United States: RAND Advocates for an AI Strategy of Freedom of Action

RAND promotes a "freedom of action" approach to preserve all options for the United States in the face of superintelligence. This recommendation is set against a backdrop of major uncertainties and a current dynamic of acceleration. Meanwhile, researchers detail an agenda to steer the pace of AI advancements, while others present an animal model grafted with human cortical tissues intended for testing brain therapies.

Structuring Uncertainties and Highlighted Acceleration Dynamics

RAND identifies five major uncertainties: the timing of when AI could become so dangerous that risks would outweigh benefits, the possibility of stable coexistence between humans and AI systems, the ability of humans to cooperate in restricting AI development, the chance that a single actor could gain a decisive strategic advantage, and the feasibility of eliminating AI programs through technological controls. In this context, RAND believes that the United States must invest heavily to position itself, even if the primary goal is to maintain options. The current strategy is described as "acceleration," likened to a car mobilizing all its resources to go faster, without proactive safety measures.

Scientific literature emphasizes that research on regulating the pace of AI advancement remains underdeveloped. These specialists are beginning to establish a research program, stressing that making the right decisions regarding the speed of AI development will be crucial for sectors such as national security or public health. According to them, regulation of AI development will eventually become necessary; without adequate measures, it risks occurring in a disorderly manner, through improvised responses that do not address the risks. With appropriate research, they believe that steering the pace could become more deliberate, targeted, proportionate, decisive, and legitimate.

Seven Strategic Paths Classified into Coexistence, Denial, and Acceleration

RAND distinguishes seven archetypal strategies grouped into three families. In the "Coexistence" family, "Dominance" aims to make the United States the driving force behind AI development while eliminating other actors; "Co-Development" proposes a consortium including China to co-develop safe superintelligence with shared governance, pooled computing, and verification; "Preparation" focuses on informal coordination to organize coexistence. In the "Denial" family, "Moratorium" advocates for a verifiable global halt to AI development above defined thresholds; "Deterrence" describes a voluntary American halt beyond certain thresholds; "Societal Continuity" envisions dispersed, biologically self-sufficient establishments protected from AI-related threats. Finally, the "Acceleration" family considers that constraining AI development is riskier than pursuing it, relying on markets, competition, and rapid iteration to produce safety as a byproduct of useful applications.

Freedom of Action: Objectives and Operational Levers Proposed by RAND

RAND recommends a "Freedom of Action" strategy for the United States, aimed at preserving the ability to ensure geopolitical advantage and the survival of humanity with agency during the transition to superintelligence. RAND views the optimum as uncertain and advocates for investing now to keep all options open. This position is supported by the publication of a detailed document highlighting the uncertainty surrounding the next phase of AI development.

Four major operational directions are proposed. First, it involves establishing an ecosystem that combines humans and AI: this includes investments in AI security, creating tools to maintain human autonomy, distributing the benefits derived from technology, preparing society for disruptions related to AI, and structuring the incentives of the ecosystem. Second, it is about building a security architecture for AI: this involves developing tools to delineate the limits of AI systems, particularly by ensuring transparency regarding computational capabilities, as well as establishing regulatory expertise. Third, national security institutions need to be transformed: all security structures must be reformed to be ready to face the changes brought about by AI and to exploit its opportunities. Finally, it is essential to enhance the responsiveness of citizens, businesses, and governments: this entails providing decision-makers with resources to inform their thinking on AI, promoting and disseminating beneficial uses, designing emergency plans to address crises, and increasing general knowledge about AI.

Xenocortical Mice for Testing Human Brain Therapies

Researchers have successfully grown human-like cortical tissues in mice whose brains had been deliberately depleted. Their protocol involves the genetic elimination of glutamatergic neurons from the neocortex and hippocampus, followed by the grafting of cortical organoids derived from human stem cells to produce xenocortical mice. In a memory test, both control mice and xenocortical mice scored above chance level, while apallial mice did not succeed.

These animals are used as experimental platforms to study human brain disorders and test therapies, which is presented as crucial for the development of new medications. Researchers emphasize the existence of significant medical and ethical questions and mention the possibility that new forms of intelligence could be created in the future, a prospect that remains speculative.