There was a time when the shape of a person's cultural life was determined by a modest set of forces: the record shop within walking distance, the friend with unusual opinions about cinema, the secondhand bookstall that happened to stock a battered copy of Rilke. These encounters were limited and often haphazard, yet they produced something distinctive. Taste was assembled slowly, from accidents and arguments, and it bore the fingerprints of a particular life.

Today, a considerable portion of that assembly has been delegated. The songs that soundtrack our mornings, the films that occupy our evenings, even the books we believe we chose freely: many arrive through recommendation systems engineered to predict what we will consume next. These systems are not malign. They are remarkably good at what they do. But what they do is subtly different from what a cultivated life requires.

John Dewey argued that experience becomes meaningful when it involves genuine interaction, when the self is changed by what it encounters rather than merely confirmed. This offers a useful lens. If leisure is to be more than the pleasant passing of hours, we must ask not only whether our cultural diet is enjoyable, but whether it is ours. What follows is a framework for understanding algorithmic influence, engaging with it deliberately, and rebuilding the discovery practices that keep taste alive and self-authored.

To manage an influence, one must first see it clearly. Recommendation systems generally work by identifying patterns in what you have played, paused, finished or abandoned, then matching those patterns against the behaviour of millions of others. The result is a prediction: people who lingered over this also lingered over that. The logic is statistical rather than critical. It holds no view on whether a work is profound, only on whether you are likely to stay.

This distinction matters more than it first appears. A good critic, bookseller or knowledgeable friend recommends on the basis of what might enlarge you. An algorithm recommends on the basis of what resembles your past. The former is oriented toward who you could become, the latter toward who you have already been. Over months and years, that difference compounds into something significant: a cultural life that grows denser in familiar territory while its borders quietly stop moving.

The costs of algorithmic convenience are rarely felt as costs. They appear instead as a gentle flattening. Difficult works are structurally disadvantaged: the novel that rewards patience, the album that sounds unpleasant on first listen. Early abandonment teaches the system to offer less of them. Meanwhile the friction that once accompanied discovery, the effort of seeking and choosing, disappears. With it goes much of the investment that made a discovery feel like a personal achievement rather than a delivery.

There is also a social dimension. When millions draw from the same predictive engines, cultural conversation tends to cluster around whatever those engines amplify. Niches still exist, often richly, but they are increasingly niches the platform has identified and labelled on our behalf. We are sorted into taste communities before we have had the chance to discover our own affinities, and we may mistake a well-fitted category for an authentic identity.

Mapping this influence is a practical exercise. Consider the last ten cultural works that genuinely moved you, and trace how each one arrived. How many came through an autoplay queue, a personalised feed or a 'because you watched' row? How many came through a person, a place or a deliberate search? The ratio is rarely what people expect, and simply seeing it tends to change how one relates to the next recommendation.

Takeaway

An algorithm recommends from the logic of who you have been. A cultivated taste requires sources oriented toward who you might become.

None of this argues for abandoning recommendation systems. They are extraordinary instruments, capable of surfacing obscure recordings and forgotten films that no single shop could ever stock. The question is not whether to use them but in what spirit. The sophisticated approach treats the algorithm as a capable research assistant rather than a curator. It is useful for carrying out an intention and unreliable as a source of one.

The first discipline is to arrive with a direction. Searching deliberately for Malian guitar music or postwar Japanese cinema, then letting recommendations unfold from that starting point, produces very different results from opening an app and accepting whatever appears. In the first case, you have set the terms and the algorithm extends them. In the second, the algorithm sets the terms entirely, drawing on a profile you never consciously constructed.

The second discipline concerns how you train the system. Every interaction is data, and most of it is given carelessly: background listening, half-watched series, tracks skipped out of mood rather than judgement. Some people keep separate profiles for ambient consumption and serious exploration, so that the casual does not contaminate the considered. Others periodically clear their histories, or use private modes for anything they do not wish to shape future suggestions. The principle is to feed the system the self you are trying to develop.

The third discipline is deliberate perturbation. Because algorithms reward continuity, occasional discontinuity keeps both the system and you from settling. Try following a recommendation chain three or four steps further than feels comfortable, or seeking out a genre you have previously dismissed. Either introduces data the system would never have predicted, and its subsequent offerings widen accordingly. In effect, you are using the system's responsiveness against its tendency toward repetition.

Finally, there is the matter of stopping. Autoplay and infinite feeds are designed to remove the moment of decision, and that moment is precisely where taste is exercised. You can reintroduce it by finishing an album and consciously choosing the next one, or by closing the app when a film ends rather than letting the next one begin. These small acts restore agency in a meaningful way. Choice, repeated, is how preference becomes character.

Takeaway

Treat the algorithm as an assistant that carries out your intentions, not a curator that supplies them. Every casual click is a vote for the self it will reflect back to you.

If algorithms optimise for resemblance, the counterweight is to cultivate sources that operate on different principles. These are older technologies of taste, and their apparent inefficiency is precisely their value. They introduce what an economist might call noise and an artist might call chance. They produce encounters that could not have been predicted from your history, and therefore cannot merely confirm it.

The most powerful of these is the human recommender. Think of a friend who knows your sensibility but does not share it, a bookseller with strong opinions, or a critic whose judgement you trust even when you disagree. Each offers something an algorithm cannot: a recommendation with reasons attached. The reasons matter. They invite you into a way of seeing, rather than simply placing an object before you. A small, varied circle of such voices is among the most valuable investments a cultural life can make.

The second source is place. Libraries, independent cinemas, record shops, museums and the season programme of a local concert hall are curated by people with commitments. They confront you with adjacent works you did not seek. Browsing a shelf is a fundamentally different act from scrolling a feed. Your eye falls on spines you would never have searched for, and the decision to pick one up is wholly yours.

The third source is lineage. Nearly every work of substance points beyond itself, toward its influences, its contemporaries and its rivals. You might read the interviews of a director you admire and then watch the films she cites, or trace a jazz musician back through his sidemen and teachers. Either way, you build a cultural map that is coherent rather than merely correlated. Discovery becomes genealogical, and taste acquires depth as well as breadth.

These practices can be formalised without becoming rigid. Some people adopt a simple ratio: for every three works that arrive through algorithmic suggestion, one must arrive through a person, a place or a lineage. Others keep a written list of recommendations from trusted voices and draw from it deliberately. The aim is not purity but proportion. A cultural diet needs enough unpredictable inputs that it can still surprise, and therefore still change, the person consuming it.

Takeaway

Serendipity is not pure luck. It is the product of placing yourself among sources that operate on principles other than your own history.

The algorithmic age has made cultural abundance effortless, and that is a genuine gift. Yet abundance without authorship produces a peculiar poverty: a life full of things to watch and hear, but with fewer encounters that genuinely reshape us. Dewey's insight endures. Experience becomes meaningful through the active participation of the person undergoing it.

A workable framework follows from the three movements above. Map where your significant cultural experiences actually originate. Engage algorithms with intention, training them deliberately and reclaiming the moment of choice. Cultivate discovery channels (people, places and lineages) that operate independently of your history.

None of this requires renouncing convenience. It asks only that convenience serve a larger purpose. Taste, after all, is not a fixed possession but a continuing conversation between who we are and what we encounter. The question worth sitting with is simple: who, or what, is doing most of the talking?