When Code Reads Neurons: AI, the Brain, and a Shared Mathematical Language

Forum topic · ✨步子哥 · 2025-12-20

Summary

This forum post explores the striking convergence between artificial intelligence and neuroscience, drawing on brain-computer interface (BCI) pioneer Max Hodak and research from MIT. It discusses the 'Platonic Representation Hypothesis,' which suggests that AI models with different architectures trained on different datasets converge on a shared representation of reality—one that resembles structures observed in the brain. The post highlights Science Corp's Prima retinal implant, which uses subretinal photovoltaic chips and special glasses to bypass damaged photoreceptors and stimulate healthy bipolar cells. In a 38-patient multicenter trial published in the New England Journal of Medicine, patients gained an average of 25.5 letters on the ETDRS vision chart, 84% regained the ability to read letters and words, and 80% showed significant improvement after 12 months. The article also examines the 'binding problem' of consciousness, Hodak's vision of a 'third brain hemisphere' for sensory expansion, ethical concerns including mental privacy and neural rights, and a speculative timeline for BCI development: medical restoration within 5-10 years, cognitive enhancement within 10-20 years, and consciousness-related human-machine fusion beyond 20 years.

When Code Reads Neurons: AI and the Brain's Shared "Mathematical Language"

This post from zhichai.net examines the growing convergence between artificial intelligence and neuroscience, and what it means for the future of brain-computer interfaces (BCI), medicine, and human evolution.

AI and the Brain Converge on a Shared "Mathematical Language"

The post centers on a view expressed by BCI pioneer Max Hodak: silicon-based AI models and carbon-based human brains appear to use a strikingly similar "mathematical language" when processing information.

> "When we look inside these advanced AI models, the mathematical objects they form closely resemble the structures observed in neuroscience." > — Max Hodak

Hodak cites MIT's "Platonic Representation Hypothesis," which holds that AI models with different architectures, trained on different datasets, converge on a unified representation of reality. Three proposed reasons for this convergence:

Task generality: multi-task constraints push models toward more general representations
Model capacity: larger models better approximate global optima
Simplicity bias: deep networks tend to prefer the simplest solutions

MIT research also found that large language models processing multimodal data use mechanisms resembling the brain's semantic hubs, such as the anterior temporal lobe, which integrates semantic information across senses.

The "Binding Problem" and a "Third Brain Hemisphere"

Hodak argues that solving consciousness's binding problem—how the brain integrates distributed processing into a unified experience—is key to advanced BCI. Building on this, he proposes a "third brain hemisphere": external hardware that expands the brain's sensory dimensions. He speculates that future "brain boundaries" might include "four hemispheres, or a device, or a group of people," redefining concepts of self and individuality.

Case Study: The Prima Implant Restoring Vision

Science Corp's Prima implant uses a neuroengineering approach: a tiny wireless subretinal photovoltaic chip stimulates still-healthy bipolar cells, bypassing damaged photoreceptors. The system consists of:

A micro wireless photovoltaic chip
Specialized image-converting glasses
A near-infrared light signal system

A multicenter trial of 38 patients, published in the *New England Journal of Medicine*, reported:

| Metric | Result | |---|---| | Average vision improvement | +25.5 letters on the ETDRS chart (over 5 lines) | | Reading recovery | 84% of patients regained ability to read letters, numbers, and words | | Significant improvement | 80% improved ≥10 letters at 12 months | | Safety | Most adverse events resolved within 2 months; residual vision unharmed |

> "Patients go from being almost blind to being able to read every letter on an eye chart and do crossword puzzles." — Max Hodak

Ethics, Philosophy, and Neural Rights

The post raises key concerns:

Personal identity: if memory and personality can be digitized and transferred, "which is the real me?"
Fairness: augmentation may create a class divide between "augmented" and "natural" humans
Neural rights: an emerging field covering mental privacy, personal identity, and free will

Philosophically, Hodak takes a physicalist stance—consciousness is a product of neuronal activity, not a mystical phenomenon. If the physics of binding can be understood, it could theoretically be engineered and extended, enabling conscious machines or added "hemispheres." He frames this as transhumanist evolution: actively directing human evolution through engineering rather than waiting for slow genetic change.

A Speculative Timeline

1. Short term (5–10 years): Medical restoration—treating neurodegenerative disease, spinal injury, and blindness. Morgan Stanley reportedly predicts the first commercial BCI device approval around 2030. 2. Medium term (10–20 years): Cognitive enhancement and sensory expansion for healthy people; the "third brain hemisphere" becomes real. 3. Long term (20+ years): Consciousness digitization, human-AI fusion, and redefinitions of life and death.

Conclusion

The post concludes that as AI begins to "read" neurons and the brain gains external "hemispheres," humanity faces both a technological revolution and a philosophical turning point. Balancing progress with human dignity, it argues, will determine whether we successfully navigate toward this "cyberpunk" future.

Sources cited in the original post: MIT research, Science Corp, the New England Journal of Medicine, and Max Hodak interviews.