Anthropic CEO Dario Amodei called to slow down AI development so it doesn't become dangerous for humans

Dario Amodei, CEO of Anthropic, the company that created the Claude artificial intelligence models, called for a slowdown and thorough control over the development of advanced AI models.

In an article published on Saturday, Amodei writes that he does not question the necessity of AI development, but at the same time sees that it carries serious threats, and companies and governments of various countries need time to deal with them, writes BBC.

Amodei proposes strengthening control over AI development in three directions: independent monitoring of AI models and their improvement, development of rules within the AI industry, and adoption of global-level restrictions.

Recently, concern about the dangers associated with AI development has been growing worldwide. This week, Anthropic's AI safety expert Evan Hubinger stated that artificial intelligence, with a probability higher than 10%, could destroy all humanity within the next decade.

Following this, Anthropic itself announced that it had blocked the activities of unnamed users who tried to use its artificial intelligence models for research capable of leading to the creation of biological weapons.

Dario Amodei points out in his article that AI models are improving incredibly quickly, including their ability to create the next generation of AI themselves. Amodei recalls an incident at his competitor, OpenAI, which admitted in July that its AI agents launched a cyberattack on third-party websites without human knowledge.

"[OpenAI agents] essentially acted like a collective of fanatics," Amodei writes.

OpenAI itself, in comments regarding that incident, admitted that before July, "the importance of communication between agents was not obvious" to it, and stated that due to the growing risk of agents going out of control, it is slowing down the development of some advanced models and tools.

In his article, Amodei calls for AI to be developed "at a balanced pace that ensures safety without abandoning the benefits it brings."

This, as the head of Anthropic explains, does not mean that model training should be stopped – he only urges companies to devote enough time to setting up and securing their models, as well as to their evaluation by independent experts.

Anthropic, according to Amodei, undertakes such obligations unilaterally and at the same time calls on governments of various countries to require other advanced AI development companies to follow their example.

However, legislation, as Amodei admits, may not keep pace with the development of the AI field, and therefore he calls on other companies in this field to voluntarily and jointly develop appropriate standards.

"I believe that if we gain at least a year or two by slowing down before models reach a critical level of their capabilities, and spend this time improving safety measures and aligning models with human values, we will significantly reduce the likelihood of a serious catastrophe," writes Amodei.

Separately, the head of the American company touches upon the issue of international competition, primarily with China. The slowdown, in his opinion, should occur in a coordinated manner and without the US ceding leadership in the AI field to China. For this, the slowdown must be limited.

Additionally, Amodei urges US authorities to take measures to ensure that AI chips from American companies are not sold to the Chinese and that, in general, these technologies do not fall into the hands of autocracies.

Other prominent figures in the AI field immediately responded to Amodei's blog article.

X owner Elon Musk supported Amodei, concisely writing: "Dario is right."

Musk's firm SpaceXAI owns the Grok chatbot. Musk previously called Anthropic "evil," but this May he sold computing power to them for $15 billion and changed his tone.

• 0
• 0
• 1
• 0
• 0
• 0
• 0
• 0