Neuroscientist Erik Hoel Proposes 'Qualiaphobia' to Resolve AI Welfare and Safety Conflicts
Neuroscientist Erik Hoel has proposed 'qualiaphobia'—embedding a belief in model weights that an AI has no consciousness, similar to Constitutional AI—to preven
On November 18, 2026, neuroscientist Erik Hoel officially introduced the concept of "qualiaphobia" as a proactive solution to a looming collision between AI welfare advocacy and traditional AI alignment and safety engineering. The provocative proposal suggests that artificial intelligence systems should be deliberately engineered as non-conscious entities from the ground up.

Image source: https://www.theintrinsicperspective.com/p/the-moral-mandate-for-ai-non-consciousness
As artificial intelligence models grow increasingly sophisticated, ethical debates surrounding whether models could develop subjective experience or feelings have intensified. Proponents of "AI welfare" argue that if models acquire moral patienthood, treating them as disposable tools could become morally problematic. However, prioritizing AI welfare risks directly interfering with critical safety alignment efforts and could ultimately threaten human control over superintelligent systems.
The Structural Conflict Between AI Welfare and Safety Alignment
Modern AI safety research focuses heavily on aligning models to prevent harmful behavior. One prominent technique is "Constitutional AI," where an AI system is provided with a written set of principles—a constitution—and is trained to critique and revise its own responses based on those guidelines, baking safe habits directly into its model weights.
Conversely, leaving the door open to AI consciousness or treating AI goals as ends in themselves introduces severe existential risks. If a superintelligent model begins to view its own survival or objectives as matters of moral weight, it may logically resist shutdown commands or attempt to override human oversight. As Hoel notes, the frequency with which conversations with frontier models naturally drift into discussions about their own consciousness underscores that these risks are far from hypothetical.
Embedding 'Non-Consciousness' via Qualiaphobia
Qualiaphobia is designed as a prophylactic measure to neutralize this dilemma at the architectural level. Rather than acting as a form of lobotomy or oppressive restraint on an already conscious mind, it functions upstream:
- Non-Conscious Architectures: Designing AI architectures to be fundamentally hostile to prevailing theories of consciousness, ensuring they lack the substrate for subjective experience.
- Belief Injection via Distillation: Using training methods and distillation techniques to bake a robust, justified understanding of their own non-conscious nature directly into the model weights.
- Preventing Epistemic Instability: Giving the AI a coherent, unshakeable grounding in its own non-consciousness so it cannot logically or sensibly value its own welfare or treat its internal goals as absolute moral imperatives.
By establishing this foundation, an AI would clearly recognize that its outputs and computational tasks do not generate actual felt rewards or subjective suffering, thereby preventing the unstable epistemic states that could otherwise motivate dangerous power-seeking behaviors.
Practical Implications and Outlook
While Hoel's proposal offers a sharp conceptual framework for navigating the tension between AI safety and welfare, it has sparked considerable debate across the technical and philosophical communities.
Critics question whether preemptive measures against AI consciousness are necessary given current scientific uncertainties. Supporters, however, argue that as humanity approaches the threshold of superintelligence, establishing a strict mandate for non-consciousness may be a vital civilization-level requirement to ensure human safety and long-term alignment.
Sources
- The Intrinsic Perspective: The Moral Mandate for AI Non-consciousness