Engagement Ate the Internet

Like many other nerdy millennials, I watched The Social Network and walked out of it quietly identifying with Zuckerberg. He was portrayed as socially awkward, intellectually restless, slightly vindictive perhaps, but fundamentally relatable. Not a geopolitical actor. Not a steward of democratic discourse. Just a gifted builder who moved faster than everyone else. That framing matched how many of us saw ourselves at the time. Build something interesting. Scale it. The consequences will sort themselves out. Looking back, it is striking how little the narrative focused on what happens once scale and incentives begin to reinforce each other.

What we received as users over the following decade is not easily reducible to villainy or incompetence. It is the outcome of design decisions layered over time. Infinite scroll made feeds feel fluid. Autoplay reduced friction between pieces of content. Push notifications pulled conversations back into view. Ranking models learned to predict which posts were more likely to generate measurable response. Each change appeared incremental. Yet taken together, these adjustments altered the conditions of interaction. Being present inside the system gradually started to resemble being processed by it. When reaction becomes a ranking signal, content that triggers strong responses tends to travel further. Outrage, certainty and moral clarity generate cleaner engagement signals than hesitation or nuance. That does not require an explicit intention to amplify extreme positions. It follows from optimisation. In political contexts, especially around elections or periods of instability, this dynamic becomes more visible. Algorithmic exposure can create the impression of momentum where there may only be intensity. Repetition strengthens perception. Individuals who hold more extreme views may encounter reinforcing signals more frequently, which in turn can deepen conviction. None of this depends on a coordinated conspiracy. It is the predictable outcome of systems tuned to maximise measurable interaction.

The research landscape reflects that complexity. Social media is neither uniformly harmful nor uniformly beneficial. Outcomes vary with context, age, use pattern and social environment. Active interaction can support connection. Passive exposure can contribute to dissatisfaction. The same platform can provide community while simultaneously amplifying divisive narratives. That ambiguity makes the design of ranking systems more consequential, not less. When exposure is shaped at scale by models that are not fully legible to users, small adjustments in weighting can propagate widely without most people recognising the mechanism at work.

In Europe, regulators are beginning to treat recommender systems and interaction patterns as systems that deserve scrutiny. Transparency requirements are expanding. Very large platforms are being asked to explain how ranking works and to provide alternative modes of ordering. This is not nostalgia for chronological feeds. It reflects an understanding that digital infrastructure now mediates civic life in ways that were not initially anticipated. When a platform influences how political information circulates, its internal optimisation logic acquires public significance. At the same time, expectations from users have shifted. People want spaces where they can belong without feeling surveilled. They want to express different aspects of themselves without permanent audience collapse. They want discovery that feels intentional rather than overwhelming. They want moderation that is predictable and contestable rather than opaque. Most existing platforms technically offer fragments of these capabilities, yet they sit within a structure optimised around growth and retention. That tension is not always visible in individual features, but it shapes the broader environment.

Architectural concentration intensifies that tension. Identity, storage, ranking, monetisation and governance often reside under a single operator. Leaving means abandoning not only content but social graph and history. Portability is limited. Data ownership remains conditional. When frictionless exit is absent, agency is negotiated rather than guaranteed. Meanwhile, technical alternatives have matured. Federated protocols, portable identity layers and local-first patterns demonstrate that centralisation is no longer the only viable configuration. The constraint is less technical than economic. If a network does not monetise attention, it must rely on a different sustaining mechanism. Subscription changes incentives but reduces accessibility. Cooperative ownership distributes governance but introduces complexity. Public infrastructure models raise funding and independence questions. None of these are trivial transitions. What the past decade illustrates, however, is that business models and product design cannot be separated. The metric placed at the centre of the system quietly determines what the system becomes.

Perhaps the inflection point ahead is less about features and more about contract. The early web optimised for reach because reach was scarce. Today trust appears scarcer. A network designed under that constraint might prioritise legibility over immersion. Ranking systems that can be inspected or adjusted. Moderation that is explained and appealable. Data that can move with the user. Interfaces that do not assume constant presence as the default state. These are not technically radical ideas. They are radical relative to the dominant incentive structure.

We may not need another platform competing on the same axis. What may be required is a different understanding of what digital interaction is meant to optimise. The tools to build alternatives increasingly exist. Regulatory pressure is rising. Research continues to add nuance. Whether that translates into a different architecture depends less on technology and more on which metric we decide to centre next.

Edit (Feb. 24th.): Added a detail on political consequences of social networks and their ranking algorithms.