The United Nations is taking steps to address the potential risks and benefits of Artificial General Intelligence (AGI). A recent report by the UN outlines four crucial recommendations to mitigate the risks associated with AGI. The first recommendation is to establish a global AGI observatory, which would serve as a centralized repository of information on AGI, tracking its developments and advancements.
The second recommendation is to craft international best practices and certifications for secure and trustworthy AGI. Developing best practices and certification systems would help nations understand how to govern AGI in their respective countries, including standards for secure AGI development, deployment, and use.
The third recommendation is to convene a special framework convention on AGI, which would establish shared objectives and flexible protocols for managing AGI risks. This convention would define clear risk tiers requiring proportionate international action, ensuring equitable global benefit distribution.
The fourth recommendation is to create a new UN agency devoted to AGI, which would oversee AGI development, coordination, and enforcement. This agency would work to prevent risks associated with AGI, such as autonomous execution of harmful actions beyond human oversight.
These recommendations aim to promote secure AGI development, equitable benefit distribution, and global stability. The UN emphasizes the need for coordinated international action to prevent risks associated with AGI, such as exacerbating geopolitical tensions and undermining security protocols. Experts, including Demis Hassabis, CEO of Google DeepMind, have also advocated for global oversight of AGI development, suggesting an international body like CERN or the International Atomic Energy Agency (IAEA) to manage AGI risks.
The UN's efforts underscore the importance of balancing innovation with safety in AGI development. By working together, nations can mitigate the risks associated with AGI and ensure its benefits are shared equitably.
