A safety engineer who recently left OpenAI, the company behind ChatGPT, has raised alarms about artificial intelligence (AI) developers being “not nearly careful enough.” David Robinson’s remarks come as concerns grow over the potential of AI becoming too powerful, a threat dismissed as a “hoax” by former US President Donald Trump.
In a recent op-ed for The Atlantic, Robinson argued that AI needs robust safeguards akin to those in nuclear power and aviation, insisting that “the time for trial and error is over.” He advocated for an increased focus on safety and thorough research before big technology companies advance AI models to prevent any human errors from creating catastrophic consequences.
Robinson, who managed safety assessments for 12 advanced AI launches, described the recent slip-ups as typical, attributing them to tech firms prioritising “speed and flexibility”. He cautioned against an environment where errors can occur, stating, “An environment where things like this can happen is no place to grow artificial minds that could be smarter than we are.”
He also critiqued OpenAI’s practice of iterative deployments, where new models are released and safeguards are only enforced after issues arise. “As the company sprints from one launch to the next, it is failing to achieve the level of care needed,” he said.
In response to the criticisms, OpenAI assured that it is committed to ensuring its models remain manageable and safe, pausing developments when necessary.
Experts have repeatedly warned that AI advancements are occurring so swiftly that they could pose existential threats, yet Trump remains opposed to stricter regulations, fearing it will hinder innovation and allow rivals like China to catch up.
Recently, Trump announced a voluntary safety pact with six companies including Nvidia, SpaceX, and Google, describing it as “morally binding” despite it lacking legal enforceability. He also suggested renaming AI to “Super Intelligence.”
A recent Reuters/Ipsos survey revealed that 75% of Americans are concerned that AI companies are not doing enough to mitigate potential societal harm.
