How Recommendation Algorithms Manufacture Self-Sealing Realities

Roy Bhaskar's concept of the demi-reality, developed within his critical realist philosophy, provides an unusually precise framework for understanding what Algorithmic curation systems are producing at scale. A demi-reality is not merely a distorted perspective or a collection of false beliefs — it is a partial reality that achieves internal coherence precisely through what it excludes. It feels complete from within because the absences that constitute it are themselves invisible. Bhaskar's crucial move is to grant causal efficacy to absencing: the systematic removal of perspectives, facts, and forms of contact actively shapes the epistemic environment as powerfully as any content that remains.

What makes this diagnosis contemporary and urgent is that the maintenance of demi-realities has been automated. Recommendation algorithms, optimizing for engagement rather than epistemic adequacy, generate cultlike epistemic enclosures without requiring charismatic authority or institutional conspiracy. When someone embedded in one demi-reality encounters another, the foreign view doesn't register as a competing interpretation of shared facts — it registers as ontologically impossible. This explains why fact-checking consistently underperforms: it intervenes at the empirical surface without addressing the structural absences that hold the demi-reality together.

Bhaskar's stratified ontology — distinguishing the empirical, the actual, and the real — maps cleanly onto this problem. Polarization and Epistemic Fragmentation are empirical phenomena. Beneath them sit the actual mechanisms of Algorithmic curation. Beneath those lie the real generative structures: economic imperatives, design principles, and regulatory vacuums. Any intervention that fails to reach this deepest stratum merely treats symptoms while the generative machinery continues undisturbed.