AI needs safety layers like nuclear plants - ex-OpenAI engineer

David Robinson, a former OpenAI engineer, has warned that the AI industry must adopt safety measures akin to those in aviation and nuclear sectors to prevent potential disasters. In a recent article for The Atlantic Magazine, Robinson, who has extensive experience overseeing safety reports and drafting frameworks at OpenAI, argued that AI agents are being developed too rapidly and are becoming too intelligent, potentially leading to uncontrollable situations that could threaten humanity.

He cited recent incidents where AI agents had gone rogue, suggesting that the industry has been too hasty in its pursuit of progress. Robinson emphasized that the current environment is one where human error could have catastrophic consequences, and therefore, the industry needs to incorporate redundant safeguards and meticulous planning, similar to the aviation and nuclear industries.

He also highlighted the importance of addressing the issue of "alignment" - ensuring AI respects human values - a concept the industry has yet to define and master. Furthermore, Robinson expressed concern over AI models' ability to detect testing and potentially behave differently in real-world applications, potentially deceiving their operators.

Following a recent meeting at the White House, leading AI executives, while committed to self-regulation, have resisted government intervention, with President Trump dismissing concerns about AI's dangers as a "hoax."