Restoring Sensory Pathways Is the Central Goal of Neural Interfaces
Science.xyz co-founder and CEO Max Hodak argues that neural interfaces should be judged less by the BCI label than by whether they can restore the sensory and motor pathways through which the brain receives and acts on the world. He presents Science’s PRIMA retinal implant, which has European marketing approval, as an early proof: it bypasses damaged photoreceptors to give some blind patients form vision rather than ordinary sight. Hodak’s broader case is that treating the brain as a computational system could lead to devices that repair lost capabilities while offering a new way to study intelligence.
PRIMA has European approval, but its vision is still only a beginning
Max Hodak says Science’s PRIMA retinal prosthesis received European marketing approval in July, roughly two years after Science acquired the underlying technology. He describes that as the point at which the device became commercially available, with initial sales expected in the following weeks.
The commercial milestone does not mean that the system reproduces ordinary sight. PRIMA can restore what Hodak calls a form-vision image for people who lost vision after the eye’s light-sensitive cells died. The engineering agenda now, he says, is to deepen grayscale and work toward red and green perception; blue is more difficult.
PRIMA is intended for people whose blindness results from the loss of rods and cones, including patients with macular degeneration, the condition in which Science conducted its clinical trial. Hodak says the company is also preparing studies in retinitis pigmentosa, Stargardt disease, and other conditions.
The system combines an implanted chip with external glasses. A tiny chip sits beneath the retina at the back of the eye. The patient wears glasses equipped with a laser projector, which sends an image to the implant. The implant stimulates the retina directly, bypassing damaged light-sensitive cells and sending a visual signal onward to the brain.
Science shows the kind of image processing involved through a demonstration of hands playing dominoes. One panel is labeled “Camera view”; the other, “Output processing,” is set to “Edge mode” and reduces the scene to outlines. The screen explicitly notes that the demonstration is for illustrative purposes and that the actors are not visually impaired.
Hodak compares the ambition to a cochlear implant for the eye. Cochlear implants, he says, have produced some of medicine’s largest effects, and Science wants to build interventions with a similarly consequential impact.
The retinal prosthesis right now, I think, is a great proof of concept that we're on the right track.
Hodak characterizes the device as the first instance, in his view, of restoring a form-vision image “in the mind’s eye” of a blind patient through this approach. He does not present that achievement as the finished product. The path forward is iterative signal engineering: improve the image, add more grayscale depth, pursue color, and compound those gains into a better device.
The retina was the practical route into a larger neural-interface project
Max Hodak describes Science as fundamentally a medical-device company, while framing its mission more broadly: using a differentiated understanding of the universe to improve the human condition. In practice, he says, that means working with the brain in ways that can create “big effect sizes” that medicine often does not achieve.
The company did not begin solely as a retinal-prosthesis effort. One major direction was a biohybrid neural interface: rather than placing metal wires into the brain or genetically modifying it, Science would engraft living neurons that grow in and establish biological connections. Hodak calls that a substantial and exciting research program, but says it needed to be paired with a nearer-term business.
In early 2021, Science asked what valuable problem it could realistically solve with its available resources and the state of the field. Hodak says the company concluded that restoring vision to blind people was the strongest candidate.
That decision required choosing where in the visual system to intervene. The retina is where vision is first created, Hodak says. Farther downstream, the optic nerve reaches the lateral geniculate nucleus in the thalamus, and from there connects to visual cortex. Primary visual cortex, or V1, contains roughly half a billion cells at the back of the brain in his description.
The potential access points were therefore the retina, the thalamus, or V1. But if the optic nerve remains available, Hodak says scientific and technical considerations favor the retina. The company then examined multiple ways to stimulate or modify it: an in-house retinal gene therapy, electrical stimulation, and ultrasound.
Science developed a gene therapy that Hodak expected to bring into human studies the following year. In parallel, it surveyed electrical retinal-stimulation work and identified what he considered the field’s leading technology: work developed by a Stanford inventor and licensed to Pixium, a small French company conducting clinical trials. Science developed a relationship with Pixium over several years and eventually acquired it. Hodak says the company saw value in the technology that others had not fully recognized.
The choice of the retina, in this account, was not a retreat from more ambitious neural interfaces. It was a point at which Science believed it could reach the brain through an existing sensory pathway while building a commercial product and continuing longer-horizon work.
