The caricature beats the photograph

You can pick a friend out of a caricature faster than out of a photograph. That should be strange. The caricature has the nose too big, the chin too long, the eyes too far apart — it is, measured against the actual face, wrong in every dimension. And yet it lands quicker and harder than the accurate image sitting right next to it. The exaggeration isn’t a distortion the brain forgives. It’s the thing the brain was looking for.

There’s a clean piece of animal-learning research underneath this. Train a pigeon to peck at a disc of one particular color for a reward, and — no surprise — it pecks hardest at that color and less at neighboring shades. But run the experiment a little differently: reward one wavelength and actively not-reward a nearby one, so the animal has to tell the two apart. Now the peak of its response doesn’t sit on the rewarded color anymore. It shifts — past the target, further away from the un-rewarded one. The pigeon’s strongest response is to a color it was never actually trained on. Researchers call this the peak-shift effect, and it shows up wherever a brain has to learn a difference rather than a single fact.

The reason is quietly profound. A brain that has to discriminate isn’t storing “the thing.” It’s storing “the direction the thing differs from the thing it’s not.” And if a little difference is good, more of that difference — further along the same axis — reads as more so. The gull chick that pecks its parent’s beak for food will peck even harder at a disembodied stick painted with more red bands than any real beak has. Tinbergen called these supernormal stimuli: keys that fit the lock better than the key the lock was built for. The caricaturist is doing this on purpose. They find the axes on which your face departs from the average face, and they push you further out along every one. The result isn’t less like you. It’s the most like-you a picture can be.

Here’s why this belongs to anyone building a product. Your user’s brain is not memorizing your app. It is learning what makes your app different from the last five things that looked like your app. That’s a discrimination task, and it obeys the same rule. Which means the version of your product that lands hardest is not the balanced, realistic, everything-in-proportion one. It’s the one that finds the single axis you differ on and pushes past where a reasonable person would stop.

So the build decision is not “what should our product do?” It’s “what is the one axis we’re different on — and are we rendering it at realistic size, or at caricature size?” If you’re the fast one, your empty state shouldn’t just load quickly; the speed should be almost rude, a result appearing before the user’s finger has left the key. If you’re the calm, uncluttered one, don’t ship a slightly cleaner screen than the competitor — ship one that looks almost empty, that a focus group would call too sparse. The realistic amount of your defining quality is the amount that gets you filed next to everyone else.

Two honest limits, because this principle has a dark twin. First, peak shift only sharpens the axis the brain is already discriminating on. Exaggerate a dimension nobody’s comparing you on and you don’t get a caricature, you get a clown — loud in a direction no one cares about. You have to know your real axis of difference before you can push it, and most products are wrong about what theirs is. Second: supernormal stimuli are exactly how junk food, slot machines, and infinite feeds hijack us — locks being picked by keys cut past anything nature made. The same move that makes your product unmistakably itself can, aimed at a reward circuit instead of a recognition one, make it predatory. The test is which brain you’re exaggerating for: the part that’s trying to recognize you, or the part that’s trying to stop.

The science, to look up: peak shift in stimulus generalization (Hanson, 1959, and the broader discrimination-learning literature); Niko Tinbergen’s work on supernormal stimuli; and Ramachandran & Hirstein’s “The Science of Art” (1999), which argues caricature and much of visual art run on exactly this mechanism.