The Brain’s Best Guess
Three years after losing my sight, I started seeing things—people, bikes, plants—that weren’t there. The weirdest part? They always appeared in context: a phantom keyboard materialized whenever I sat down to type, and a crowd of white-cane learners popped up during my orientation training. At first I feared madness, but the truth turned out to be stranger—and far more revealing—than any psychiatric diagnosis. My brain wasn’t failing me; it was doing exactly what it’s wired to do: predict the world, even when the evidence tells it to stop.
The Quiet Collapse of Input
Charles Bonnet Syndrome isn’t a disease. It’s a cascade. The moment light stops hitting the retina—whether from macular degeneration, glaucoma, or retinal detachment—the visual system gets cut off mid-thought. No signal reaches the primary visual cortex, but the higher layers keep firing, trying to finish what was started. It’s like a concert where half the musicians have walked offstage, yet the conductor still raises the baton and expects a symphony.
The medical term for this cutoff is deafferentation. But “cut off” undersells it. What really happens is more like a feedback loop spinning out of control: the brain expects input that never arrives, detects the mismatch, and then doubles down on its own predictions. In a normal eye, your visual system compares incoming light with what it anticipates and corrects the difference in real time. In CBS, the correction signal vanishes, and what remains is pure prediction—unmoored from reality.
Clinically, CBS shows up as complex, formed hallucinations: faces with expression and texture, animals in mid-stride, buildings with windows and chimneys. Crucially, patients retain insight—they know the images aren’t real. That’s what separates CBS from psychosis. You’re not delusional; you’re hallucinating with your eyes wide open and a clear head.
Top-Down Runs the Show
Here’s where predictive processing rewires our intuition. The brain doesn’t passively record the world; it actively constructs it, using Bayesian inference to weigh prior experience against current data. As philosopher Andy Clark puts it in The Experience Machine, “We perceive what would need to be present for our sensations to make sense.”
When I lifted my hands toward the keyboard, my brain didn’t wait for photons from the keys. It invoked the entire motor program: finger curvature, tactile landmarks, the clack of keystrokes—then filled in the visual part last. Since I’d typed thousands of times with sight, the prediction was effortless and automatic. The visual cortex didn’t hallucinate a keyboard; it hallucinated my keyboard—the one with the scratched F key and the slightly sticky Enter. That specificity is impossible to explain with hyperexcitability alone.
Functional MRI work by Ffytche and colleagues in 1998 confirmed this hierarchy. When participants reported faces, face-selective regions lit up; when they saw patterns, texture-processing zones fired first. The hallucination wasn’t raw noise—it was semantic content riding on spontaneous activity.
Why the Keyboard Disappeared
The hallucinations didn’t last forever. After about two years, they faded—some receding gradually, others vanishing overnight. That’s where neural plasticity enters the story.
The visual cortex isn’t static. In response to diminished input, it begins to reorganize: neurons prune unused synapses, rewiring for efficiency. Over time, the hyperexcitable network stabilizes. The internal predictions settle into a new baseline that matches the current sensory reality: no keyboard, because my brain learned to trust touch and sound rather than vision.
Marschall and colleagues (2020) describe this as cortical adaptation to deafferentation. The brain doesn’t “heal” the retina, but it compensates by recalibrating its generative model of vision. The hallucinations subside not because the brain gives up, but because it succeeds—finally updating its expectations to fit the world as it now is.
The Power of Prediction, Not Paralysis
This reframing matters—not just for people with CBS, but for anyone who’s ever misread a fallen branch as a snake. We don’t see the world; we see our brain’s best guess about the world, sculpted by history and context. In CBS, that guess runs wild because the usual checks are gone. But in everyday life, prediction is our greatest tool: it lets us drive on autopilot, recognize a friend’s voice across a noisy room, and walk confidently into the dark.
Knowing that my hallucinations stemmed from an overachieving predictive system—rather than a failing mind—was liberating. It turned a source of fear into a window onto how perception itself works. The brain isn’t trying to trick me; it’s just doing its job, one well-calibrated guess at a time.
Bottom Line for Patients and Clinicians
About 10–30% of visually impaired adults experience CBS, especially those with advanced macular degeneration. The clinical triad is simple: complex visual hallucinations, documented vision loss, and preserved insight. No lab test confirms it—just a good history and the confidence to ask the right question without stigma.
Management is equally straightforward: education and reassurance. Let patients know CBS is common, benign, and evidence of a brain that works too well, not one that’s broken. Then add practical tips: increase ambient light to disrupt spontaneous activity, shift gaze rapidly during an episode, or engage in conversation to activate top-down controls. Optimizing refractive error or considering cataract surgery can reduce episodes dramatically if the visual pathway remains intact.
Pharmacologic options exist—SSRIs, anticonvulsants, low-dose antipsychotics—but they’re rarely needed and often overprescribed because clinicians fear missing psychosis. When in doubt, remember: the CBS patient knows it’s not real. That single fact should anchor your response.
Predictive processing doesn’t just explain CBS; it explains how vision itself works. We’re not passive receivers of light—we’re active predictors, constantly stitching the world together from fragments and expectations. When that machinery loses its anchor in reality, it doesn’t crumble; it improvises. And sometimes, the improvisation looks like a keyboard where there is none.
