Are Social Networks Still Social? / nullreference

What makes a social network social?

The obvious answer would seem to be other people.

The early promise of social networking was fairly literal. Facebook showed you what your friends were doing. Twitter showed you people you chose to follow. Instagram showed you their photos. The content was organized, imperfectly, around a social graph.

You chose the people. The people produced the content.

Open many of those apps today and the order is increasingly reversed.

The platform chooses the content. You discover the people afterward.

There is still plenty of social interaction. People comment, argue, flirt, message, share, and occasionally make friends. But social interaction can happen inside something without being its organizing purpose. A bar is social. We do not call it a social network.

So if the social graph no longer organizes much of what we see, what exactly have social networks become?

From people to content

There is an obvious problem with building an entertainment product around someone's friends.

Most people are not professional entertainers.

Your friends post occasionally. Their photographs are not optimized for retention. Your uncle does not A/B test his vacation pictures. Eventually you run out of new things they have made.

The wider internet does not have this problem.

Somewhere, someone has made something almost perfectly suited to keeping you interested for another thirty seconds. Recommendation systems have become very good at finding it.

The old social graph asks:

What have the people you care about posted?

The recommendation system asks:

What are you most likely to engage with next?

The second question has a nearly unlimited supply of answers. It is not difficult to understand why platforms moved toward it.

Nor is the change obviously bad. There is no reason the people we happen to know should produce all the information and entertainment we consume. Most books are written by strangers. So are newspapers, television shows, films, and songs.

Nobody complains that The Sopranos failed to show them what their cousin ate for lunch.

Consuming media made by strangers is not new. We have done it for centuries.

What changed is that social networks increasingly became media networks.

The television that watches back

Seen this way, TikTok or the modern Instagram feed has an older relative: television.

Traditional mass media had a relatively small number of producers and editors deciding what a large audience would see. Television networks programmed a schedule. Newspaper editors decided what made the front page. Publishers decided which books were worth printing.

The internet radically lowered the cost of producing and distributing media. Almost anyone could publish.

That solved one bottleneck and created another.

There is now far more media than anyone could possibly consume. Something still has to decide what reaches us.

The editor did not disappear. It became an algorithm.

The difference is that this editor can make a different newspaper, television channel, or front page for every person.

And this television watches back.

Every pause, replay, like, search, share, and swipe can help determine what appears next. Millions of people produce the programming while each viewer continuously gives the system information about which tiny fraction of it they should see.

Traditional television had few producers and a mass audience.

Recommendation media has mass producers and personalized audiences.

That is not quite television and not quite a social network.

It is personalized mass media.

The algorithm is still an editor

Calling the editor an algorithm does not make it neutral.

A recommendation system has an objective. Engagement is an obvious one, but it can have others. A platform can demote certain material, prohibit words or subjects, promote trusted sources, filter spam, or decide that some kinds of content should not be recommended at all.

These choices can be sensible, ideological, commercial, legal, or several at once. What matters is that they affect what people are likely to see.

Even a seemingly neutral instruction like "maximize engagement" is still a choice about what the system should value.

Users then adapt to the filter.

The strange vocabulary of the internet makes this unusually visible. Words like "unalive," altered spellings, euphemisms, and coded emojis emerge partly because users believe ordinary language will be suppressed or punished by moderation systems.

Sometimes those beliefs about the algorithm may themselves be wrong. It hardly matters. Once enough people believe the filter exists, they begin writing for it.

The filter changes the language passing through it.

This is not entirely new. Newspapers had editorial standards. Television had broadcast rules. Publishers had editors. Every medium has shaped what could be said through it.

What is different is the scale, speed, and personalization.

The system can continuously decide what millions of individual people see while those same people continuously modify their behavior according to what they think the system rewards.

A million different televisions

Traditional media did not show us an unfiltered reality either.

There are more events than can fit into a newspaper or nightly broadcast. Choosing ten stories means ignoring thousands of others. Choosing a headline, photograph, or interview affects how those stories are understood.

The old media critique was partly about who controlled that selection.

Recommendation media does not eliminate the problem. It changes its architecture.

Traditional mass media could give millions of people roughly the same selective picture of the world.

Personalized media can give millions of people millions of different selective pictures.

That has real advantages.

An unknown person can document something with a phone and reach millions of people. An expert who would never be invited onto television can build an audience directly. People with obscure interests can find each other. A story ignored by established media can become impossible to ignore.

The old gatekeepers buried things too.

But the same architecture can make fringe beliefs appear common, ordinary behavior appear unusual, or unusual behavior appear ordinary.

No conspiracy is required. Selection alone can change the apparent frequency of something.

If one unusual behavior reliably holds attention, a recommendation system may show us far more of it than we would encounter by randomly sampling the world. From inside the feed, however, the sample can start to look like the world.

This creates a strange distinction.

Modern social media may be becoming less social as a network while becoming more consequential socially.

A person scrolling through recommended videos from strangers is not necessarily socializing. But millions of people watching those videos can change how people dress, speak, date, eat, exercise, travel, decorate their homes, or understand what behavior is normal.

Some of this is trivial. A recipe becomes popular. Everyone buys the same water bottle. A phrase appears everywhere for six months and then disappears.

Other changes are more interesting.

Imagine someone in 2002 regularly taking carefully posed photographs of themselves eating, exercising, traveling, or standing in front of a mirror, then printing those photographs and putting them somewhere acquaintances and strangers could inspect them.

It would at least have required an explanation.

Today the underlying behavior is ordinary enough that doing it rarely requires one.

The technology did not merely give us a more convenient way to perform an existing social ritual. In some cases, it helped create and normalize the ritual itself.

Media does not simply transmit culture. It participates in producing it.

So there is an odd possibility here. These platforms may no longer be particularly social in the old sense of connecting us primarily to people we know.

