You know that feeling when Spotify hands you a song you instantly love, you play it into the ground for a week, and then somewhere around day ten you can’t stand to hear it again? That’s not just you being fickle. A new study suggests the very algorithm that found you that perfect song might be the thing quietly draining the fun out of your listening — and out of your movies, your shows, and your reading too.
The research comes from Samsun Knight, an assistant professor at the University of Toronto’s Rotman School of Management who also happens to be a published novelist. His paper, “Engagement-based curation and the evolution of taste,” appeared in the Journal of Cultural Economics. Knight started wondering about this after his own odd experience with Spotify: he’d fall in love with a recommended song, and then the app would keep shoving that same song at him until he couldn’t bear it. Why, he asked, would a company that badly wants you to stay happy keep making you miserable?
Here’s the core idea. The more you listen to a certain style of music, the better you get at appreciating it — Knight borrows the economists’ term consumption capital for this. But appreciation follows an upside-down U. A moderate amount of exposure makes you like something more. Too much exposure makes you sick of it. This is really a story about the mere exposure effect — the well-documented tendency to like things more simply because they’ve become familiar — running straight into its own limit. Familiarity builds liking, right up until it tips over into “please, anything but this.”
Now here’s the problem Knight built a mathematical model to expose. Recommendation algorithms optimize for what keeps you clicking today. They test content over a few weeks or months. But real human taste evolves over ten or twenty years. So Knight ran simulations — a thousand separate trials — pitting different kinds of algorithmic “curators” against a simulated listener whose tastes slowly shifted over time. One curator naively assumed that high engagement just meant high quality. It never realized that its own past recommendations were the reason a song felt familiar and got clicked in the first place.
What happened? The precise, engagement-hungry algorithm stopped exploring almost entirely. When it showed the listener something unfamiliar and got a lukewarm response, it decided that whole genre was bad and buried it. Its exploration rate dropped to zero. Then it played the safe favorites until the listener was thoroughly bored — a self-fulfilling prophecy of monotony. There’s even a name in the paper for the trap: straddling, where the system overplays a great song until you’re sick of it, while occasionally testing a mediocre one just enough to confirm it’s mediocre, never realizing that simply resting the good song would bring the joy back.
And the punchline, the part I found genuinely surprising: a worse algorithm did better. When Knight added a little random noise — forcing the system to occasionally toss in something unfamiliar — the simulated listeners discovered new styles, built appreciation for them, and got a break from their overplayed favorites. The slightly imperfect system made people happier in the long run. His sharpest example is hip-hop. It took a lot of listeners years to learn how to hear it; early on it sounded abrasive to ears raised on rock and roll. Knight points out that if a 1980s Spotify had ranked hip-hop by people’s initial distaste, the genre might have been buried before it ever got off the ground.
I spent years in e-learning and watched my own son disappear into video games that were engineered to keep him engaged minute by minute, and this paper put words to something I’d half-noticed for a long time. The systems that are best at giving us what we want right now can be terrible at helping us become people with bigger, richer tastes later. There’s a real difference between a tool that satisfies you and a tool that helps you grow.
So what can you actually do with this? Be your own source of randomness. Once in a while, hand the keys to a human — a friend’s playlist, a librarian’s pick, a critic whose taste runs different from yours. Deliberately rest the songs and shows you love instead of binging them flat. And when an algorithm keeps serving you the same comfortable loop, treat that as a signal to go wander somewhere it would never send you. The boredom you’re feeling might not be a sign that there’s nothing good left — it might just be a sign that you’ve been fed the same thing one too many times.
If you’re studying psychology, scroll down — I’ve pulled out the key concepts this research illustrates, with plain-language definitions you can use for an exam.
Psychology Terms in This Article
Mere exposure effect — The tendency to develop a preference for things simply because we’ve encountered them repeatedly. This study is built on the upside of that effect: the more you’re exposed to a style of music or art, the more you learn to appreciate it. The twist is that the same familiarity that builds liking eventually overshoots into boredom — so an algorithm that maximizes familiar content rides the mere exposure effect right past its sweet spot.
Habituation (satiation) — A decrease in responsiveness to a stimulus after repeated or prolonged exposure. In the model, listeners get “sick of” a favorite song because their response to it weakens every time it’s replayed. Knight’s “straddling” trap is essentially habituation in action: the algorithm keeps replaying a great song until the listener habituates, never realizing a rest period would reset the response.
Reinforcement — In operant conditioning, any consequence that strengthens the behavior it follows. Recommendation systems treat your clicks and plays as reinforcement signals, “rewarding” whatever you engage with by serving more of it. The paper shows the danger of a system that only follows immediate reinforcement: it optimizes for the next click while quietly narrowing the range of things you’ll ever enjoy.
Sensation seeking — A personality trait describing the drive to pursue novel, varied, and stimulating experiences. The research highlights what gets lost when an algorithm refuses to explore: the novelty that lets tastes evolve. The “noise” that improved long-term satisfaction in the simulation is essentially a manufactured dose of novelty — the thing sensation seeking naturally pushes us toward and that over-precise systems strip away.
References
Knight, S. (2026). Engagement-based curation and the evolution of taste. Journal of Cultural Economics. https://doi.org/10.1007/s10824-026-09591-3
Reporting by Eric W. Dolan, PsyPost (June 2, 2026).
Leave a Reply