.”
Thought-to-image is more interesting to him, partly because images are not generated in the same way sentences are spoken. He can imagine a system where a person guides an image as it is being created, with a model detecting subtle internal cues and adjusting the output. That kind of feedback loop could make the brain signal more useful because the interaction is iterative rather than purely extractive.
Still, even there, he is careful. The field has a habit of making single demonstrations sound like the arrival of a complete category. Swechinsky is more interested in what happens after the demo, when the system has to work across people, contexts, moods, movement, noise, and hardware variation.
“If you want to build products and not just papers, you have to engineer for the messy case from day one,” he says.
That lesson came directly from brain.space. The company’s goal was to create a large dataset of brain activity and train models on top of it to quantify mental states. Large-scale data collection appealed to Swechinsky because he believed measurement had to come before improvement. Yet collecting the data was only the first fight.
Brain signals are not tidy. Data collected from real people in real environments does not behave like idealized lab output. Before researchers can train models, they often have to solve layers of preprocessing and analysis that sound less glamorous than “mind reading” but decide whether the work is meaningful at all.
“Physics trained me to look for clean structure,” Swechinsky says. “EEG made me earn every useful signal.”
That is why he is supportive but cautious about EEG-based monitoring companies that hope to gather large amounts of data from many users with a very small number of electrodes. The long-term idea appeals to him. A system that can monitor stress, fatigue, or mental load could become useful in daily life. He can imagine an app that recognizes when someone is tired and tells them it is time to log off.
His concern is whether limited electrode coverage can reliably detect the signals these products claim to need. He would rather see stronger proof with broader coverage before assuming the smallest possible sensor setup can carry the field.
“Large-scale data collection is important,” he says. “But you have to prove the signal is there before you build the story around it.”
His opinions are not limited to consumer neurotechnology. Swechinsky is also interested in how neuroscience may shape the study of AI. As AI systems become more complex, he expects researchers to probe and analyze them in ways that resemble how people study the human brain. He also believes the structure of the brain may offer useful ideas for more efficient neural networks, especially when it comes to systems with multiple pathways handling different kinds of computation.
Even psychedelic research fits into his broader view of mental states, though he believes the topic is often sensationalized. He supports research into psychedelic substances and their potential to improve mental well-being, but he wants the field to model the phenomena accurately instead of leaning on spectacle.
That is the thread running through his work: the mind is powerful enough without exaggeration.
Swechinsky is now focused on non-invasive brain-computer interfaces aimed at modulating mental states rather than only reading them. He points to sleep as a likely starting point because it is measurable, universal, and tied to a problem nearly everyone understands, but he does not frame the work as a finished consumer promise. To him, the larger question is whether neurotechnology can move from monitoring the brain toward helping people influence their own internal states in a safer, more deliberate way.
That would still be a major shift, even without the language of science fiction. A tool that helps someone enter a useful state at the right moment may not sound as cinematic as reading a private thought. It may prove far more valuable.
For a field crowded with speculation, Swechinsky’s standard is useful. He is not asking neurotechnology to dream smaller. He is asking it to survive contact with reality. The future he cares about will depend on systems that can handle noise, models trained on data that has been cleaned and understood, and tools that make internal states less mysterious without pretending the brain is simple.
The standard he keeps returning to is not whether a demo sounds futuristic. It is whether the system survives variation, ordinary use, and the stubborn mess of real signals.
That is where Swechinsky believes neurotechnology becomes more than an announcement. It becomes something that works.