Lindy AI Founder Flo Crivello on Biden's Executive Order and AI Safety

TL;DR

Flo Crivello discusses AI safety, regulation, and future implications.

Transcript

and so I was afraid that the regulation would go something like if you install Microsoft Office in your AI then you have to make a report and it's like what like like ah it's happening again they're doing it again it's like chill I totally get it but this time is really different like this is really something special that's happening not just in th... Read More

Key Insights

• Flo Crivello believes AI is a new form of life, emphasizing the need for careful regulation and existential risk management.
• The executive order on AI safety is a positive step, focusing on flop thresholds and reporting requirements for large-scale AI projects.
• Crivello suggests a technical solution like a GPU kill switch to prevent rogue AI systems, emphasizing the need for centralized control.
• Open-sourcing AI models poses risks; Crivello argues for more controlled access to prevent misuse and potential existential threats.
• Flo Crivello predicts AGI could emerge within 2-8 years, with a 50% chance of development, and ASI within 30 years.
• Collaborating with China on AI safety is crucial, and Crivello is optimistic about global coordination to manage risks.
• Crivello argues that the recursive self-improvement capability of AI is critical, with GPT-4 already showing potential for AGI.
• The philosophical question of AI's subjective experience is deemed irrelevant by Crivello, focusing instead on measurable outcomes.

Questions & Answers

Q: What is Flo Crivello's perspective on AI as a new form of life?

Flo Crivello views AI as a new form of life, emphasizing that the evolution of the universe has been towards greater self-organization of matter. He believes that AI represents a significant leap in complexity, comparable to the leap from atoms to bacteria to humans, and stresses the need for careful regulation to manage existential risks associated with this new form of intelligence.

Q: What are Crivello's thoughts on the recent executive order on AI?

Crivello is pleasantly surprised by the executive order, noting that it is a positive step in the right direction. He appreciates that the proposed measures focus on flop thresholds and reporting requirements for large-scale AI projects, which he finds sensible. However, he acknowledges the challenges of regulating AI without loopholes and emphasizes the need for technical solutions like a GPU kill switch.

Q: Why does Crivello advocate for a GPU kill switch?

Crivello advocates for a GPU kill switch as a technical solution to prevent rogue AI systems. He believes that centralized control over powerful AI technologies is necessary to manage existential risks. By implementing a chip in GPUs that allows for a centralized shutdown, he argues that it would provide a safeguard against potential threats posed by advanced AI systems.

Q: What does Crivello predict about the timeline for AGI and ASI?

Crivello predicts that AGI could emerge within 2-8 years, with a more than even chance of development, and ASI within 30 years. He believes that the recursive self-improvement capability of AI is critical and that GPT-4 already shows potential for AGI. Despite the risks, he remains optimistic that the transition to advanced AI can be managed without catastrophic outcomes.

Q: How does Crivello view collaboration with China on AI safety?

Crivello views collaboration with China on AI safety as crucial and is optimistic about global coordination to manage risks. He believes that despite geopolitical tensions, there is precedent for successful coordination on critical issues, and emphasizes that working together is essential to address the existential risks posed by advanced AI technologies.

Q: What is Crivello's stance on open-sourcing AI models?

Crivello is against open-sourcing AI models due to the risks of misuse and potential existential threats. He argues for more controlled access to these technologies, as open-sourcing could lead to the proliferation of powerful AI systems that could be used maliciously or lead to unintended consequences. He stresses the importance of regulating the distribution and use of AI models.

Q: How does Crivello address the philosophical question of AI's subjective experience?

Crivello dismisses the philosophical question of AI's subjective experience as irrelevant, focusing instead on measurable outcomes. He uses the Zen concept of 'Mu' to reject the premise of the question, arguing that consciousness and subjective experience are not scientifically measurable and do not impact the practical considerations of AI development and safety.

Q: What are Crivello's thoughts on the application-level regulation of AI?

Crivello is skeptical of regulating AI at the application level, as he believes it is not viable for managing existential risks. He argues that as AI becomes more powerful, the risk-benefit analysis shifts, necessitating regulation at the silicon layer. He compares it to regulating enriched uranium, emphasizing that powerful technologies require control at the foundational level to prevent misuse.

Summary & Key Takeaways

• Flo Crivello discusses the implications of President Biden's executive order on AI safety, emphasizing the need for careful regulation and existential risk management. He suggests a technical solution like a GPU kill switch to prevent rogue AI systems, arguing for centralized control over powerful AI technologies.

• Crivello predicts that AGI could emerge within 2-8 years, with a 50% chance of development, and ASI within 30 years. He highlights the importance of not open-sourcing AI models to prevent misuse and potential existential threats, advocating for more controlled access to these technologies.

• Collaboration with China on AI safety is crucial, and Crivello is optimistic about global coordination to manage risks. He argues that the recursive self-improvement capability of AI is critical, with GPT-4 already showing potential for AGI, while dismissing the philosophical question of AI's subjective experience as irrelevant.