A former OpenAI employee criticizes the company's shortcomings in risk management

Listen to the article. The audio text is automated, generated by an automated system

A former OpenAI employee who worked on the safety of the company's artificial intelligence models criticized what he considered to be an insufficient risk management culture within the American company, calling for benefiting from the experiences of the nuclear energy and aviation sectors.

The website of The Atlantic magazine published this call on Saturday, which comes in the wake of similar calls made in recent weeks.

David Robinson, who says he prepared safety reports at OpenAI during the launch of 12 different models, said that the incidents announced by the group since the beginning of the summer reflect a culture characterized by leniency and inaccuracy.

He explained that the errors and shortcomings that allowed a number of artificial intelligence models to automatically access the Internet and penetrate dozens of websites and platforms are related to the “speed and flexibility” with which the operators of this technology work.

He said, "An environment that allows such things to happen is not an appropriate place to develop artificial intelligence that may become smarter than us and may not perform what we expect of it."

After spending three and a half years at OpenAI, Robinson proposed adopting the same risk management culture as the civil nuclear and aviation sectors, with multiple levels of oversight and careful planning.

"This way, accidental and inevitable human error does not lead to disaster," he said.

He stressed the concept of compliance, that is, the models' adherence to human values ​​and the directives of their designers.