A kill switch won’t save us from rogue AI, tech leaders say amid growing concerns

There’s an industry-wide debate about risks concerning advanced artificial intelligence (AI) systems. From tech leaders to researchers, several have raised concerns about the challenges of keeping capable AI systems under human control. Many have suggested AI kill switches as an answer to uncertainties. However, researchers and leaders say a kill switch will not be enough to stop a rogue or potentially uncontrollable superintelligent AI system.

Geoffrey Hinton, a computer scientist who is also known as the godfather of AI, told CNN that “I don't think it'll work in the long run,” about kill switches as an answer to stop AI. When it's superintelligent, it'll be much better than people. So it will be able to persuade the people in charge of the switch not to pull the switch,” he added. The statement has raised concerns about the potential existential threat to humanity.

Must read: ‘If you’re not confident... don’t release it’: Nvidia’s Jensen Huang on how companies should take responsibility for AI safety

AI kill switch debate: Is it the solution?

While Anthropic and OpenAI share a similar vision for AI slowdown, Meta’s Mark Zuckerberg and Nvidia’s Jensen Huang emphasised AI safety. Now, the debate has also reached the US Congress. On September 16, Senator Rand Paul blocked a bill introduced by Senator John Kennedy that would have required advanced AI systems to have a kill switch.

In theory, an AI kill switch could allow companies to shut down their own AI models or cut them off from resources, external tools, servers, and computing power. However, AI experts argue that a kill switch may not be enough to control highly capable AI systems. They believe that the mechanism could only be activated once the damage has already escalated, or could serve as one layer of a broader AI safety framework.

Must read: OpenAI, Anthropic and Google have been discussing AI safety and risks for weeks: All details

Nate Soares, an AI-safety researcher, said in a Sky News interview that a kill switch may only work when the AI system is tied to a physical location, but it may not work if the system escapes or replicates itself in critical infrastructure.

Previously, Anthropic CEO Dario Amodei mentioned that an AI kill switch is “no panacea”, which means it cannot solve every AI safety problem by itself. Whereas Sam Altman said that there is no “magic red button” that can stop AI from going rogue.