Watch: Superintelligence (SI)
A hypothetical system whose cognitive performance greatly exceeds the best human minds in nearly every domain. No system has reached it, and there is no agreed test for when one would.
Transcript
Imagine a machine that does not just match human intelligence, but far exceeds the very best of us in almost every field, from science and strategy to social reasoning. This is superintelligence. It represents the next step beyond artificial general intelligence, or AGI, which only aims for human-level performance. Right now, no system has reached either milestone.
The idea goes back to 1965, when mathematician I. J. Good envisioned an intelligence explosion where machines would design even better machines, leaving humans far behind. Today, the most common definition comes from philosopher Nick Bostrom, who described superintelligence as any intellect that greatly exceeds human cognitive performance in virtually all domains. This could happen in a few ways: a system that thinks just like us but at lightning speed, a collective network of smaller minds working together, or a quality superintelligence that is simply smarter in kind.
We already have narrow superhuman performance. Machines can defeat chess champions, master the game of Go, and predict complex protein structures. But these systems are highly specialized. To be truly superintelligent, a system must hold that kind of lead in almost every domain at the same time.
Though no lab has built such a system yet, the race is on. Major players like Meta, OpenAI, and the startup Safe Superintelligence are all actively working toward this goal. For now, superintelligence remains a future destination, but it is the ultimate destination the AI world is chasing.