In the previous article we left the beach expert wondering how far his plausibility could go. It was talked about for months under every sun umbrella, with one acronym.
AGI.
Under every sun umbrella someone names it with the certainty of someone reading a thermometer: “we’ll get there in two years,” “no, five,” “it’s already here and we don’t know it.” As if AGI were a line in the sand, and all you had to do was wait for the right wave.
It isn’t. Nobody knows how to draw that line. General compared to what? Intelligence measured how? Everyone means something different: fully human, superior, economically autonomous, conscious. Debating when we’ll reach AGI is debating when we’ll reach Neverland: the one from the song, which may or may not exist, and nobody can point to on a map.
The second problem is worse. We don’t even know how it comes about. The mechanism of this machine is to guess the most probable next piece: scaled to an unimaginable degree, it produces something that argues and reasons — and not even those who build it can quite explain why. Predicting when something undefined will arrive, generated by an unexplained process: it’s a prediction squared.
There’s a third trap too. Every time a machine surpasses a threshold we used to call intelligence — chess, Go, a bar exam — we say “ah, but that’s not real intelligence.” The goalposts move back deliberately as capability advances. And today AI already beats you on certain things — memory, speed, breadth — while falling behind on others; it can’t even count the letters in a word. It isn’t a tide rising toward the “human” line. It’s a jagged skyline: towering heights here, deep gaps there.
Before trusting the next confident date, ask yourself who’s giving it to you. The labs raising funds need AGI to be close. The prophets of doom need it to be close and terrible. It’s not a conspiracy — it’s just that a date on AGI says more about what the speaker is selling than about what will actually happen.
That said, I have no interest in playing the failed prophet either, and I don’t want to dismiss a prediction squared with one cubed. What’s certain is that capabilities are advancing — in a disruptive way — with or without a threshold to cross.
I think the winning move is to stop waiting for the threshold, stop building models on something we don’t yet know the shape of, and try to govern as well as possible what is already changing from next Monday. Which is already hard enough.