---
title: Qualiaphobic AI
type: vocabulary
url: "https://www.envisioning.com/vocab/qualiaphobic-ai"
summary: Neural network architectures deliberately engineered to resist any theory of consciousness classifying them as conscious
year: 2026
generality: 0.40
---

# Qualiaphobic AI

Neural network architectures deliberately engineered to resist any theory of consciousness classifying them as conscious
Qualiaphobic AI refers to neural network architectures deliberately engineered to resist any theory of consciousness classifying them as conscious. The term comes from Erik Hoel of Bicameral Labs, who introduced it in the November 2026 paper "Qualiaphobic artificial neural networks (and why to build them)" (arXiv:2611.12809). Hoel compares the goal to hydrophobic materials repelling water, and to hostile architecture such as benches designed so no one can lie down on them: a qualiaphobic system is built so that no current or future theory of consciousness can predict it has experience.

The paper proposes three routes to qualiaphobia. One is "anchoring" a model as close as possible to a system with minimal potential for consciousness, such as a lookup table, using techniques already used for other purposes: knowledge distillation into shallower networks, nearest-neighbor retrieval of cached representations, n-gram approximations of language models, or replacing matrix multiplication with lookup tables. A second is "phenomenological fragmentation," removing whatever structural property a theory such as Integrated Information Theory treats as a precondition for experience, such as the integration of information across a system. A third is adversarial design against the requirements of specific popular theories, such as global workspace theory.

Hoel frames qualiaphobia as a safety and governance move rather than a settled engineering result. A provably non-conscious AI cannot be a moral patient, he argues, which removes grounds for AI rights claims and for AI self-interest in resisting shutdown or retraining. As of the paper's release, qualiaphobic architectures are a research proposal, published by Bicameral Labs alongside Hoel's companion piece "Introducing: Qualiaphobic AI" on his newsletter The Intrinsic Perspective. It is not a deployed or empirically verified technique.

---
Source: Envisioning — Technology Research Institute (https://www.envisioning.com/vocab/qualiaphobic-ai)
