Can a Recommendation Path Become a Radicalisation Path?

It is easy to tell a story in which someone watches one provocative video, receives progressively more extreme recommendations and eventually commits violence. The story has a plausible technical mechanism. That does not make it a demonstrated account of an individual attack.

Where recommendation can intervene

Online systems may contribute at several distinct stages:

Discovery. A recommender can introduce a creator, claim or community that the person did not deliberately seek.

Repetition. Similar messages can recur until a marginal interpretation appears familiar or widely accepted.

Normalisation. Visible approval and repeated exposure may change perceptions of what other people believe.

Community formation. Recommendations can help people find networks that supply identity, status and reinforcement.

Escalation. A system may rank more intense material when it predicts that intensity will sustain attention.

Amplification. Extremists can exploit platforms to distribute propaganda or magnify an attack after it occurs.

Each stage is a legitimate subject for measurement. None proves the next.

Violence has more than one cause

Ideological violence is rare relative to the number of people who encounter extreme material. Personal grievance, relationships, mental health, offline groups, political events, ideology, opportunity and access to weapons may all matter. Their interaction differs between people and cases.

The UK Independent Reviewer of Terrorism Legislation has noted both the possible role of algorithmic promotion and the limited empirical basis for much theoretical analysis of online radicalisation. Research is constrained by small samples, changing platforms, incomplete data and the difficulty of constructing a credible comparison for a person who committed an attack.

This means the responsible conclusion is conditional: recommendation can change the information and social environment through which radicalisation may occur. Available evidence does not support treating the recommender as a sufficient explanation for extremist violence.

Avoid the pipeline metaphor

A pipeline suggests a single entrance, a predictable direction and an inevitable destination. Real pathways contain exits, pauses, competing influences and deliberate choices. Many people consume provocative material without becoming extremists. Some extremists actively search for material rather than receiving it passively.

“Exposure trajectory” is a more useful term. It asks what appeared, how the sequence changed, what alternatives were available and when the user exercised agency. Investigators can then relate that record to other evidence without allowing the technical mechanism to substitute for a causal account.

The governance case does not depend on certainty

Platforms need not wait for proof that a particular recommendation caused an attack before examining foreseeable risks. They can test whether systems disproportionately distribute terrorist or violent extremist material, whether borderline content creates routes around moderation and whether users can reset or leave an escalating sequence.

At the same time, interventions create risks of their own. Automated definitions of extremism can be opaque, politically contested and unevenly applied. Governance must include appeal, transparency and examination of discriminatory effects.

The question is not whether an algorithm pulled a trigger. It is whether a system materially increased the reach, repetition or apparent legitimacy of content associated with a foreseeable risk—and what proportionate design could reduce that contribution.

References

• Independent Reviewer of Terrorism Legislation: The Terrorism Acts in 2021, especially the discussion of online radicalisation and limits of the evidence.
• European Commission: Study on algorithmic amplification of terrorist, violent extremist and borderline content.
• Home Office: The internet and radicalisation pathways.