Historians Won’t Be Able To Agree On When Exactly We Reached AGI: Sam Altman

Most people working in AI seem to be focused on achieving Artificial General Intelligence, and have their own predictions on when it’ll be reached, but one of the most prominent voices in the field say that “reaching AGI” might not be the milestone that many are expecting.

OpenAI CEO Sam Altman says that it’s unlikely that reaching AGI will be an event that takes place on a definite date. “I think we’re like in this period where it’s going to feel very blurry for a while,” he said in an interview. “(People will wonder if) is this AGI yet, or is this not AGI, or it’s just going to be this smooth exponential. And probably most people looking back in history won’t agree when that milestone was hit. And we’ll just realize it was like a silly thing,” he added.

“Even (for) the Turing test, which I thought always was like this very clear milestone, there was this like fuzzy period. It kind of went whooshing by and no one cared. But I think the right framework (for AGI) is it’s just this one exponential. That said, if we can make an AI system that is materially better at all of OpenAI at doing AI research, that does feel to me like some sort of important discontinuity,” he added.

“It’s probably still wrong to think about it that way. It probably still is the smooth exponential curve. That feels like a real milestone,” he added.

Altman seemed to be comparing reaching AGI to how humanity behaved when the Turing test was passed. For decades, the Turing Test was supposed to be the ultimate measure of computer intelligence — if a computer could pass off as a human while having a conversation without the human knowing, the Turning Test would have had been deemed to have passed. But since late 2022, when OpenAI released ChatGPT, computers have been steadily getting better at talking to humans, and now it’s clear that the Turing Test has been conclusively breached. But it’s impossible to pin-point an exact date, or through which model it was passed — model performance kept improving gradually until it became apparent to everyone that the Turning Test had been conclusively cleared.

Altman seemed to be saying that AGI might follow a similar pattern. It helps that AGI doesn’t have a precise definition, so it’ll be harder to build consensus on when exactly it’s reached. But Altman seems to be saying that AI models will keep improving their capabilities until it’s apparent that they’re better than humans at most tasks, and at that point everyone will agree that AGI has arrived, but there will be no definite date as to when the milestone was reached. But Altman’s ambivalence about a definite date for AGI is in contrast to the predictions he’d been making — he’s previously said that AGI will arrive in 2025, but now seems to be saying that AGI will arrive, but it’ll be hard to tell when exactly it’s here.