AI Catastrophic Risks and National Security: Taking Stock of Perceptions and Approaches

Introduction

Hello everyone, and thank you for coming today. I am Paul Shan, Executive Vice President and Director of Studies at the Center for a New American Security (CNAs). It's an exciting time for artificial intelligence (AI), where we see exponential growth in data, computing power, and algorithms. Policymakers are being challenged to understand this rapidly developing technology and anticipate potential risks and opportunities it presents.

The Significance of AI

We're witnessing tremendous breakthroughs in AI, and if current trends continue, we are on track to see systems in 2030 trained with 1 million times more computational power than today's state-of-the-art. This raises significant questions about how we prepare for all the associated opportunities and risks. There's a mix of excitement and fear surrounding AI, with concerns ranging from AI-enabled surveillance to existential risks threatening humanity. This discussion aims to explore catastrophic risks and understand their nuances.

Introduction to the Report

The recent report by CNAs researchers Bill Drexel and Caleb Withers seeks to bring clarity to these discussions, providing a sober and thorough analysis of AI and its associated catastrophic risks. It offers practical recommendations for policymakers and decision-makers striving to govern this critical technology effectively.

Key Recommendations from the Report

Drawing from Bill Drexel's presentation, here are the five primary recommendations outlined:

Differentiate Risks Deliberately:
- Policymakers, technologists, and journalists should avoid using terms like catastrophic and existential risks interchangeably. Precise and deliberate use of terminology is essential in preventing confusion and effectively communicating the nature and extent of risks.
Holistic Risk Assessment:
- Continue to evaluate risks across various dimensions, such as new capabilities, technical challenges, system integration, and development conditions. Historically, disasters often result from multiple factors coalescing.
Expand Testing and Evaluation:
- Support comprehensive testing and evaluation for advanced AI models to ensure their safety when used in high-impact domains.
Plan for International Catastrophes, Especially from China:
- Recognize that catastrophic risks from AI, particularly those originating abroad (like in China), can affect the United States due to global interconnectivity. Strengthening defense mechanisms and keeping an eye on international AI development is crucial.
Promote Risk Mitigation Measures Internationally:
- Continue promoting international norms and agreements for responsible AI use, like the political declaration for responsible military use of AI and autonomy.

Panel Discussion Insights

The event transitioned to a panel discussion featuring Wyatt Hoffman from the Department of State, Alexander Seymour from the House Committee on Homeland Security, and Michael Kaiser from the Department of Homeland Security (DHS). Their insights and perspectives were vital in understanding national security's different facets concerning AI.

Michael Kaiser

Michael emphasized the importance of building consensus among various communities to understand AI-related catastrophic risks practically. DHS focuses significantly on biosecurity and is working on regulatory approaches that focus on the physical transition of what's developed in AI environments.

Alexander Seymour

Alexander discussed cyber security and critical infrastructure protection, emphasizing the balance between regulation, promoting innovation, and adopting AI Technologies safely and securely. He highlighted Congress's pragmatic approach to evaluate existing laws, identify gaps, and determine risks they are willing to accept.

Wyatt Hoffman

Wyatt pointed out both the risks and opportunities AI presents in the military domain. The State Department focuses on mitigating unintended consequences of AI application in military settings by promoting international norms ensuring explicit, well-defined use cases, and transparency in the development and deployment of AI technologies.

Interactive Panel and Audience Questions

The panel answered various questions about the exponential growth of AI computing power, the dual-use nature of AI research, and the complexities of global consensus on AI safety standards. They also discussed the challenges and opportunities of collaboration with China on AI-related security risks and how their respective offices are navigating these issues.

Conclusion

Understanding and navigating the rapid advancements in AI requires insightfully defined policies and global cooperation. The work by CNAs and colleagues in various governmental sectors is crucial in shaping a secure environment that embraces the vast possibilities AI presents while preemptively managing its risks.

Keywords

AI catastrophic risks
National Security
CNAs
Artificial Intelligence
Exponential Growth
Computing Power
International Collaboration
Cybersecurity
Critical Infrastructure
Dual-Use AI
AI Governance

FAQ

Q: What is the primary goal of the CNAs report on AI catastrophic risks?

A: The primary goal of the CNAs report is to provide policymakers and decision-makers with a thorough, clear understanding of AI catastrophic risks and practical recommendations to govern this technology effectively.

Q: How do AI catastrophic risks differ from existential risks?

A: Catastrophic risks pertain to events causing substantial, wide-scale damage and disruption but not necessarily threatening humanity's survival. Existential risks refer to scenarios where AI could potentially imperil humanity's ongoing existence.

Q: What is the significance of AI computing power growth by 2030?

A: By 2030, AI systems could have 1 million times more computing power, raising questions about new capabilities and associated risks, emphasizing the need to understand, monitor, and prepare for these advancements.

Q: How should policymakers address the risks from international AI development, especially from China?

A: Policymakers should recognize that international AI developments, especially from China, can impact global security. It's essential to monitor these risks, strengthen defenses, and promote international norms and agreements for responsible AI use.

Q: What role does the balance between regulation and innovation play in AI governance?

A: It's crucial to balance regulation and innovation to ensure AI's safe, secure, and responsible use without stifling technological advancements. Policymakers continuously evaluate existing laws, identify gaps, and discern acceptable risks to foster an environment conducive to innovation.