Should there be just one western AGI project?

Abstract

Centralizing Western AI development, likely necessitating substantial US government involvement, presents potentially significant strategic implications. While a single project could streamline resource allocation and strengthen infosecurity by reducing attack surface, it also raises concerns regarding power concentration and limitations on innovation. A centralized project might reduce racing between Western entities, but its impact on the US-China AI race remains uncertain. Increased US development speed could paradoxically decrease racing by widening the lead, but it might also prompt China to accelerate its own efforts, potentially negating the advantage. Furthermore, factors like bureaucracy and reduced innovation could offset the benefits of resource amalgamation. A single project also presents challenges in ensuring broad access to advanced AI services, which could disempower actors outside the project and increase integration with the US government, potentially exacerbating power concentration risks. Conversely, multiple projects, while potentially more susceptible to security breaches and prone to racing, foster competition and broader access to AI capabilities. The optimal path involves interventions robustly beneficial in either scenario, such as enhancing infosecurity research and establishing governance structures that mitigate power concentration and uncontrolled racing. – AI-generated abstract.