Lesson 3 of 6

Toward AGI?

7 min read

Everyone argues about 'AGI'. But what would it actually take — and why can't smart, informed people agree on whether it's close?

A spectrum, not a switch

AGI — artificial general intelligence — usually means one system about as capable as a person across almost any task, not just one. Today's AI is narrow getting broad: a single model now handles many tasks, but still stumbles in ways no person would. Think of capability as a spectrum — narrow → broad → general — with today's frontier somewhere in the broad middle, and nobody sure exactly where.

AGI is the far end of a spectrum, and we're somewhere in the broad middle.

Why it's genuinely contested

One camp thinks it's close: scaling keeps working, and each model clears tasks the last one couldn't. The other thinks a real gap remains — reliability, genuine understanding, and learning from little data may need breakthroughs we don't have. Some go further and imagine superintelligence beyond human level; others doubt we'll reach even 'general'. Honest experts disagree by decades. Be sceptical of anyone naming a confident date — the honest answer is a wide range.

The timeline is a genuine range, not a known date — treat confident predictions with suspicion.

Watch for the word 'AGI' doing marketing work. There's no agreed test for it, so a company 'reaching AGI' often means whatever they choose it to mean.