Alignment Problems in AI Governance

The talk provides a thoughtful analysis of AI governance challenges, drawing on legal precedents and current regulatory frameworks. The speakers demonstrate deep expertise in both law and technology, offering a nuanced perspective on how AI systems can evade regulation. The argument is well-structured, with clear examples from copyright law illustrating the concept of anti-regulatory mechanisms. However, the talk is primarily opinion-based, lacking empirical evidence or peer-reviewed sources. The discussion of the EU AI Act is detailed but may not be accessible to a general audience. The speakers do not address potential counterarguments or limitations of their proposed solutions. Overall, the talk is valuable for its insights into the intersection of AI and law, but it would benefit from more rigorous sourcing and consideration of alternative viewpoints.

The title accurately reflects the content, which focuses on alignment challenges in AI governance.

7/10

The talk is given by experts in AI governance and law, with references to legal cases and regulatory frameworks. However, it is an opinion/expert talk without peer-reviewed sources, and some claims are not fully substantiated.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 02:00Introduction to the Sony Betamax case and the concept of safe harbor.
• 05:00Discussion of Napster and Aimster cases, illustrating anti-regulatory mechanisms.
• 10:00Introduction to the taxonomy of aversion in the context of the EU AI Act.
• 15:00Explanation of benchmark shopping and model sandbagging as avoidance strategies.
• 20:00Discussion of federated learning as a way to split computation and avoid systemic risk thresholds.
• 25:00Analysis of how AI safety science can be used as a mechanism of change.
• 30:00Discussion of companies' lobbying efforts and the role of voluntary frameworks.
• 35:00Proposal for adaptable regulatory systems and robust oversight.

• Simons Institute Event Page — Official event page with details about the talk and speakers.

• EU AI Act — The regulatory framework discussed in the talk.

• No discordant sources found — The talk does not present conflicting sources.