Doom Debates: Holly Elmore vs. Yudkowsky on AI Safety, Consent, and Power

SPEAKER_1 (0:01)

When the future feels too important to leave alone, the arguments get intense fast. Welcome to Doom Debates, where Liron Shapira talks with Holly Elmore about Eliezer Yudkowsky's controversial ideas on AI safety and civilization's long-term fate. And there's a major question at the heart of this conversation: was Yudkowsky's early vision visionary or fundamentally undemocratic? Critics say his plans implied an AI-driven takeover, including this idea of a "pivotal act" that could lock in the future without broad human consent. Right, but defenders push back hard. They argue that Coherent Extrapolated Volition, or CEV, was meant as an idealized framework for solving humanity's coordination problems, not a literal coup.

Holly Elmore isn't buying that framing. She makes a crucial point: you cannot retroactively get consent to radically reshape everyone's world, especially not by scanning brains or handing control to a system that claims to know what humanity would want. For her, the core issue is self-determination, not just technical alignment. That brings us to something really interesting— the history of the Pause AI movement and PAI US, the organization Elmore founded. She describes a major shift in the Overton window, where AI pause arguments moved from fringe activism into congressional discussion and even legislative drafting. What once looked like educational outreach now looks like direct political influence. And that success creates its own branding debate. Some people prefer "Pacing the Frontier" because it sounds broader and more viral.

But Elmore treats that as a softer, less honest compromise. The central message remains simple: pause first, deliberate publicly, and secure democratic buy-in before racing ahead. From there, the conversation turns into a critique of the rationalist and effective altruist world around Yudkowsky. Elmore says the inner circle often operated like a secretive community whose shared knowledge about AI danger was part of its glue. She argues this culture helped fuel the talent pipeline for companies like OpenAI and Anthropic, even as it claimed to warn the world. Liran and Elmore also debate Yudkowsky's personality itself. At moments, he's described as a genius with genuine intellectual honesty; at others, as someone whose arrogance, contempt for "normies," and provocative style damage the movement he helped build. And things get more concrete when they discuss photos, polyamory, and "consensual non-consent" culture. Elmore says these choices make AI safety look cultish and off-putting, especially when the same people are trying to influence policymakers. She stresses that "no has to mean no."

The John Wentworth controversy sharpens that point. His grant from the Survival and Flourishing Fund was pulled after public concerns about conduct and community norms—a test of whether private behavior can be separated from research credibility. Throughout, there's a growing sense that AI safety has entered a more serious phase. The movement is managing reputation, internal factionalism, and the risk of public backlash.

That's why the final takeaway lands so forcefully: "First they ignore you, then they laugh at you, then they fight you, then you win."

And in this episode, Elmore clearly believes the fighting has already begun.