OpenAI, Anthropic and Google DeepMind are working to create a body to supervise the risks linked to artificial intelligence (AI), OpenAI told AFP on Tuesday, September 15, as calls to better control the excesses of AI are increasing.
The idea was born from a proposal from Demis Hassabis, general manager of Google DeepMind before becoming president in early August, who suggested an entity similar to the American Financial Industry Regulatory Authority (FINRA), a self-regulatory body for the financial sector.
The creation of FINRA was authorized by federal law and is placed under the supervision of the American market watchdog, the SEC.
Also readPanic on board the AI
Evaluation process
For several months, major AI players have been calling on the American government to regulate the sector and supervise AI security.
The Trump administration set up a process for evaluating new AI models at the beginning of August, but it is based on voluntary participation and its mechanism has not been communicated.
After the call to slow down the development of AI launched on Saturday by Anthropic boss Dario Amodei, Donald Trump and several members of the presidential majority reiterated their wish not to constrain the sector, officially for fear of seeing China gain the upper hand.
The establishment by the profession of its own supervisory body would likely require a law, in particular to prevent the AI industry from being accused of cartel and competition infringement.
In his message posted online in mid-July, Demis Hassabis imagined a structure financed by companies in the sector, whose council would include independent experts as well as representatives of the open ecosystem.
Also readFreed, delivered: AI beyond all control?
The latter includes developers of open AI, that is to say freely accessible, downloadable and modifiable, unlike closed models such as those of OpenAI or Anthropic.
Once the organization is in place, the most advanced AI companies would have to submit their new models to a review, particularly focused on security, before they can bring them to market.
This new body could also initiate a coordinated slowdown in the development of AI, according to Demis Hassabis, in the event that the apprehension of risks does not keep pace with the progress of artificial intelligence.
With AFP
