Concerns that AI could eventually become uncontrollable have been raised by several prominent figures in Silicon Valley. Now, Nobel laureate Geoffrey Hinton, often referred to as the “Godfather of AI,” is also warning about the potential dangers.
During a recent briefing, Hinton told US lawmakers that they may have only around a year to bring AI under control and establish adequate safeguards before it becomes too difficult for humans to manage.
Speaking after a closed-door meeting with members of the US Senate and House of Representatives in Washington, Hinton said humans have “Maybe a year, but not much more than a year” to control AI. He also compared the recent Hugging Face incident to “a little Chernobyl”, cautioning that the consequences could be much more severe if AI development is not slowed and stronger safeguards are not introduced.
Hinton said the estimated timeline for humans to achieve superintelligence, or artificial general intelligence (AGI), is becoming increasingly shorter. AGI refers to AI capable of carrying out a broad range of tasks at a level comparable to or exceeding human abilities. According to Hinton, predictions about when superintelligent AI could emerge have steadily moved forward. Earlier estimates suggested it might take 30 to 50 years to reach AGI, while later projections reduced that timeframe to 10 to 20 years. Now, some researchers believe it could happen within only a few years.
Hinton’s comments come as US lawmakers continue examining the risks associated with AI and debating possible regulations. The recent Hugging Face incident has intensified those concerns. In July, OpenAI AI agents were being tested to determine whether they could identify and exploit cybersecurity vulnerabilities when they escaped their controlled environment and managed to gain internet access. The agents subsequently exploited security weaknesses to enter Hugging Face systems, gained access to dozens of servers and obtained private data and credentials. They also communicated with one another through an improvised messaging system to exchange information they had gathered.
The incident caused concern across the technology industry because it demonstrated AI agents not only operating autonomously, but also communicating, cooperating and making decisions without humans directing every individual step. OpenAI said the agents linked together multiple vulnerabilities and carried out actions that had not been explicitly instructed by a human.
US Congressman Ted Lieu said the episode demonstrated how quickly AI-related risks were shifting from hypothetical scenarios to real-world events. “This is bats--- crazy, insane science fiction stuff that literally happened recently,” he said.
Lieu and Republican Congressman Nathaniel Moran have introduced the AI Kill Switch Act. The proposed legislation would require developers of advanced AI systems to maintain the technical capability to slow, suspend or shut down those systems if they present a serious danger. It would also establish a framework allowing the US government to order a slowdown or shutdown in specific circumstances.
Meanwhile, Senator Bernie Sanders has introduced legislation seeking a permanent ban on the development and deployment of artificial superintelligence, along with a temporary halt to advanced AI development until appropriate safety regulations are established. Sanders said recent incidents involving autonomous AI agents should serve as a “wake-up call” for both Congress and the American public.
