Don't Use AI Like Short Videos

When code throws an error, or when you run into a question you have not quite thought through, the easiest move now is to hand the problem straight to AI. A few seconds later, a fluent explanation and a ready-made fix are served right in front of you. You click accept, the problem disappears, and an immediate rush of “I’ve got this handled” rises up inside. Once you get used to saving effort this way, the next time you hit a small snag, your hand switches back to the AI chat box faster than your brain starts working.

Without a conscious effort to keep a bit of resistance, using AI becomes just as addictive as scrolling through short videos. This kind of addiction is even harder to spot than scrolling through short videos. Spend two hours on short videos, and you know very well that you are just passing the time; asking AI for ready-made answers round after round, on the other hand, looks on the surface like efficient work and study. I wrote about this illusion in Understanding Can Be Generated, Judgment Cannot: because you understand the explanation AI generated, you assume you have also developed the corresponding judgment. A Wharton study tracked nearly a thousand high school students learning math and found that students with unrestricted access to a GPT tutor scored higher on everyday practice, yet once AI was removed for a closed-book exam, their scores were actually 17% lower than those of students who had not used AI. The reason is not complicated: AI had walked through the very steps where they originally needed to get stuck, make mistakes, and figure things out for themselves.

Why does following our instincts to take ready-made answers actually leave us more numb the more we do it? To see this clearly, it helps to look first at how our senses are “fed directly” in everyday life.

The biologist Tinbergen ran a famous experiment: newly hatched herring gull chicks peck at the red spot on the tip of their mother’s beak to beg for food; if you put a thin wooden stick painted with three red stripes in front of a chick, it pecks at the stick even more eagerly than at a real beak. As long as a key signal is distilled and amplified, it can make our senses react more strongly than they do in everyday life. Saccharin, neon signs, and short-video feeds all work in a similar way to cater precisely to our instinctive preferences. The psychologist Steven Pinker once compared this practice to cheesecake: nature rarely offers high sugar and high fat in the same place, whereas engineered foods concentrate them together and press the brain’s reward button directly.

The problem is that when you bypass the intermediate process and serve up the most immediately pleasing stimulus right in front of you, the resulting pleasure also fades the fastest. Human senses have a quirk: when the same stimulus repeats, the pleasure it brings drops off quickly. Food researchers call this phenomenon “sensory-specific satiety,” and experiments have found that even when people drink a sweetener with almost no calories, as long as their taste buds repeatedly receive the same sweetness, how much they like it drops noticeably within a few minutes. Short videos work the same way. The novelty of a clip starts fading in seconds, and all you can do is swipe a finger and rely on the next video to wash away the numbness that has just crept in.

That leads to a contrast worth asking about: visual and auditory arts also distill colors and amplify rhythm, so why does saccharin grow cloying the more you eat it and why do short videos leave you emptier the more you scroll, while good works of art hold up so much better?

Looking only at how strongly they stimulate the senses makes it hard to tell the two apart. The real difference lies in whether something follows your instincts and feeds you ready-made pleasure directly, or deliberately leaves a bit of resistance so your brain has to take the last step on its own.

Works of art that hold up take the latter path. They first draw you in with color, melody, or imagery, and then deliberately put a small obstacle in the way. Researchers Van de Cruys and Wagemans point out that many works first help the audience build a familiar expectation and then gently break it, like the blurred patches of color in an Impressionist painting or a chord in music that delays resolution. At first glance or first listen, the audience feels that something is slightly “off.” After looking a moment longer or listening for a few more seconds, you suddenly recognize light and shadow in the patches of color, or hear how that stretch of melody resolves, and only then does the pleasure really well up. Follow-up experiments also confirm that a viewer’s liking for a work often jumps noticeably at the exact moment of “figuring it out for oneself.” Of course, this obstacle cannot be set too high; if no clues are given at all and the audience cannot piece the whole together no matter how hard they try, all that remains in the end is frustration and the feeling of “I don’t get it.”

Putting these two paths back into everyday work makes it clear how we should treat AI. This does not mean we should refuse AI just for the sake of suffering. As I wrote in Learn to Adjust the Friction in Learning, learning should not always aim to be as smooth as possible, yet resistance also comes in two kinds: purely exhausting mechanical chores like looking up material, copying and pasting, and adjusting formatting should be handed over to AI in the first place; what you really cannot give away is the process of first reading the error message yourself, first guessing at a cause, and sitting with uncertainty before making a trade-off.

The same holds true when we encounter new tools and new things. If you only try something out in passing for ten minutes, or ask AI to generate a ready-made summary, and then tell yourself, “So that’s all it is,” you have not actually crossed its real threshold. Research in the psychology of aesthetics has found that when people face something complex, they first need to use their heads to sort out the structure and relationships inside it before they can finally experience the pleasure of understanding it. Feeling that you have already “demystified” something after a single glance usually happens because you got stuck outside the barrier and never went in; constantly moving on to try the next new thing just to keep up the novelty of “I’ve seen this too” is, at bottom, still scrolling through short videos. As I wrote in Halfway Through Building an English-Learning Tool, I Decided to Use One That Already Existed, only after building half of it myself and running firsthand into the constraints behind “importing subtitles” could I go use an existing product without either blindly worshiping it or casually dismissing it.

In the end, saccharin, short videos, unrestrained AI use, and sampling new things in passing are all doing the same thing: bypassing the trouble in the middle for you and feeding ready-made satisfaction straight to your mouth. Meanwhile, art that holds up, learning that builds real skill, and genuine demystification all follow the same path: tools and techniques can help you filter out noise and save you from mechanical trouble, but that final step of “seeing it for yourself, building it for yourself, and figuring it out for yourself” has to be left for you to walk on your own.