Proposed Bill Seeks Moratorium on Advanced AI, Bans Artificial Superintelligence

Senators Bernie Sanders and Greg Casar have put forth legislation targeting the development of artificial intelligence systems. The proposed bill, titled the ‘Artificial Superintelligence Act’, seeks to ban the creation of AI deemed capable of superseding human control or subverting shutdown commands. This includes systems that could potentially destabilise governmental functions or critical infrastructure.

The legislation defines artificial superintelligence as any AI system that meets or exceeds human cognitive ability across a broad range of tasks, potentially surpassing human capacity in areas such as scientific creativity, general knowledge, or social skills. The bill further specifies that such systems must demonstrate the ability to act autonomously, setting their own goals and adapting to achieve them, even in direct opposition to human direction. A key prohibition within the proposed act targets AI that could independently initiate or manage operations critical to national security without human oversight.

Beyond outright prohibition, the bill advocates for a moratorium on the development of AI models that exhibit advanced reasoning and self-improvement capabilities. This pause would allow for the establishment of a robust regulatory framework. The proposed framework includes mandatory safety testing, external auditing requirements, and clear legal accountability for developers of high-risk AI systems. The bill suggests a new federal agency or an expansion of an existing body to oversee these regulations, focusing on ensuring AI development aligns with public safety and democratic principles.

The Act also calls for a comprehensive risk assessment of all advanced AI models before deployment. This assessment would evaluate potential societal impacts, including economic disruption and algorithmic bias. Non-compliance with the proposed regulations could lead to significant penalties, encompassing both financial sanctions and potential operational restrictions for developers. The proposed legislation raises questions about the feasibility of defining and regulating future AI capabilities.