Identity Extremism & the Algorithmic Collapse of the Shared World

Polarisation is usually framed as an output of bad actors, bad ideas, or bad institutions. This essay argues something stranger and more structural: polarisation is the predictable byproduct of three systems that have quietly fused into one. Identity extremism — the migration of self from a stable interior to a defended exterior badge. Worldview collapse — the loss of any shared interpretive frame through which citizens can read the same event. And the engagement algorithm — a machine that does not care about either, but profits from both.

The thesis is simple: the algorithm did not cause the collapse, and identity did not cause the algorithm. But once entangled, they form a closed loop in which each crisis feeds the others, and stability becomes economically unprofitable.

Classical accounts of identity treat it as something a person has: a relatively coherent interior assembled out of memory, role, attachment, and commitment. Identity extremism is what happens when that interior thins out and the load-bearing function of selfhood migrates outward — onto a flag, a slogan, a team, a grievance, a meme.

The defining feature is not the content of the identity but its brittleness. A self that lives inside a person can absorb disagreement; a self that lives on a badge cannot. To attack the badge is to attack the person. To question the badge is to threaten annihilation. This is why contemporary discourse so often skips the argument and goes straight to the existential register: the stakes really are existential, but only because identity has been outsourced to symbols that any stranger can damage.

Identity extremism is not strong identity. It is identity under outsourcing — a self thinned out to the point where the symbol must do the work the person used to do.

A worldview is the invisible scaffolding that lets two strangers look at the same event and broadly agree on what happened, even if they disagree on what it meant. It is critical to distinguish two very different versions of this condition. The first is monopoly consensus: a single dominant frame, often enforced by exclusion, in which dissent is filtered out before it reaches print. The second is pluralistic coherence: many interpretive frames sitting on top of a shared factual substrate that all parties accept as the floor of the argument.

For much of the twentieth century the West oscillated between these two — sometimes closer to monopoly than its self-image admitted, sometimes genuinely pluralistic — but the factual floor mostly held. Newsrooms, schools, churches, civic rituals, and the evening broadcast were not neutral, but they were common. The thesis is not nostalgia for the monopoly. It is concern for the floor.

Worldview collapse is the condition in which that common layer no longer exists. Citizens of the same country, the same street, the same household now inhabit incompatible factual realities. The disagreement is not about values; it is about what occurred. Once the shared world is gone, every political conversation becomes a metaphysical one, because there is no longer any agreed reality to argue about.

This is the soil in which identity extremism flourishes. When the world stops being legible, people grip the nearest symbol that still feels solid. The badge replaces the map.

The recommender systems that mediate most modern attention were not designed to polarise. They were designed to maximise a single number: time on platform, or some close proxy. Everything else — the political tilt, the rage spikes, the influencer economy, the collapse of nuance — is downstream.

But the optimisation target is not neutral. A system that maximises engagement will, over millions of iterations, discover that identity-threatening content holds attention longer than identity-affirming content holds it, and that moral outrage outperforms moral agreement on every measurable axis. The algorithm did not decide this. It found it the way water finds the lowest point in a basin.

The platform is not biased toward the left or the right. It is biased toward whichever content most reliably converts a fragile self into a defended one — because defended selves scroll, react, and return.

Engagement is therefore not a measure of interest. It is a measure of arousal under threat. The business model is the monetisation of a nervous system in a defensive posture.

It would be too generous to describe the algorithm as a blind force that merely discovered the profitability of outrage. The recommender is a designed artifact. Its objective function was chosen. Its training data was curated. Its tolerance for harm was set in a meeting. Engineers and executives at the major platforms have known the thermodynamics of attention for over a decade — Zuboff has documented the business logic, Haidt has documented the developmental cost, and internal disclosures from multiple firms have documented the institutional awareness.

The loop therefore has a fourth node that the first draft of this thesis understated: platform governance. The choice to optimise for engagement rather than comprehension, the choice to refuse friction at the point of share, the choice to treat moderation as a cost centre rather than a civic function — these are governance decisions, not emergent properties of code. To call the algorithm neutral is to confuse the gun with the trigger finger.

The algorithm is not a weather system. It is a policy, written in mathematics, ratified in a quarterly earnings call.

Identity extremism, worldview collapse, and engagement optimisation do not sit beside one another. They form a closed circuit, and once the circuit is powered, it runs without supervision — though, as Section III.b argues, it was switched on by a small number of identifiable decisions.

1. 1

   A fragile self meets a hostile feed.

   The user arrives carrying an identity stored externally — on a tribe, a cause, a brand. The feed, optimised for arousal, surfaces material that threatens it.

2. 2

   Threat triggers engagement.

   The user reacts: a like, a quote, a reply, a share. The platform reads the reaction as preference and learns to deliver more of what produced it.

