Lesson 9 — What Comes Next?
How Does the Brain Actually Work?
Learning Material
1 pagesLesson 9 — What Comes Next?
Understanding the Complex: How Does the Brain Actually Work?
The brain is the hardest problem that science has ever taken seriously. And yet, by almost any measure, progress over the past decade has been faster than in the preceding century. New tools have given researchers access to the brain at scales — from single molecules to whole-organ activity — that weren't imaginable twenty years ago.
Here's what the near and medium-term future looks like, organized by what's already happening, what's in clinical trials, and what remains genuinely speculative.
Brain organoids: brains in a dish
In 2013, Madeline Lancaster at the Institute of Molecular Biotechnology in Vienna grew the first cerebral organoid: a walnut-sized cluster of stem cells that spontaneously organized into layered structures resembling the developing human cortex. It was not a brain. It had no sensory input, no blood supply, no motor output, and no evidence of anything like experience. But it had neurons that fired action potentials and formed functional synapses.
Organoids have since become a standard model system for studying brain development, testing drugs, and modeling neurological diseases in human tissue rather than rodent models. Researchers have created organoids that replicate early development of the visual cortex, the hippocampus, and regions implicated in autism and schizophrenia.
The ethical questions are early but real. If an organoid developed enough complexity to process information in sophisticated ways, would that matter morally? Current organoids are far below any threshold that neuroscientists consider neurologically significant — but as the technology advances, the question won't stay hypothetical.
Optogenetics as medicine
Optogenetics — the same technology Karl Deisseroth developed to switch on memories in mice — is beginning to move into clinical applications. In 2021, a team led by José-Alain Sahel published the first evidence that optogenetics could restore partial vision in a patient blinded by retinitis pigmentosa. A gene encoding a light-sensitive protein was injected into retinal cells; the patient subsequently detected light and perceive some visual information. It was a modest but real recovery.
More ambitious applications are in development. Optogenetic therapies are in trials for several forms of blindness, and researchers are exploring whether the technology could be used to treat Parkinson's disease, epilepsy, and certain forms of depression — conditions where specific circuit malfunction is well understood. The challenges are significant: getting viral vectors into the right neurons, maintaining expression over time, and delivering light to deep brain structures. But the proof of concept in retina has shifted the conversation.
BCIs: from restoration to augmentation
The trajectory of brain-computer interfaces in the next decade is almost certainly toward less invasive, more capable devices.
Current high-performance BCIs require surgical implantation. But research into high-density EEG and fNIRS (functional near-infrared spectroscopy) is advancing rapidly, and companies including Kernel and Neurosity are building non-invasive systems that can detect meaningful neural signals without surgery. The resolution is currently far lower than implanted arrays, but the engineering gap is narrowing.
For patients with paralysis, locked-in syndrome, or ALS, the near-term BCI future involves better speech decoding — the ability to reconstruct intended speech from motor cortex activity in patients who can no longer speak. A 2023 study at UCSF demonstrated decoding of 78 words per minute with 25% error rate from a patient with ALS — still imperfect, but sufficient for communication. Progress is rapid.
The longer-term augmentation scenario — implants that enhance cognition in neurologically healthy people — remains far off, and its benefits are scientifically uncertain. There's no clear evidence that artificially increasing neural activity in any given region produces better cognition overall; the brain is a tightly balanced system, and pushing one variable tends to displace others.
The AI-neuroscience frontier
The most intellectually exciting intersection is between AI research and neuroscience. The two fields are increasingly learning from each other.
Neuroscience continues to provide architectural inspiration for AI: attention mechanisms (loosely analogous to selective neural gating), predictive coding (the idea that the brain generates predictions and only processes surprises — which has influenced several architectures), and memory-augmented networks (inspired by hippocampal memory consolidation).
AI, in turn, is becoming a tool for neuroscience. Large language models can be used as "neural language models" to probe which aspects of human language processing are captured by statistical patterns in text. Deep learning systems trained on visual recognition tasks produce intermediate representations that correlate with neural activity in the primate visual cortex — which tells neuroscientists something about what the cortex is computing.
This cross-pollination is still in its early stages. But the prospect of using AI to help decode the brain — and using the brain to improve AI — is one of the most generative research programs of the next decade.
What remains unknown
The most fundamental question in neuroscience — how physical brain activity gives rise to subjective experience — remains unresolved. The "hard problem of consciousness" (the philosopher David Chalmers's term) hasn't become easier. We can correlate neural activity with reported experiences; we cannot explain why there is experience at all.
This isn't a failure of neuroscience. It may be a problem that requires new conceptual frameworks not yet invented. That possibility — that the hard problem requires not just more data but different ideas — is what makes neuroscience both humbling and endlessly compelling.
Next lesson: What If...? — Three thought experiments about where the brain-machine interface might lead.
Reading time: approx. 10–11 minutes