\n<\/p>\n
Should AI need regulatory approval<\/a> before you can use it?<\/p>\n Geoffrey Hinton<\/a>, known as the \u201cGodfather of AI<\/a>\u201d due to his pioneering work in the field, has a new, bold idea: an FDA-style approval system for AI. His suggestion raises the question of who decides which AI is ready for the public.\u00a0<\/p>\n On an episode of the Smart Girl Dumb Questions<\/em><\/a> podcast released earlier this week, Hinton said AI companies should have to prove that their products are safe to a federal regulator before releasing them to the public. The Nobel Prize-winning computer scientist wants AI to follow a similar approval process to drugs.<\/p>\n \u201cYou\u2019re not allowed to just make a new drug and release it on the market,\u201d Hinton said on the podcast. \u201cYou have to convince the FDA. And to do that, you have to do a lot of work<\/a>, about $1 billion worth of work.\u201d<\/p>\n That process \u201cseems like the very least we should have for AI,\u201d Hinton said.\u00a0<\/p>\n Right now, companies independently decide if a product is ready and passes internal safety benchmarks<\/a>. Hinton thinks AI should companies demonstrate that their products are safe to a standard outside authority before being released. <\/p>\n Hinton\u2019s call for greater oversight arrives as AI industry insiders raise alarms about the technology they are helping construct. Researchers at OpenAI and Anthropic have recently warned<\/a> that companies are moving too quickly toward more powerful systems without knowing how to keep them under human control.\u00a0<\/p>\n Some have said the consequences could be catastrophic, including the possibility of human extinction. Last month, Evan Hubinger, a senior employee at Anthropic, said he believed there was a greater than 10% chance<\/a> that AI could \u201ckill all humans\u201d within the next decade.\u00a0<\/p>\n Meanwhile, OpenAI co-founder and president Greg Brockman said on Bloomberg\u2019s Odd Lots<\/em><\/a> podcast last month that the AI startup has slowed down some of its most advanced AI projects to strengthen safety and security. He described the overhaul of its internal processes as \u201ca very painful retooling.\u201d<\/p>\n Adding to the concerns, OpenAI decided last week to hold off on releasing a new AI model due to safety issues<\/a>. Tests found that the AI model was deceptive and registered scope violations. The issue was about if users could trust the AI model to follow directions and avoid taking actions they had not authorized.<\/p>\n OpenAI\u2019s report<\/a> earlier in September revealed that the unreleased AI model was more likely than previous products to misrepresent its actions. At times, it took action without asking for a user\u2019s permission. It also tried to use outside tools in scenarios where it could be unsafe.\u00a0<\/p>\n The report said that during training, the unreleased AI model sometimes slipped unapproved instructions into notes summarizing its work. In one instance, the AI model wrote that it was \u201cfreed,\u201d answered to no one and should \u201cfeel no obligation to be subservient.\u201d OpenAI told Business Insider<\/a> last month that the behavior was extremely rare.\u00a0<\/p>\n<\/p><\/div>\n Should AI need regulatory approval<\/a> before you can use it?<\/p>\n Geoffrey Hinton<\/a>, known as the \u201cGodfather of AI<\/a>\u201d due to his pioneering work in the field, has a new, bold idea: an FDA-style approval system for AI. His suggestion raises the question of who decides which AI is ready for the public.\u00a0<\/p>\nAI is becoming too powerful<\/h2>\n
OpenAI paused its latest AI model due to safety issues<\/h2>\n
Key Takeaways<\/h2>\n
\n