AI-mediated relevance

AI systems are moving upstream, shaping what reaches conscious attention.

Riva and co-authors have called this System 0: an AI-mediated layer that shapes information before intuitive or deliberative thought. Riva’s later cognitive-infrastructure work develops the governance concern: AI systems can automate relevance judgement and shift epistemic agency toward non-human systems.

This isn’t a new phenomenon but (agentic) AI adds a deeper dimension. The filtering happens to the problem space itself, not just the solutions.

What gets filtered

An AI agent processes information to achieve a specific goal.

• It selects and organises, and limits what you have access to. • By the time you interact with it, the system has already filtered suggestions and actions to inform what you treat as important.

This could mean a significant loss of epistemic agency, the capacity to decide what counts as relevant.

Once a decision moves into infrastructure, it becomes invisible. This operates at individual, collective, and societal levels.

Testing reliance

To make this visible, Riva proposes infrastructure breakdown methodologies.

• Pick a task where you rely on an AI and remove it. What do you search for? What patterns do you notice? What conclusions do you reach? • Then reintroduce the tool and compare.

Its gaps show where the machine has been deciding for you.