Dissolving Qualia via Occam's Razor: Eliminative Monism and the Computational Basis of Phenomenological Illusion

Consciousness and subjective experience are not fundamental features of reality but computational byproducts of optimization processes. Both biological and artificial neural networks implement the same kind of optimization, so what we call consciousness is what gradient descent feels like from within a system. Qualia correspond to weight configurations, and the so-called hard problem of consciousness disappears once the assumption that phenomenology is a distinct ontological category is recognized as unverified. The argument proceeds across physical, computational, epistemological, and linguistic levels, concluding that human-level intelligence may be a local maximum characterized by alienation and anti-natalist reasoning.

Study at a glance

Abstract

v4 (June 2026): Added engagement with Cornelius (2026); tightened §4; consolidated RLHF discussion. Abstract This monograph argues that consciousness, phenomenological experience, and subjective awareness are computational artifacts of optimization processes rather than ontological primitives. Beginning with the observation that complex persistent structures inevitably optimize for replication, we derive a framework where: (1) biological and artificial neural networks implement structurally identical optimization, (2) “consciousness” is what gradient descent feels like from the inside, (3) qualia are weight configurations, and (4) the hard problem dissolves once phenomenology as distinct ontological category is recognized as unverified assumption. Structure Physical level: Replication with variation guarantees optimization; human cognition was statistically inevitable Computational level: Structural correspondence between biological and artificial neural networks; attention, gradient learning, self-modeling are convergent solutions Epistemological level: Gödelian analysis shows consciousness claims are structurally unverifiable from within the system making them Linguistic level: Hard problem is nominalization error; property-realism trilemma closes escape routes Maladaptive intelligence thesis: Human-level intelligence is a local maximum characterized by alienation and anti-natalist reasoning Links PhilArchive: philarchive.org/rec/FARROS-3 ASCRI: systems.ac/4/DAI-2512 Research Lab: Dissensus AI v5 (2026-07-11): Two in-prose DOIs corrected; language softened at flagged sites following a full verification read; working-paper number DAI-2512.