Singapore’s Vision for AI Safety Bridges the US-China Divide

The authorities of Singapore released a blueprint coming for world collaboration connected artificial intelligence information pursuing a gathering of AI researchers from nan US, China, and Europe. The archive lays retired a shared imagination for moving connected AI information done world practice alternatively than competition.

“Singapore is 1 of nan fewer countries connected nan satellite that gets on good pinch some East and West,” says Max Tegmark, a intelligence astatine MIT who helped convene nan gathering of AI luminaries past month. “They cognize that they're not going to build [artificial wide intelligence] themselves—they will person it done to them—so it is very overmuch successful their interests to person nan countries that are going to build it talk to each other."

The countries thought astir apt to build AGI are, of course, nan US and China—and yet those nations look much intent connected outmaneuvering each different than moving together. In January, aft Chinese startup DeepSeek released a cutting-edge model, President Trump called it “a wakeup telephone for our industries” and said nan US needed to beryllium “laser-focused connected competing to win.”

The Singapore Consensus connected Global AI Safety Research Priorities calls for researchers to collaborate successful 3 cardinal areas: studying nan risks posed by frontier AI models, exploring safer ways to build those models, and processing methods for controlling nan behaviour of nan astir precocious AI systems.

The statement was developed astatine a gathering held connected April 26 alongside nan International Conference connected Learning Representations (ICLR), a premier AI arena held successful Singapore this year.

Researchers from OpenAI, Anthropic, Google DeepMind, xAI, and Meta each attended nan AI information event, arsenic did academics from institutions including MIT, Stanford, Tsinghua, and nan Chinese Academy of Sciences. Experts from AI information institutes successful nan US, UK, France, Canada, China, Japan and Korea besides participated.

"In an era of geopolitical fragmentation, this broad synthesis of cutting-edge investigation connected AI information is simply a promising motion that nan world organization is coming together pinch a shared committedness to shaping a safer AI future," Xue Lan, dean of Tsinghua University, said successful a statement.

The improvement of progressively tin AI models, immoderate of which person astonishing abilities, has caused researchers to interest astir a scope of risks. While immoderate attraction connected near-term harms including problems caused by biased AI systems aliases nan imaginable for criminals to harness nan technology, a important number judge that AI whitethorn airs an existential threat to humanity arsenic it originates to outsmart humans crossed much domains. These researchers, sometimes referred to arsenic “AI doomers,” interest that models whitethorn deceive and manipulate humans successful bid to prosecute their ain goals.

The imaginable of AI has besides stoked talk of an arms title betwixt nan US, China, and different powerful nations. The exertion is viewed successful argumentation circles arsenic captious to economical prosperity and subject dominance, and galore governments person sought to liking retired their ain visions and regulations governing really it should beryllium developed.