Erik Hoel: qualiaphobic AI architectures are already the dominant strategy in welfare debates

erikphoel · x · 2026-11-13

A debate on AI consciousness and model welfare. The counterargument: even if qualiaphobic architectures (avoiding attributions of qualia) would eliminate AI welfare issues, you can't force everyone to adopt them, and the policy rests on the unproven assumption that current models aren't conscious.

Erik Hoel's reply: far more companies and users would accept qualiaphobic architectures than take on the position of model consciousness with direct welfare measures. Treating models as non-conscious is already the dominant strategy — qualiaphobia would at least justify that assumption and prevent moral harms. He also notes the piece's second sentence explicitly disclaims the assumption in question.

A representative exchange in the "avoidance architecture vs. active welfare" dispute within AI ethics.

Original post →

Erik Hoel: qualiaphobic AI architectures are already the dominant strategy in welfare debates

More from AGI Musings