Biden's Executive Order and AI Safety with Flo Crivello, Founder of Lindy AI

Episode Summary

Executive Summary: In this episode of The Cognitive Revolution, host Nathan LeBenz interviews Flo Crivello, founder of Lindy, about AI safety and regulation. They discuss the recent White House executive order, the UK AI Safety Summit, and the existential risks of advanced AI. Flo argues that AI represents a new form of life and that precautionary regulation is necessary to prevent potential human extinction, while also acknowledging his accelerationist views on other technologies like self-driving cars. He proposes a 'kill switch' for GPUs as a technical solution and emphasizes the need for international coordination, including with China.

Main Topics: AI Safety and Regulation (Priority: 5/5): Discussion of the White House executive order, UK AI Safety Summit, and the need for sensible regulation to manage existential risks from advanced AI. AI as a New Form of Life (Priority: 4/5): Flo argues that AI represents a literal new form of life, part of the universe's evolution toward greater self-organization of matter, and that this transition must be managed carefully. Accelerationism vs. Precaution (Priority: 4/5): The hosts explore the tension between being accelerationist on most technologies (e.g., self-driving cars) but precautionary on AI due to existential risks. Technical Solutions for AI Safety (Priority: 4/5): Flo proposes a 'kill switch' for GPUs and regulating at the silicon layer to prevent rogue ASI, drawing parallels to controlling enriched uranium. AGI Timelines and Capabilities (Priority: 5/5): Flo predicts AGI within 2-8 years and ASI within 30 years, arguing that GPT-4 may already be a seed AI capable of recursive self-improvement. International Coordination (Priority: 3/5): Discussion of the need for collaboration with China on AI safety, citing game theory and existing choke points like NVIDIA as leverage. Subjective Experience in AI (Priority: 2/5): A philosophical discussion on whether AI systems have subjective experience, with Flo invoking the Zen concept of 'Mu' to reject the premise.

Key Arguments: AI represents a new form of life, part of the universe's evolution toward greater self-organization, and must be managed carefully to avoid human extinction. Regulation should target the silicon layer (e.g., GPUs) rather than applications, as it provides a choke point to control dangerous AI development. A 'kill switch' for GPUs, similar to a GPS chip that can be broadcast to, could provide a technical solution to shut down rogue AI. GPT-4 may already be a seed AI capable of recursive self-improvement, making it critical to stop open-sourcing such models. International coordination with China is possible and necessary, as the AI race involves few players in an iterated game. The precautionary principle applies to AI because even a 10% risk of human extinction is unacceptable. Self-driving cars are already safer than human drivers in many cases, and overregulation could stall progress.

Data Points: Compute threshold for regulation: 10^26 flops - The White House executive order sets this threshold for mandatory reporting to the government. AGI timeline: 2-8 years - Flo predicts a more than even chance of AGI emerging within this timeframe. ASI timeline: 30 years - Flo expects ASI to emerge within 30 years, within the lifetimes of the hosts. Risk of human extinction from AI: 10% - Flo considers even a 10% risk worth being very worried about and taking action. GPT-4 performance improvement from prompt: Significant - A paper found that asking GPT-4 to 'take a deep breath and take a step back' improves performance.

Pivotal Quotes: "There is a new form of life that's being birthed. And this is, we are like in new territory, and we need to be careful right now." — Flo Crivello: Flo argues that AI is not just another technology but a new form of life requiring precaution. "You are such a leap compared to an atom, right? Or compared to a bacteria, that there is no reason to expect that there wouldn't be another thing above you that is as much more complex or bigger than you as you are to the bacteria." — Flo Crivello: Flo explains the evolutionary perspective on AI as the next step in the universe's self-organization. "I don't expect ASI to take more than 30 years, so I expect that you and I, in our lifetimes, we're going to see ASI." — Flo Crivello: Flo makes a striking prediction about the timeline for artificial superintelligence.

Implications: Listeners should recognize the urgency of AI safety and the need for sensible regulation. The discussion highlights that even a small risk of extinction warrants action, and technical solutions like GPU kill switches may be necessary. The AI community must balance accelerationism with precaution, especially as AGI may be closer than many think.