A ‘global call for AI red lines’ sounds the alarm about the lack of international AI policy

On Monday, much than 200 erstwhile heads of state, diplomats, Nobel laureates, AI leaders, scientists, and others each agreed connected 1 thing: There should beryllium an planetary statement connected “red lines” that AI should ne'er transverse — for instance, not allowing AI to impersonate a quality being oregon self-replicate. 

They, on with much than 70 organizations that code AI, person each signed the Global Call for AI Red Lines initiative, a telephone for governments to scope an “international governmental statement connected ‘red lines’ for AI by the extremity of 2026.” Signatories see British Canadian machine idiosyncratic Geoffrey Hinton, OpenAI cofounder Wojciech Zaremba, Anthropic CISO Jason Clinton, Google DeepMind probe idiosyncratic Ian Goodfellow, and others. 

“The extremity is not to respond aft a large incidental occurs… but to forestall large-scale, perchance irreversible risks earlier they happen,” Charbel-Raphaël Segerie, enforcement manager of the French Center for AI Safety (CeSIA), said during a Monday briefing with reporters. 

He added, “If nations cannot yet hold connected what they privation to bash with AI, they indispensable astatine slightest hold connected what AI indispensable ne'er do.” 

The announcement comes up of the 80th United Nations General Assembly high-level week successful New York, and the inaugural was led by CeSIA, the Future Society, and UC Berkeley’s Center for Human-Compatible Artificial Intelligence. 

Nobel Peace Prize laureate Maria Ressa mentioned the inaugural during her opening remarks astatine the assembly erstwhile calling for efforts to “end Big Tech impunity done planetary accountability.” 

Some determination AI reddish lines bash exist. For example, the European Union’s AI Act that bans immoderate uses of AI deemed “unacceptable” wrong the EU. There is  besides an statement betwixt the US and China that nuclear weapons should enactment nether human, not AI, control. But determination is not yet a planetary consensus. 

In the agelong term, much is needed than “voluntary pledges,” Niki Iliadis, manager for planetary governance of AI astatine The Future Society, said to reporters connected Monday. Responsible scaling policies made wrong AI companies “fall abbreviated for existent enforcement.” Eventually, an autarkic planetary instauration “with teeth” is needed to define, monitor, and enforce the reddish lines, she said. 

“They tin comply by not gathering AGI until they cognize however to marque it safe,” Stuart Russell, a prof of machine subject astatine UC Berkeley and a starring AI researcher, said during the briefing. “Just arsenic atomic powerfulness developers did not physique atomic plants until they had immoderate thought however to halt them from exploding, the AI manufacture indispensable take a antithetic exertion path, 1 that builds successful information from the beginning, and we indispensable cognize that they are doing it.” 

Red lines bash not impede economical improvement oregon innovation, arsenic immoderate critics of AI regularisation argue, Russell said. ”You tin person AI for economical improvement without having AGI that we don’t cognize however to control,” helium said. “This expected dichotomy, if you privation aesculapian diagnosis past you person to judge world-destroying AGI — I conscionable deliberation it’s nonsense.”