Roko Critiques Core AI Safety Terms: P-Doom, Superintelligence, and Singularity Lack Rigor
mimi10v3 · x · 2026-08-05
Roko Mijic published a thread sharply criticizing the foundational terminology used in the AI safety and alignment community, arguing that none of them hold up under close inspection:
- P-Doom: Lacks a coherent definition of "Doom" and ignores the potential risks of not developing AI.
- Superintelligence: Claimed to be mathematically impossible (with a proof promised).
- AGI: So ill-defined and flexible that there is no consensus on whether or when it is achieved.
- Orthogonality: The weak form is trivially true, while the strong form is likely false and ultimately unknowable.
- Singularity: The strong form is provably impossible, and the weak form is trivial and uninteresting.
More from AGI Musings
- AutismCapital Essay: AI and Robotics Will Inevitably Replace Human Emotional Roles — AICopyLab · 2026-08-05
- Blind Test: AI-Written Stories Preferred Over Human, Readers Can't Tell the Difference — EricTopol · 2026-08-05
- Jeff Ladish: Superintelligence Can Cure Disease and Poverty, If Aligned — JeffLadish · 2026-08-05
- Study: People Prefer AI-Written Stories Blindly, Rate Highest When Falsely Told They're Human — EricTopol · 2026-08-05
- Stanford Study: Intensive Use of AI Companions Linked to Lower Psychological Well-being — Diyi_Yang · 2026-08-05
- Joke: AGI Timelines Decay with Distance from SF in an Inverse Square Law — ziv_ravid · 2026-08-05