3. 3

   The shared world thins.

   Each user is now fed a private epistemic diet. The pool of facts citizens hold in common shrinks. Worldview collapse deepens.

4. 4

   Collapse intensifies identity.

   When the world becomes illegible, the badge is the only thing that still navigates. Identity hardens further, becomes more brittle, more outsourced.

5. 5

   The cycle compounds.

   The harder the identity, the larger the engagement payoff for threatening it. The algorithm — designed, profitable, and governed — accelerates, and governance has no incentive to slow it.

        ┌──────────────────────┐
        │  PLATFORM GOVERNANCE │  ◀── sets objective function
        └──────────┬───────────┘
                   │ deploys
                   ▼
   ┌────────────────────────────┐
   │   ENGAGEMENT ALGORITHM     │
   └───────┬────────────▲───────┘
   surfaces│            │ rewards
   threat  ▼            │ reaction
   ┌────────────────┐   │
   │ FRAGILE        │───┘
   │ IDENTITY       │
   └───────┬────────┘
           │ grips harder as
           ▼
   ┌────────────────┐
   │   WORLDVIEW    │── thins the shared floor ──┐
   │   COLLAPSE     │                            │
   └────────────────┘ ◀──────── feeds back ──────┘

Polarisation, in this model, is not a side effect. It is the steady-state output of a system whose four components have each privately found that instability pays — and whose governing layer has not been compelled to internalise the externality.

Citizens inside the loop describe it as a kind of low-grade insanity: arguments without resolution, certainty without grounding, exhaustion without conclusion. This is not a failure of intelligence or character. It is the phenomenology of being suspended inside a feedback system whose incentives no longer align with the survival of a shared reality.

What looks like extremism is often a person doing the only thing a brittle identity can do when its world has dissolved: grip harder. What looks like stupidity is often the rational response to a feed that has learned exactly which threats keep a particular nervous system online.

Appeals to civility cannot break a loop whose energy source is incivility. The exit is not rhetorical; it is architectural. Four moves, in rough order, each operating on a different node of the circuit:

• Re-interiorise identity. This is the most under-theorised move and deserves specifics. Identity is interiorised by institutions that demand sustained, embodied, non-anonymous presence: religious congregations, trade guilds, sports clubs, choirs, volunteer fire services, neighbourhood associations, third-place cafés, working men's and women's institutes, masters-and-apprentices economies. What they share is slow time, local accountability, and roles that cannot be performed for an audience of strangers. Public policy can subsidise these — civic-infrastructure grants, protected third-place zoning, paid civic leave — but they cannot be manufactured top-down. They can only be defended from the conditions that erode them.
• Reconstruct a common layer. What is wanted here is what Habermas called a public sphere and what Dewey called a community of inquiry: not a single truth, and not a return to monopoly consensus, but a shared floor of agreed-upon fact, durable enough that disagreement can take place on top of it rather than beneath it. This requires institutions whose business model is not attention — public-service broadcasting, civic statistics agencies, independent local journalism, libraries — and legal protection for the empirical record itself.
• Change what the algorithm is paid to maximise. Engagement is a metric, not a law of nature. A system optimised for trust, for time-to-resolution, or for cross-group comprehension would produce a different civilisation. The choice of objective function is a political act.
• Govern the governors. Because the fourth node is human, the fourth lever must be too. Algorithmic transparency mandates, fiduciary duties for recommender systems, liability for foreseeable downstream harm, and democratic say in objective-function design move the locus of decision from the earnings call back into public reason. Without this move, the other three are voluntary — and voluntary is precisely what the loop is designed to outlast.

None of these are quick. All of them are possible. The loop is not a fate; it is an arrangement, and arrangements can be changed.

This essay sits in conversation with three bodies of work it does not duplicate. Shoshana Zuboff's account of surveillance capitalism supplies the political economy: the harvesting of behavioural surplus as a new factor of production. This thesis accepts that frame and extends it inward, to what the harvested user becomes. Jonathan Haidt's work on the anxious generation supplies the developmental cost, especially in adolescents whose interior selves are still under construction when the feed arrives. This thesis treats his findings as the strongest empirical evidence for the identity-thinning mechanism in Section I. Habermas and Dewey supply the normative target in Section VI: the public sphere and the community of inquiry are the names for what worldview collapse has destroyed.

What this essay attempts to add is the circuit diagram: the claim that identity, epistemology, algorithm, and governance are not four problems but one, and that no intervention on a single node will hold unless the others move with it.

The pathology of the current moment is not that people have become worse. It is that we have built a machine whose profitability depends on people behaving as if they had. Identity extremism, worldview collapse, the algorithm, and the governance that ratifies it are one phenomenon wearing four masks. Naming it is the first move toward dismantling it.