Max Hodak's retinal implant tests the brain-as-computer idea

The most consequential idea in Max Hodak's conversation with Sarah Guo is also the easiest to dismiss as philosophy: the brain is an information-processing system, so medicine may sometimes improve it more effectively by engineering interfaces than by trying to understand every molecule inside it. Science Corporation's PRIMA retinal implant gives that claim a concrete test. It does not restore a lost eye by rebuilding the eye. It sends a new visual signal into the remaining retinal circuitry and asks the brain to use it.

Hodak is the co-founder and CEO of Science Corporation. In the episode, he describes Science as a medical-device company whose first commercial product is PRIMA, then uses that product to explain a much wider program in neural interfaces, brain science, and human identity. 1

PRIMA combines a tiny photovoltaic implant under the retina with glasses that project near-infrared light onto it. The implant stimulates retinal cells that remain after severe central vision loss, while the glasses provide the incoming image, power, and data. Science describes the implant as a 2 mm by 2 mm subretinal device paired with specialized glasses, rather than as a replacement eye. 2

That arrangement matters because it defines the kind of claim Science is making. The company is not claiming that a chip has recreated normal sight. The company is routing a usable signal around damaged photoreceptors and letting the patient's existing visual system interpret it. Hodak says the first version produces a narrow field of view and black-and-white form vision. He describes the next engineering steps as adding grayscale and, potentially, some color. 1

The clinical results give the device more weight than a futuristic demonstration. Science reports that a multicenter trial followed 38 patients at 17 sites in five countries. The company says the patients gained an average of 25.5 ETDRS letters, 84% could read letters, numbers, and words, and 80% achieved at least a 0.2 logMAR improvement in prosthetic visual acuity after 12 months. The same report says the serious adverse events occurred mainly during the first two months and that 95% resolved within two months. 2

Those numbers establish a meaningful proof of function. They do not establish ordinary sight, a cheap treatment, or a general solution for blindness. Hodak's own description keeps the boundary visible: patients can read and recognize structured forms, but the image remains constrained. That limitation is part of the story because an engineered system can improve along identifiable dimensions. Field of view, resolution, grayscale, color, and the number of eligible retinal conditions can each become a separate development target.

Hodak's explanation of Science's path to PRIMA is less glamorous than the talk of brain uploading, and more useful. The company considered several routes into the visual system, including stimulation at different points and different physical mechanisms. Science eventually acquired a retinal-implant program developed by the French company Pixium, then spent about two years bringing it toward European commercialization, according to Hodak. 1

The retina won because the optic nerve already provides a route from the eye into the brain. A device placed there can work with existing biological wiring instead of requiring the company to decode and replace the entire visual cortex. Science's official account describes PRIMA as an artificial-photoreceptor array that stimulates remaining retinal cells and sends visual information to the brain. 2

Hodak contrasts that approach with the uncertainty of drug discovery. A drug program can spend years turning over molecular hypotheses before a clinical trial produces a decisive answer. A device program can also fail, but a working device exposes a more legible improvement path: increase the usable field, improve the signal, add color, reduce surgical burden, and expand the diseases it can treat. The comparison does not make engineering easy. The comparison explains why Science sees a tractable product loop where conventional biotech often sees a long biological search. 1

Science's business plan depends on that loop continuing. The company announced a $230 million Series C in March 2026 to fund PRIMA commercialization and additional clinical trials, and said it was expanding work toward Stargardt disease and retinitis pigmentosa. The same announcement identified Hodak as co-founder and CEO and said the company had submitted a European CE-mark application while pursuing U.S. regulatory approval. 3 In the interview, Hodak says European marketing approval arrived in July and that initial sales would begin in the following weeks. 1

The conversation becomes more speculative when Guo and Hodak move from restoring lost vision to changing the boundary of the human mind. Hodak separates ordinary communication with a device from adding a new structural capability. A system that turns an intended action into text still works like a communication channel. A system that supplies vision, hearing, balance, or motor control could become part of the user's practical body.

That distinction gives the BCI category a more useful shape than the usual promise of "mind control." Silent-speech tools, motor decoders, retinal implants, and systems aimed at richer sensory input solve different problems. Hodak argues that restoring a missing sense may be more consequential than making an existing action slightly more convenient, because the implant becomes part of the loop through which the brain experiences the world. 1

Hodak also treats continuity as central to identity. A software copy that answers like a person may preserve information about that person, but he doubts that the copy would feel like the same continuing self. A gradual change that preserves continuous experience could feel different from an abrupt replacement, even if the replacement reproduces the same memories and responses. The interview does not resolve that problem. It identifies a design constraint for any future technology that claims to preserve or extend a person.

The final bridge in the episode connects Science's neural work to AI models. Hodak points to the "platonic representation hypothesis": the idea that models trained on the world may converge on internal structures that resemble representations found in biological brains. He says Science has used alignments between animal neural recordings and model representations as a practical research clue, while acknowledging that the hypothesis remains controversial and incomplete. 1

This argument is stronger as a research method than as a theory of consciousness. AI models are inspectable in ways that living human brains are not. Researchers can vary the training data, measure internal representations, and test how a model changes when its inputs or objectives change. If some of those structures resemble neural patterns, models could provide a tractable environment for asking questions that are currently difficult to test in people.

The gap remains large. Similar geometry does not prove that a model experiences anything, or that a brain and a model implement the same computation in the same way. Hodak names connectomics—the mapping of neural connections—as one missing piece, and he says a detailed human connectome remains far away. 1

The episode's most defensible conclusion is therefore modest but important. PRIMA shows that a neural interface can restore a constrained human capability and improve through an engineering process. The larger vision—brain augmentation, substrate independence, and shared representations between brains and models—remains a program of research. The device earns the right to make that program interesting; it does not settle the philosophical questions on its own.