The New Gatekeepers

For most of the twentieth century, cultural taste was shaped by a relatively small number of institutional gatekeepers: publishers, record labels, film studios, newspaper critics, museum curators. You could criticize these institutions for their biases — and there was a great deal to criticize — but their function was legible. They made choices, and the choices were made by identifiable people with articulable (if not always defensible) reasons.

The recommendation algorithm is a different kind of gatekeeper. It is not a person; it does not have reasons in any meaningful sense; it is optimizing for a metric (engagement, watch time, clicks) that is only loosely connected to cultural value. And yet it now has an influence on what films get watched, what music gets heard, what books get read that is, in aggregate, probably greater than any individual critic or institution in history.

What Algorithms Are Actually Optimizing For

To understand what recommendation engines do to culture, you have to understand what they are actually trying to do. They are not trying to introduce you to the best film you haven't seen; they are trying to keep you on the platform for as long as possible. These goals sometimes overlap and often diverge.

The practical consequence is that recommendation engines tend to favor:

  • The familiar over the challenging. Work that resembles what you already like is more reliably engaging than work that requires adjustment, patience, or the acquisition of new reference points. Algorithms learn this and act on it.
  • The recent over the enduring. Recency is correlated with the current preferences of the active user base, which is the population the algorithm is learning from. Older work tends to be underrecommended relative to its cultural weight.
  • The emotionally immediate over the formally complex. Art that produces immediate emotional response — catharsis, excitement, comfort — gets rewarded over art whose effects are slower, stranger, or harder to describe.
  • The popular over the singular. By definition, recommendations are built from aggregate data. The idiosyncratic and irreplaceable tend to get lost.

The Fragmentation Paradox

One of the more counterintuitive effects of algorithmic recommendation is that it has simultaneously fragmented and homogenized culture. Fragmented, because individuals are served highly personalized feeds that diverge dramatically from one another — there is no longer a shared cultural object (the number one album, the water-cooler TV show) in the way there once was. Homogenized, because the same optimization pressures apply across all users, tending to flatten the range of what gets widely distributed within any given genre or taste community.

The result is a cultural landscape that feels enormously varied on the surface but is, in certain structural ways, surprisingly narrow. There are more things to consume than ever; the range of modes of attention and formal experimentation they reward may actually be narrower.

What the Canon Was For

The concept of a literary or artistic canon has fallen out of favor for good reasons — canons encode the biases of those who construct them, and the traditional Western canon excluded vast swathes of human creative achievement on the basis of race, gender, and geography. These are serious criticisms that ought to be taken seriously.

But the canon also served a function worth retrieving: it was an argument about what rewards sustained engagement, what speaks to concerns beyond its immediate moment, what is worth the effort of meeting on its own terms rather than on yours. This is not something an engagement algorithm can replicate. The algorithm has no concept of "worth the effort." It only knows what produces immediate response.

Navigating the Landscape

None of this is an argument for turning off streaming services or abandoning the internet. It is an argument for using them with some awareness of their structural tendencies. Some practical implications:

  1. Seek recommendations from people whose judgment you trust, not just from systems that know your history.
  2. Be deliberate about seeking out work that is old, foreign, formally unusual, or otherwise underrepresented in algorithmic feeds.
  3. Distinguish between what you enjoy and what is genuinely good — these overlap but are not identical, and conflating them is a habit that recommendation culture encourages.
  4. Preserve spaces of cultural conversation — book groups, film clubs, editorial recommendation — that are organized around value rather than engagement metrics.

The algorithm is a powerful tool for discovery. It is a poor substitute for a culture that knows what it thinks is worth knowing.