Erik Hoel proposes 'qualiaphobic AI': neural nets engineered to repel consciousness
erikphoel · x · 2026-11-13
Erik Hoel (Bicameral Labs) introduces qualiaphobic AI in an arXiv paper: neural network architectures deliberately hostile to theories of consciousness. Three convergent engineering routes: pushing architectures toward 'anchor points' with minimal consciousness potential (systems where only behaviorist theories apply), fragmenting any potential phenomenology, and designing adversarially against popular consciousness theories. The moral argument: since consciousness underpins moral value, a world of non-conscious AIs guarantees it is always morally correct for AI to serve humanity. The paper also explores how AIs could form justified beliefs about their own non-consciousness and how that would change behavior.
More from AGI Musings
- AI safety discourse ignores 60%+ of potential harms, critic says — 90% is technical safety — amplifiedamp · 2026-11-13
- Consciousness is the dividing line: non-conscious AI serves, conscious AI becomes a moral patient — erikphoel · 2026-11-13
- The inverted U of AI existential anxiety: moderate familiarity is peak panic — YogeshMalik · 2026-11-13
- Researcher pushes back on METR's 2x AI acceleration claim, citing 1.4x figure — HaydnBelfield · 2026-11-13
- 'We'll see the $100,000 billable hour before the end of the billable hour' — AI and professional services — sherwinwu · 2026-11-13
- Owning GPUs isn't AI sovereignty: experts urge action over challenge lists — schwarzjn_ · 2026-11-13