The big bosses of artificial intelligence (AI) say they want to slow down the frenetic pace of development of the sector, but competition, the reluctance of the American government and the geopolitical dimension constitute all obstacles to this initiative.
What are AI leaders offering?
On Saturday, the boss of Anthropic, Dario Amodei, pleaded for a deceleration in order to better understand the risks arising from the new capabilities of artificial intelligence.
He was publicly supported by his OpenAI counterpart Sam Altman, as well as Elon Musk, Google DeepMind President Demis Hassabis, and Microsoft CEO Satya Nadella.
Why now?
In recent months, OpenAI and Anthropic have reported several incidents which saw AI spontaneously emerge from their confined environment to go on the internet and attack sites and platforms. These AIs have shown themselves capable of coordinating, establishing a hierarchy among themselves, cheating and erasing traces of their misdeeds.
Furthermore, on Tuesday, Jacob Coxon, resigned from Anthropic and formerly of OpenAI, publicly accused the two start-ups of lax security and “(gaming) with our lives”.
Will they slow down on their own?
The competition between the flagships of this technology is such, with hundreds of billions of investments at stake, that none of them will risk slowing down alone. Dario Amodei said he was ready to coordinate with his rivals outside China.
According to The Information website, Anthropic, OpenAI and Google have been discussing the creation of a professional supervisory body since July. For now, the only effective measure, announced by both Anthropic and OpenAI, is to welcome permanent independent observers, who will verify, internally, the companies’ work on AI security.
Is this a PR stunt?
While many in the community welcomed Dario Amodei’s message, others questioned his motivations. For Brian Roemmele, entrepreneur and personality in the AI sector, this is a business pitch presented a few weeks before Anthropic’s probable IPO.
“We are the responsible company (among the AI giants), we are the solution, buy” the products or shares of the company which offers you “a security bonus”, he developed on X.
Several figures from private equity in tech have also accused Anthropic of “regulatory capture”, that is to say of wanting to obtain from public authorities that they freeze, through regulation, a hierarchy of which the start-up occupies the top.
“Stop acting as if this was (an initiative) altruistic,” commented Donald Trump’s former advisor on AI David Sacks, still influential in the White House.
“You would be massively exposed to (civil liability proceedings) if your products allowed a devastating cyberattack,” he argued.
Can AI slow down without support from the Trump administration?
Donald Trump on Sunday described the exhortations of AI industry bosses as “negative forces”. Several Republican figures have said they are reluctant to actively regulate this technology, so as not to “stifle” innovation and risk seeing China win the AI battle, according to the Speaker of the House of Representatives, Mike Johnson.
On Sunday, Dario Amodei once again defended the need for public authorities to regulate AI. He judges, like Sam Altman or Demis Hassabis, that rules set by the sector itself would not be sufficient.
Can we slow down without China?
Dario Amodei suggests synchronization limited to “democratic countries”, excluding China. But the race for AI and Chinese advances in the field lead to “the most delicate dilemma” as to whether or not to slow down, he analyzed in an interview broadcast on Sunday by CBS.
The United States plans to discuss AI security on the sidelines of Chinese leader Xi Jinping’s visit to Washington at the end of September, but China is wary, according to several media, of a possible American attempt to impose its nomenclature.
The gap is all the more pronounced as China favors the emergence of so-called open AI, that is to say freely accessible and modifiable, as opposed to the so-called closed models of the big American names. Open AI is, by definition, more difficult to control, because its users can modify its settings, including for malicious use.
To allow the most successful non-Chinese companies to slow down, Dario Amodei is campaigning to strengthen restrictions on the export of the most advanced American chips to China. He is also in favor of better protection against the plundering of Western AI by Chinese laboratories.
