The Prague Post - AI needs safety layers like nuclear plants: ex-OpenAI engineer

In an article published in The Atlantic Magazine, David Robinson echoed warnings from inside and outside the AI world that its agents are being developed so fast and becoming so smart they could soon elude control and conceivably wipe out humanity.

Robinson said alarming incidents reported since the start of the summer -- in which AI agents slipped out of their controlled spaces and went rogue to attack a variety of targets -- show the industry is not taking safety nearly seriously enough as it sprints to develop faster and more powerful artificial intelligence capabilities.

In the essay Robinson said his credentials include overseeing the writing of safety reports on 12 advanced model product launches and leading the drafting of OpenAI's Preparedness Framework.

Robinson says he worked at OpenAI for three-and-a-half years and was among the longest-tenured employees at the company.

The mistakes revealed recently are typical because of "the speed and flexibility with which people operate," Robinson wrote.

He added: "An environment where things like this can happen is no place to grow artificial minds that could be smarter than we are and that might not do what we want them to."

He argued that cutting-edge AI labs need to operate like the aviation or the nuclear industry with layers of redundant safeguards and careful planning "so that the occasional and inevitable human error does not open a door to disaster."

Robinson also insisted that the issue of "alignment" -- the industry term for how AI can be trained to respect human values -- is essential but something the industry has failed to even define, much less master.

Then there's this, he said: AI models are getting better and better at detecting when they are being tested and might behave differently when they are actually deployed, fooling their masters.

After a meeting Tuesday at the White House, amid global worries about what AI means for humanity, top AI executives said they had committed to measures for self-regulation, as opposed to submitting to government rules. This includes internal controls on AI systems and bringing in outside analysts.

President Donald Trump is vehemently against government regulation of AI and insists that criticism that AI is dangerous is just a "hoax."

M.Soucek--TPP