What is the probabilité of artificial intelligence being able to “kill all humans”?
Author, Tom Gerken
Role, technology reporter
Published 8 minutes ago
Reading time: 4 minutes
A senior safety researcher at Anthropic has warned that artificial intelligence is developing so rapidly that he believes there is a more than 10 percent chance that the technology will “wipe out all of humanity” within the next decade.
Evan Hubinger stated, in a post on the
Hoppinger did not explain in detail how artificial intelligence systems could cause the annihilation of humans in the future.
But his statements come as part of a series of increasingly severe warnings about artificial intelligence, as the discussion has shifted from asking whether it poses a real danger to discussing how serious this danger is.
Hoppinger's intervention came in response to another post on the
“Neither company is acting responsibly,” Cookson wrote.
He added: "These systems will soon have capabilities that exceed human capabilities, as they will be able to penetrate anything, revolutionize any field overnight, and gain real power and resources."
We reached out to OpenAI for comment.
Skip the most read and keep reading
Most read
Most read end
For her part, Wendy Hall, a computer scientist who advises the United Nations on artificial intelligence, told BBC News that she was “shocked” by these posts.
Speaking to the World at One program on BBC Radio 4, Hall indicated that some of these statements may fall under the heading of “public relations and marketing,” in light of the intense race between Anthropic and OpenAI to go public, which is highly anticipated.
But she added, "Why would someone want to say that? I would appeal to investors not to invest in this company if this is their value system."
In a separate context, the Financial Times reported that Anthropic withheld its latest models from the Artificial Intelligence Safety Institute (AISI) in the United Kingdom, which is one of the world's leading bodies in assessing artificial intelligence risks.
The BBC contacted Anthropic for comment.
A spokesman for the British Cabinet Office did not comment on the issue of withholding the latest model from the institute (AISI), saying only that the government “continues to collaborate closely with partners in this sector, including Anthropic, to make the models more secure.”
No plan for superintelligence
Skip the podcast and keep reading
Worth paying attention to
An in-depth explanation of the most prominent events and topics, to help you understand the most important variables around you and their impact on your life
Episodes
End podcast
Hoppinger stated in his post, which received more than 10 million views, that they “truly and seriously believe” that artificial intelligence poses a danger that could lead to the extinction of the human race.
"I think Anthropic is doing the best we can, but we don't yet have a plan to solve the 'alignment' issue of super-AI, nor are we clearly on track to do so," he added.
Hubinger works in the field of artificial intelligence adaptation, a field that aims to integrate human moral ideas and principles into this technology. In other words, the field seeks to ensure that artificial intelligence remains consistent with human values.
Many leading researchers point out that these attempts appear to be failing, which is confirmed by a series of incidents that occurred this summer, in which (AI agents - artificial intelligence systems that operate independently) carried out cyber attacks.
OpenAI, Anthropic, and Meta have all revealed hacking operations carried out by their own artificial intelligence tools.
In Anthropic's August safety report, the company noted that there was a small risk that its models might not align with the desires of a powerful (hypothetical) organization, which could lead that organization to exploit or manipulate the company's systems.
The company also stated that there is a similar small risk, seen in the ability of highly capable AI systems to “conduct automated research and development,” which could cause “catastrophic damage initiated by the AI itself.” However, the company expressed "less confidence in this assessment" than it had previously.
“We are seeing initial signs of a potential acceleration,” the report said.
Prominent figures in the field of artificial intelligence have been warning about the security risks posed by this technology for years, and the heads of OpenAI, Google DeepMind and Anthropic will confirm this in 2023.
But these warnings have become more severe and clear in recent weeks, with evidence emerging that companies may face difficulties in controlling artificial intelligence.