Yet they are profoundly social in another sense.

They help determine what society sees itself doing.

A television that makes itself for you

Recommendation systems personalize which media we see. Generative AI introduces another possibility: personalizing the media itself.

Today, two people interested in the same event might be shown different videos about it. There is no technical reason they must continue receiving the same versions of those videos.

This process has already begun at the edges. YouTube, for example, allows creators to test different titles and thumbnails to see which presentation performs better with viewers. The purpose is mundane. Find the packaging people are most likely to click.

Generative media removes the technical reason to stop at the packaging.

Imagine a product review that can change for every viewer.

One person responds to enthusiasm, so the review emphasizes what the product does well. Another distrusts enthusiastic reviews, so they receive a skeptical version. One wants technical detail. Another wants a story. Another wants the conclusion in thirty seconds.

At first, the underlying facts could remain identical. Only the presentation changes.

But presentation is part of how people understand facts.

The next step is selecting which facts receive attention. A positive version spends longer on advantages. A skeptical version spends longer on flaws. Neither has to contain a lie.

Eventually there may be no canonical version at all.

Traditional mass media gave millions of people one program.

Recommendation media gives different people different existing programs.

Generative media can create a different program for every person.

At that point, the idea of an echo chamber almost seems quaint.

An echo chamber still has a chamber. A group of people reads the same stories, repeats the same arguments, and reinforces the same beliefs.

Hyperpersonalized media does not require the group.

The chamber can have one person in it.

When the objective is only engagement

The obvious fear is propaganda, but malicious intent is not required.

Suppose the system wants only to maximize engagement.

It learns that you stop watching when something challenges a particular belief. It learns that you stay when information confirms something you already suspect. Nobody has to instruct it to make you more certain. Certainty might simply retain you better.

Someone else might respond to the opposite. They enjoy being challenged, so their media becomes increasingly contrarian.

One person responds to optimism. Another to outrage. Another to conspiracies. Another wants everything explained through economics. Another prefers personal stories.

A sufficiently capable system does not merely learn what subjects interest you.

It can learn how you prefer reality to be narrated.

People already talk about training their For You pages. They like, hide, follow, search, and linger until the feed begins to feel like theirs.

Generative media could complete that process.

Instead of searching through an enormous supply of human-produced media for the thing most likely to retain you, the system can produce it.

The supply becomes infinite.

And you may have no reason to know what everyone else saw.

What happens to a shared truth?

None of this makes truth disappear.

An election still has a result. A bridge either collapsed or it did not. A company has revenues that can be measured. Reality does not become personalized because media does.

What can disappear is the shared path by which people encounter it.

Disagreement over interpretation is normal. Two newspapers can observe the same event and frame it differently.

But there is something useful about being able to put the same newspaper on a table and argue about what it says.

What happens when there is no common front page?

If every person receives a different selection of facts, a different ordering of importance, a different emotional framing, and eventually different generated media, disagreement becomes harder to locate.

Did we disagree about the facts?

Did we see different facts?

Were the same facts framed differently?

Did one of us see something because a system predicted that particular version would keep us watching?

We could end up talking past one another before the argument even begins.

The danger is not necessarily a world without truth.

It is a world with fewer shared reference points for finding it.

The world's largest game of telephone

Information already mutates as it travels.

A source publishes a finding. Someone summarizes it. A creator summarizes the summary. Someone clips the creator. Another account captions the clip. A viewer remembers the caption and tells someone else.

At every stage, information disappears and emphasis changes.

Generative media can make these transformations nearly free.

A system summarizes a source for one audience. Another rewrites that summary for another. A video generator turns it into a story. A recommendation system discovers which version spreads. Other people respond to that version rather than the original.

Nobody in this chain necessarily has to lie.

It can become an enormous game of telephone in which every participant is simply trying to make the message more relevant to the next person.

The result can still end up far from where it began.

Maybe "social media" has simply become an antiquated category.

A group chat among twenty friends and an algorithm selecting or generating videos for a billion people are both called social media despite having almost opposite organizing principles.

One begins with relationships and lets information travel through them.

The other begins with information and creates relationships, audiences, and communities afterward.

Calling both social media makes it easy to ask whether "social media" is good or bad, as though we were talking about one institution.

We probably are not.

A private Discord server, a TikTok feed, an Instagram group chat, a Reddit community, and YouTube Shorts may share technical ancestry, but they serve increasingly different purposes.

Perhaps much of what we call social media would be better understood as personalized media or media networks with social features.

That still leaves the question of what an actual social network should be.

Recreating the Facebook feed of 2011 does not seem particularly ambitious.

An online social network does not have to be a separate world where relationships are performed through posts. It could instead make relationships outside the network easier.

It could tell you that five friends are free tonight. It could remind you that someone you care about has not been around in a while. It could make "we should do this sometime" more likely to become Saturday afternoon. It could help someone find other people nearby who share an obscure interest, then become less important once they meet.

Perhaps the distinction between an online friendship and a "real" friendship should disappear in the opposite direction from where personalized media is taking us.

Instead of making the digital world sufficiently compelling that we need less of the shared one, technology could make movement between the two so ordinary that separating them stops making sense.

That creates an unusual measure of success.

A media network succeeds when you keep consuming media.

A social network might succeed when you go be social.

That is an awkward objective for a business whose revenue tends to increase with attention. It may also explain why the evolution from social graph to recommendation engine was so attractive.

There are two very different directions technology can take from here.

One gets progressively better at constructing a world for each individual.

The other gets progressively better at bringing individuals into a shared world.

Social networks have not necessarily become antisocial.

Many may simply have become something other than social networks.

If so, the interesting question is not how to make our media feeds social again.

It is what we would build if connecting people were actually the point.

Are Social Networks Still Social?