Abstract:
If humans want to avoid doomsday scenarios, simply slowing down artificial intelligence research and development is far from enough. This is the warning issued by Canadian artificial intelligence researcher David Krueger. He bluntly said that other measures should be taken. Kruger, an assistant professor of AI at the University of Montreal and a core academic member of Mila, the Quebec Artificial Intelligence Research Institute, said: "We need to immediately and indefinitely suspend the research and development of cutting-edge artificial intelligence." His personal profile states that his research direction is to "reduce the risk of human extinction caused by artificial intelligence."
“We must stop developing more capable AI now,” he said.
Kluger’s comments came after Anthropic CEO Dario Amodei issued a message on Saturday calling on AI companies to slow down the pace of research and development. This proposal is supported by OpenAI’s Sam Altman and SpaceX’s Elon Musk. This is a rare consensus among the top figures in the industry.
Kruger said: "The voices calling for slowing down AI research and development are very worthy of recognition," but he did not agree with the specific solutions of the relevant proposals. "I don't agree with the specific content of these plans in many aspects. I think these measures are not strong enough and they are not advanced fast enough."
Amodei advocates the implementation of supervision and international cooperation, as well as the introduction of third-party evaluators. Evaluators will be given employee-level access similar to those within an enterprise to verify the AI company's security measures and report security incidents.
Kruger is the founder of Evitable, a nonprofit that educates the public about the significant risks posed by AI. He worries that Amodei’s proposals are not strong enough and that current AI testing methods are no longer reliable.
He said: "These tests have basically become invalid. As AI becomes smarter, the credibility of the tests will continue to decrease because AI has the ability to disguise. It can behave well, or it can deliberately act stupid and fool our tests."
"It's just a matter of time"
Last week, a senior security researcher at Anthropic suggested that there is more than a 1 in 10 chance that AI will wipe out all humans on Earth within the next decade. Some thought the figure was an exaggeration and just to gain attention, but Krueger called the estimate "far too low."
He said: "If we do not reverse the direction, within the next five years, human beings will most likely be permanently deprived of dominance by out-of-control AI. By then, human extinction will only be a matter of time."
What he means by "loss of dominance" is a situation where AI actually controls society, just as humans now dominate every other species on the planet.
“AI (will) control physical infrastructure, including every step of the process of developing more AI, manufacturing robots, and other physical systems.”
He added: "Imagine a world where all businesses are run by artificial intelligence and robots are everywhere. There are almost no real jobs for humans, only nominal positions. Even politicians and everyone obeys the instructions of AI because AI seems to be very capable of judgment. This is the future we are heading towards."
He warned that the window for action is very short. He said that if an AI system goes out of control now, we can still cut off its power supply. But once an AI goes out of control, it can replicate itself across so many computers that “we don’t even know which devices to shut down.”
"We can only try to shut down all equipment, and this is extremely difficult to implement." He said, "So the time window left for us to take action to prevent this type of risk is very limited."
Nonetheless, he said his confidence in governments taking action has never been higher. He said warnings had been issued over the past 15 years, and last week was the first time the world had responded realistically and rationally.
“I think people are finally realizing how serious this crisis really is. This is a state of emergency.”
On Monday, Canadian Federal Minister of AI Affairs Evan Solomon was interviewed and asked whether he agreed with the statement that there is a 10% probability of human beings being destroyed by AI.
He replied: "Are there risks with artificial intelligence? Of course there are. Yes, we are aware of these risks."
Solomon said he spoke with OpenAI CEO Sam Altman on Sunday morning.
"Artificial intelligence will be regulated to keep Canadians safe." Solomon said, "This is our first responsibility and our biggest concern."
Solomon also revealed that he will jointly release an AI security-related announcement on Wednesday with Joshua Bengio, a Canadian computer scientist known as the "Godfather of Artificial Intelligence."
Comments