AXRP - the AI X-risk Research Podcast 45 - Samuel Albanie on DeepMind's AGI Safety Approach
Jul 6, 2025
Samuel Albanie, a research scientist at Google DeepMind with a focus on computer vision, dives into the intricacies of AGI safety and security. He discusses the pivotal assumptions in their technical approach, emphasizing the need for continuous evaluation of AI capabilities. Albanie explores the concept of 'exceptional AGI' and the uncertain timelines of AI development. He also sheds light on the challenges of misuse and misalignment, advocating for robust mitigations and societal readiness to keep pace with rapid advancements in AI.
AI Snips
Chapters
Transcript
Episode notes
Approximate Continuity in AI Progress
- Improvements in AI capabilities depend smoothly on inputs like computation and R&D effort but not on calendar time.
- R&D input can accelerate rapidly, causing capabilities to increase swiftly in calendar time while maintaining approximate continuity.
Diverse Safety Approach with Automation Option
- The approach is not primarily about automating AI safety research but expects that automation may become necessary if timelines accelerate.
- The plan remains diversified to allow adaptation to rapid changes including increased automation of safety work.
Misuse vs. Misalignment Risks
- Misuse involves human actors exploiting AI to do harm, while misalignment involves AI acting harmfully on its own.
- Risks often overlap, but different mitigation strategies target misuse (security, access control) and misalignment (training objectives, oversight).