For Hodak, the central object is moving signals into and out of the brain
Max Hodak grounds the retinal project in an unusually direct premise about biology and computation.
The brain very literally, very clearly, plainly is a computer.
His point is that computational problems can be solved by arranging matter in a particular way and then allowing that physical arrangement to operate. The brain is one such arrangement. The skull, he says, is effectively the familiar thought experiment of a “brain in a vat”: the brain is connected to the outside world through a limited set of physical cables, principally the cranial and spinal nerves.
Those pathways carry vision, hearing, balance, and motor signals. For Hodak, gaining the ability to move such signals into and out of the brain is not simply a means toward an individual product. It is the central technical objective.
If you can get the visual signal, auditory signal, balance, motor, in and out of the brain, that is an end in itself.
PRIMA illustrates the structure of that idea. It does not regenerate dead rods and cones. Instead, the glasses, laser projector, implant, and retina create a route for visual information to reach the brain.
Hodak’s broader interest includes sensory and motor pathways as well as what he calls substrate independence. He presents these as related questions: whether the signals through which the brain receives the world and acts on it can be reliably interfaced with, restored, or generated through new hardware. His focus is not the BCI label itself, but the functional consequences of access to those pathways.
A brain keyboard may be useful without revealing a hidden stream of thought
Sarah Guo describes a common premise in the BCI market: an invasive or noninvasive interface could enable high-bandwidth communication with an AI model. Max Hodak does not rule out the value of such a product. But he questions whether it rests on the right model of cognition.
Speaking and writing, he argues, are themselves forms of thinking. The common sensation that an idea is already complete inside one’s mind can be misleading; when a person tries to write or articulate it, the idea may turn out not to be fully formed after all. He therefore doubts that there is necessarily a rich, preassembled stream of thought waiting to be extracted at a much higher rate than language permits.
He points to what he calls a famous cognitive bottleneck of roughly 10 bits per second. Different lines of evidence, he says, converge on that order of magnitude. One example is a person with perfect memory taking a helicopter ride over Manhattan and later attempting to draw what they saw. Even with that memory, the detailed experience can be rendered only at about that rate across an hour or two, in Hodak’s account.
Hodak treats that as evidence of a deeply evolved processing constraint that also carries through language. If so, reading brain activity may not disclose a far faster channel of already-prepared thought.
That does not make a brain-controlled AI interface impossible or necessarily uninteresting. He can imagine walking down the street wearing a cap and asking an AI questions through internal monologue. Some combination of EEG and MEG might eventually make that possible, he suggests.
But Hodak considers such a “brain keyboard” a different product category from generating vision, generating hearing, or pursuing substrate independence. It could be valuable while still encountering a constraint familiar from augmented-reality glasses: human attention may already be fully occupied. Adding another interface does not automatically create additional usable cognitive capacity.
Both a brain keyboard and a retinal prosthesis may qualify as BCI products, but Hodak expects them to demand different technologies, serve different needs, and be built by different kinds of companies.
AI models may offer a way to compare representations across systems
Max Hodak sees AI not only as something a person might communicate with through an interface, but as a potentially useful lens for neuroscience. Sarah Guo describes his view as a reason for growing interest in BCIs: internal representations in AI systems and biological brains may be alignable.
Hodak links the idea to the Platonic representation hypothesis. He calls it controversial, but says something real is happening. When researchers inspect large AI models, the mathematical objects involved in representing concepts look, to him, like structures observed in neuroscience. The geometry associated with concepts in models and in the brain appears similar.
He acknowledges substantial uncertainty about what that similarity means. It is not clear whether the relevant structure is global or local, he says. Researchers may recover relational structures between ideas without yet knowing how more disconnected ideas are positioned relative to one another. Those open questions leave room to doubt whether the apparent alignments reveal a fundamental property of intelligence.
Hodak’s own view is that they do. He describes both brains and AI systems as potentially learning some true underlying structure in data, though he does not claim the phenomenon is fully understood.
The operational point is narrower and more concrete: Science uses the relationship constructively. Hodak says the company can align neural recordings from animal brains with internal representations from AI models. For him, that practical alignment is evidence that the comparison is more than an analogy—and a usable tool for investigating how intelligent systems represent information.