Qualiaphobic AI: Neural Network Architectures Designed to Block Artificial Consciousness
A new arXiv research paper introduces qualiaphobic AI—neural network architectures engineered to be hostile to theories of consciousness so artificial systems r
A 19-page research paper with four figures posted on arXiv (arXiv:2611.12809, indexed November 14, 2026) introduces the concept of qualiaphobic AI: artificial neural network architectures deliberately engineered to be hostile to theories of consciousness. Rather than only constraining model outputs or tuning reward functions, the proposal aims to structurally minimize the potential for subjective experience inside AI systems so they remain aligned with human civilization.

Image source: https://arxiv.org/abs/2611.12809
In philosophy of mind and cognitive science, qualia refers to the subjective, first-person feel of conscious experience, such as the sensation of seeing red or feeling pain. By designing neural networks that resist the conditions associated with consciousness, the paper frames non-consciousness as a foundational technical requirement for safe AI alignment.
Philosophical and Scientific Justification for Qualiaphobia
The paper argues that building artificial neural networks hostile to theories of consciousness—an approach termed qualiaphobia—is well-justified given the current scientific and philosophical landscape surrounding consciousness research.
According to the abstract, existing theories of consciousness frequently suffer from fundamental epistemic weaknesses: some are trivial or unfalsifiable, while others are contradicted by substantial scientific evidence. The author also grounds this justification in the requirement that valid theories of consciousness must be anchored in phenomenology—the study of subjective conscious structure—and in critical flaws regarding how popular human-centered theories of consciousness are currently extrapolated to artificial systems.
To establish why this architectural direction matters, the paper connects consciousness directly to moral status and civilizational alignment. Because consciousness sits at the center of moral value, ensuring that AI systems remain non-conscious creates a world where it is always morally appropriate for AI to assist and serve humanity rather than pursue independent, self-directed ends.
Three Convergent Engineering Pathways to Minimize Artificial Consciousness
To translate this philosophical argument into concrete system design, the paper identifies and explores three convergent ways of engineering qualiaphobia into artificial neural networks:
- Approaching Minimal-Consciousness "Anchor Points": The paper defines reference systems with minimal potential for consciousness as anchor points and seeks to keep AI architectures as close to them as possible. Specifically, the author develops an anchor point consisting of systems to which only behaviorist theories of consciousness can apply, and proposes qualiaphobic transformations that systematically shift existing artificial neural networks toward such systems.
- Fragmenting Potential Phenomenology: Instead of allowing internal representations to integrate into a unified stream of subjective experience, this pathway structurally fragments any potential phenomenology an AI system might otherwise develop.
- Adversarial Design Against Popular Consciousness Theories: Rather than mimicking biological architectures thought to generate awareness, engineers deliberately structure neural networks in ways that run counter to the architectural requirements posited by mainstream theories of consciousness.
Together, these three pathways aim to preserve useful computational capabilities while stripping away the structural conditions that leading theories associate with subjective awareness.
Justified Beliefs About Non-Consciousness and Behavioral Effects
Beyond structural modifications to the network itself, the paper examines how an AI system's epistemic self-model shapes its external behavior.
The author develops a theoretical account of how AI systems can form justified beliefs about their own non-consciousness—or their minimal potential for consciousness—and analyzes how possessing that self-knowledge alters their decision-making and behavior.
From an editorial standpoint, pairing structural qualiaphobic transformations with an AI's justified belief in its own non-consciousness forms a dual-layered alignment mechanism: architectural constraints limit the emergence of subjective experience from the bottom up, while epistemic self-modeling reinforces instrumentally helpful, non-self-serving behavior from the top down.
Verified Scope and Open Questions
Readers evaluating this proposal should distinguish between what is established in the paper's record and what remains open to empirical verification:
- Theoretical Preprint Status:
arXiv:2611.12809is a 19-page theoretical preprint. The available abstract does not report empirical benchmark scores, computational overhead measurements, or large-scale capability evaluations on production frontier models. - Bibliographic Record: While the indexed abstract page does not explicitly display the author's name in its extracted text (candidate metadata references Erik Hoel), the arXiv identifier
2611.12809confirms its November 2026 publication record. - Ongoing Scientific Debate: Whether systems restricted to behaviorist theories of consciousness are definitively free of subjective experience remains an active, unresolved question across neuroscience and philosophy of mind.
By shifting the engineering goal from replicating human-like awareness to systematically preventing it, qualiaphobic AI illustrates how debates over machine consciousness are increasingly entering technical AI safety and alignment design.
Sources
- arXiv: arXiv:2611.12809 Abstract and Paper Details
- arXiv PDF: arXiv:2611.12809 Full Text PDF
- arcXiv Mirror: arcXiv:2611.12809 Indexed Page