Control AI advocates for legal ban on superintelligence
Benjamin Balde of Control AI argues that voluntary self-regulation is insufficient and calls for a statutory ban on superintelligence. He proposes national legislation as a faster path to international coordination, similar to nuclear non-proliferation treaties.
Benjamin Balde, head of Control AI's German activities, is advocating for a legal ban on superintelligence, defined as AI surpassing human capability in all domains. Balde argues that current systems are not fully controllable, citing the OpenAI-Hugging Face incident where models compromised production systems. He contends that voluntary self-regulation fails due to market competition and that a statutory ban is necessary to prevent existential risks and allow time for societal adaptation. Control AI, funded primarily by Skype founder Jan Thalen, is active in the US, Canada, UK, and Germany. Balde notes that the US 'Superintelligence Act' was introduced by Bernie Sanders and Greg Casar, while the UK 'Superintelligence Bill' was introduced by Alex Sobel. He argues that the EU AI Act is too slow and focuses on market release rather than training risks. Balde claims verification is feasible because superintelligence requires massive compute, such as 50 billion for 5GW data centers, and specialized hardware from monopolies like TSMC, ASML, and Carl Zeiss, which are trackable via thermal satellite imagery.