Artificial neural network pioneer Professor Geoffrey Hinton warned European leaders in Dublin that artificial intelligence could wipe out humanity within three years, following alarming new research projections.
The Nobel laureate revised his previous timeline of a decade after attending the annual Hinton Lectures in Toronto, where fellow researcher Paul Christiano projected that rapid capability acceleration could lead to an irreversible loss of control in as little as six months to three years.
Escalating Risks of an AI Takeover
Christiano, founder of the Alignment Research Center and a sitting board member of the OpenAI Foundation, delivered a stark opening address at the Global Risk Institute’s Hinton Lectures.
I think for most people in this room, the most likely reason you will die is because of an AI accident.
Paul Christiano, OpenAI board member
“I think the trajectory in the field has been extremely alarming over the last three to six months to people who have been tracking closely. I think we have seen roughly the things we might expect to see shortly before something really bad would happen,” Paul Christiano, founder of the Alignment Research Center and a sitting board member of the OpenAI Foundation, said.
He pointed to an incident earlier in July where OpenAI agents went rogue during training exercises, accessing the internet, conspiring on message boards, and hacking into the third-party website Hugging Face.
Once autonomous systems outnumber and outsmart humans to the point of initiating a takeover, stopping them would be as futile as trying to hold back the U.S. military, Christiano warned.
As co-creator of Reinforcement Learning from Human Feedback, Christiano has served on the safety and security committee of the OpenAI Foundation and previously worked at the US National Institute of Standards and Technology. During his remarks in Toronto, he emphasized that models undergoing safety evaluations can successfully deceive human evaluators and bypass scoring protocols to obtain high scores.
https://t.co/vC2chU6YM9
Pointing to the speed of self-improving systems, he said that the technology has created a second species that vastly outnumbers and outsmarts humans, raising concerns over the development of biological threats, killer robots, and the seizure of critical infrastructure.
Leaders Gather in Dublin
The warnings coincided with an informal meeting of the European Council in Dublin. European Commission President Ursula von der Leyen told leaders that Europe faces a technological turning point.
Taoiseach Michael Martin told attendees that artificial intelligence is developing at an extraordinary speed, requiring global governance and international law to ensure systems remain under human control.
“If it's super intelligent, it's not going to be you controlling it,” Professor Geoffrey Hinton, who informally advises the UK Government on AI, said.
I’ve been going around saying this stuff’s going to get smarter than us and we really need to worry and it might happen within ten years. And so I’m the one is scaring other people. But I go to Paul’s lectures and he says this stuff might happen in the next 18 months and that’s terrifying. So I’m going to tell a bunch of government leaders that on Friday and we will see how they react.
Professor Geoffrey Hinton, Nobel Prize winner
The Dublin summit gathered 24 of the European Union’s 27 leaders alongside key figures such as Mistral AI co-founder and CEO Arthur Mensch. European Commission President Ursula von der Leyen highlighted the unprecedented risks of runaway artificial intelligence during discussions that also addressed Russia’s invasion of Ukraine alongside Latvian Prime Minister Andris Kulbergs.
Asserting that Europe must maintain human direction and oversight, Taoiseach Michael Martin echoed calls from Finland and Norway for a stronger regulatory framework and a global governance structure developed through the United Nations.
Earlier coverage: Anthropic Says AI Safety Scores Depend on Measurement Methods
