Import AI 473: The US's superintelligence strategy; human brain in a mouse skull; and machine hermeneutics

The United States has developed a comprehensive strategy to maintain its geopolitical advantage as artificial intelligence (AI) continues to advance towards superintelligence. The RAND Corporation, a renowned research organization, has published a detailed paper outlining the US's "Freedom of Action" strategy, which aims to preserve human agency and ensure survival throughout the transition to superintelligence.

The Freedom of Action Strategy comprises four key components:

1. Building a human-AI ecosystem: Investing in AI safety, developing tools to maintain human agency, sharing technology benefits, preparing society for AI-driven disruption, and shaping AI ecosystem incentives.

2. Creating an AI-security architecture: Developing tools to monitor AI systems worldwide, verifying potential agreements, and establishing regulatory expertise to govern AI technology.

3. Adapting existing national security institutions for the AI era: Overhauling all security institutions to prepare for AI's impact and capitalize on AI capabilities.

4. Enhancing citizen, firm, and government capacity to respond: Providing decisionmakers with information to consider AI, promoting positive AI uses to adapt to changes, developing break-glass plans for crisis response, and raising AI literacy.

Seven archetypal strategies are explored within three families: Coexistence, Adaptation, and Preparedness. Coexistence strategies prioritize the development and dominance of the US in AI while suppressing competitors. Adaptation strategies focus on co-development with a consortium (including China), aiming for shared governance, pooled compute, and verification.

Preparedness strategies involve informal coordination, such as voluntary commitments, phased deployment, independent evaluations, incident reporting, and domestic capacity to absorb and recover from AI incidents.

The strategy also addresses five main uncertainties:

1. Danger proximity: Determining when AI becomes dangerous enough that its risks outweigh benefits.

2. Coexistence feasibility: Assessing the potential for humans and AI systems to maintain a stable equilibrium.

3. Restraint feasibility: Examining the possibility of global cooperation to constrain AI development.

4. Decisive strategic advantage: Evaluating whether a single actor can achieve a dominant position in AI development.

5. Suppression feasibility: Investigating whether domestic or rival AI programs can be controlled through technology.

The RAND analysis concludes that spending significant resources on AI development is crucial, even if the primary focus should be on maintaining optionality and proactive safety measures.