3
The Upside Down World of Predictive Coding 6:01 Lena: Okay, Miles, let’s talk about that "upside down" view. Traditionally, we’re taught that our senses are like windows. Light hits the eye, travels to the brain, and the brain builds a picture of what’s "out there." It’s a bottom-up process. But Friston and other proponents of the Bayesian brain say that’s actually backwards.
6:22 Miles: Right, the "Camera Myth." Most people think the world flows *in* and creates the experience. But in the predictive coding framework, the experience actually flows *out*. Your brain is constantly generating top-down predictions about what it expects to see, hear, and feel. It’s like it’s saying, "Okay, I’m in a kitchen, so I expect to see a counter, a toaster, and a mug."
6:45 Lena: And then the sensory data—the actual light hitting the eyes—is just there to check those predictions?
6:50 Miles: Exactly. The sensory data isn't the *content* of your vision—it’s the *feedback*. It’s a signal that says, "Hey, you predicted a mug, and here’s some light that matches that." Or, more importantly, it says, "You predicted coffee, but this liquid is orange. Error! Error!"
7:07 Lena: So, the only things that actually travel "up" the hierarchy of the brain are the errors? The things the brain *didn't* expect?
7:16 Miles: That’s the core of it. If your prediction is perfect, there’s no "news." The brain doesn't need to process the parts of the world it already understands. It’s incredibly efficient. It only pays attention to the "prediction errors." This is why you can drive your familiar route to work and realize you don't remember the last five miles. Your brain’s model of that road is so accurate that there were almost zero prediction errors. There was no "news" to report to your conscious mind.
7:42 Lena: That’s a bit unsettling, honestly. It implies that my conscious experience is basically a "controlled hallucination." I’m living in a simulation my brain is running, and the real world only breaks through when it proves my simulation wrong.
7:58 Miles: "Controlled hallucination" is exactly the term neuroscientists like Anil Seth use. It sounds provocative, but it’s mathematically sound within this framework. Think about it—your brain is locked in a dark, silent bony vault. It has no direct access to the world. All it gets are these electrical impulses through the optic nerve and the auditory nerve. It has to *infer* the causes of those impulses. If it receives a certain pattern of "pips," it says, "Ah, that’s probably a bird chirping." It’s a best guess.
8:29 Lena: It makes me think about those optical illusions where you see a shape that isn't really there, like the Kanizsa triangle—the one where three "Pac-Man" shapes are arranged so you "see" a white triangle in the middle. My brain is so sure there’s a triangle there that it "paints" it into my consciousness, even though there are no lines.
8:48 Miles: Perfect example! Your brain’s top-down prediction of a triangle is so strong that it overrides the "raw" sensory data. It would rather believe its model of a triangle than accept a world of weirdly aligned Pac-Men. This also explains why it’s so hard to see your own typos when you’re proofreading. Your brain predicts the correct spelling of the word because it knows what you *meant* to write, so it literally doesn't "see" the error signal coming from the page.
9:14 Lena: So, we’re essentially filtering reality through our own expectations. But if that’s the case, how do we ever learn anything new? If I’m just seeing what I expect to see, how does the "orange juice glitch" actually change my mind?
9:28 Miles: That’s where the "recursive cycle" comes in. The brain is constantly doing this dance: Predict, Compare, Resolve. When that "orange juice" error signal travels up the hierarchy, it forces the brain to update its model. It’s like an internal software update. The brain says, "Okay, my 'this is coffee' hypothesis has been thoroughly debunked by the 'it’s cold and citrusy' data. Let’s update the belief to 'this is orange juice.'"
9:54 Lena: And that update happens at different levels, doesn't it? Like, there’s a fast update for "oh, it’s juice," but maybe a slower update for "I should check the mug before I sip next time"?
10:05 Miles: Spot on. The brain uses a hierarchy. The lower levels are dealing with fast, sensory stuff—pixels, edges, tones. The higher levels are dealing with abstract concepts—objects, locations, intentions. If you’re in a familiar café, your high-level prediction of "I am in a café" stays stable even as you’re processing the "low-level" surprise of a new song playing on the speakers. This hierarchy is what gives our world stability. You don't have to rebuild your entire reality every time you blink.
10:39 Lena: It’s a very elegant system. But it also suggests that our "reality" is incredibly subjective. If you and I have different "prior beliefs" or different models of the world, we could be looking at the same thing and literally experiencing different realities.
10:54 Miles: Absolutely. And that leads to some fascinating—and sometimes troubling—places when we look at things like mental health or political polarization. But before we go there, we should talk about the "other" way the brain handles these errors. Because updating your mind is only half the story. Sometimes, instead of changing your mind, you change the world.