Why do Experts Disagree on Existential Risk and P(doom)? A Survey of AI Experts

Expert disagreement on AI risk stems from fundamentally different worldviews: The survey found that AI experts cluster into two distinct groups—those viewing "AI as a controllable tool" versus "AI as an uncontrollable agent"—with these contrasting perspectives driving divergent beliefs about the importance of AI safety research.

Knowledge gaps are widespread even among experts: Only 21% of surveyed AI experts had heard of "instrumental convergence," a foundational AI safety concept predicting that advanced AI systems will pursue common sub-goals like self-preservation. The researchers note that "the least concerned participants were the least familiar with concepts like this," suggesting familiarity with safety research correlates with concern about risks.

There is broad agreement on the need for safety research, despite disagreement on risk levels: 78% of the 111 surveyed experts agreed or strongly agreed that "technical AI researchers should be concerned about catastrophic risks," indicating substantial consensus on the importance of the field even among those who may not believe existential risks are imminent.

Online discourse has become polarized rather than substantive: The paper notes that debate over AI existential risk "has begun to turn tribal" with participants using dismissive labels like "doomer" or "accelerationist," suggesting that productive dialogue between different expert viewpoints may be hindered by polarization rather than genuine technical disagreement.