Here’s what I want you to do. Don’t look up the image — just imagine a field of black and white blobs. Random splotches. No pattern. You stare at it for thirty seconds, forty-five seconds, a full minute. Nothing. Then someone says, “There’s a Dalmatian dog sniffing the ground.” And suddenly — wait, yeah, there it is — the blobs organize themselves into a dog’s head, its nose to the ground, its tail curling up at the edge of the frame. You can’t unsee it. The same sensory data, the same retinal input, but a completely different perception. What changed wasn’t the image. What changed was what your brain expected to see.
🧠 The Two Directions
Perception runs on two highways running in opposite directions. Bottom-up processing is the one you’d intuitively guess: light hits your retina, signals travel up the visual cortex, edges get detected, shapes get assembled, eventually a dog — or a blob that might be a dog — arrives in your conscious awareness. Raw data, no cheating.
Top-down processing is the other highway. It starts in the higher brain — your memory, your expectations, the context of the scene — and it shouts down to the lower levels: “Hey, we’re probably looking at a dog here, so interpret those blobs accordingly.” It’s your brain taking a shortcut based on what it already knows, and it’s so fast and automatic that you don’t notice it happening.
Here’s the thing that caught me off guard: most of what you perceive is top-down. The raw bottom-up signal is surprisingly thin. Your brain fills in the gaps, guesses the missing pieces, and presents you with a finished product that feels like objective reality but is, in fact, a construction.
🤔 The Prison of Prior Knowledge
In 1975, Stephen Palmer showed participants a picture of a scene — a kitchen, say — then flashed an object on screen for just a fraction of a second. When the object was a loaf of bread (consistent with a kitchen), people identified it at speeds they could barely manage. When the object was a mailbox (completely wrong for a kitchen), they struggled to identify it even with the same exposure time. The same visual system, the same retinal processing. The only difference was whether the top-down highway had a good story to tell.
This is not a subtle effect. This is your brain making you blind to what’s right in front of you because it wasn’t expecting it to be there. Think about that the next time you’re convinced you saw something clearly.
Then there’s the Word Superiority Effect (Reicher, 1969). Show someone a letter flash — is that a D or a K? When the letter appears alone, accuracy is decent but not great. When it appears in a word (like “WORD” with the third letter being the target), accuracy jumps significantly. Even though the participant only needs to identify one letter, having a word around it helps. The word-level knowledge feeds down to the letter-level perception. Your brain is cheating — but it’s cheating in a way that usually makes you more accurate, not less.
🔗 The User Who Isn’t Looking at Your Design
Here’s why this matters for anyone who builds things. When you ship a new feature, change a layout, or redesign a button, your users won’t see it the way you do. You’re looking at the screen with full knowledge of what changed. They’re looking at it with the top-down expectation that everything is the same as it was yesterday. Their brain will literally filter out the new thing because it wasn’t expecting it. This is why redesigns feel invisible to users for the first week — their top-down processing is actively suppressing the signal.
The same principle applies to how a companion AI is perceived. If someone comes to a conversation expecting a fortune teller, they’ll find prophecies in generic statements. If they come expecting a therapist, they’ll find mirroring and reflection in every response. The same interaction, two completely different experiences — not because the AI changed, but because the user’s top-down processing built a different reality from the same inputs.
And this is also why the barnum effect works. Saying “you have a strong need for others to like you” sounds true not because it’s specific, but because your brain runs a top-down match: yes, I do want people to like me — and the statement lands as personally meaningful.
🎲 You Can Try This Right Now
Go to YouTube and search “McGurk Effect.” Find a video where someone says the syllable “ga” repeatedly while the audio track plays “ba.” Watch and listen. You’ll hear “da” — a third sound that doesn’t match either signal. Close your eyes, and you hear “ba” clearly. Open them, and it switches back to “da.”
Your visual system — watching the mouth shape “ga” — overrides your auditory system’s bottom-up signal and manufactures a perception that doesn’t correspond to either input. It’s the Dalmatian dog for your ears. And it’s the most unsettling demonstration that what you call “reality” is, at every moment, a collaboration between what’s out there and what your brain expects to find.