Back in part 1 we split AI into three tiers: narrow AI (real, everywhere), general AI or AGI (doesn’t exist), and super AI (pure theory). Eight articles later, you now know exactly what narrow AI is actually doing under the hood. So — are we any closer to the general kind? Depends entirely who you ask.
Quick recap: three tiers, one big gap
Everything we've covered — one skill, done brilliantly.
Human-level reasoning across any task. Not built yet.
Beyond human, everywhere. Not on the table yet.
Why the smartest people disagree
A genuinely large part of the disagreement isn’t even about the technology — it’s that nobody agrees on the definition. “General intelligence” doesn’t have one crisp, universally accepted test the way “can this model translate French” does. Different researchers use different bars, so “how close are we” gets a different answer depending on whose ruler you’re using.
the two camps
"Not even close"
Today's models are pattern-matchers with no real understanding, common sense, or persistent memory of the world. Scaling up the same approach may hit a wall long before anything like general intelligence.
"Sooner than people think"
Capabilities have jumped sharply in short spans of time before, and today's systems already handle tasks that seemed years away not long ago — so extrapolating that trend forward gets you somewhere fast.
Both camps include serious, credentialed researchers — this isn’t experts versus cranks, it’s genuine, live disagreement at the top of the field.
why it's not just an academic argument
Why the timeline actually matters
How close AGI is affects real, present-day decisions — how much money gets poured into AI safety research right now, what regulations governments write today, how companies plan their next five years. That’s why this debate keeps making headlines instead of staying a niche academic argument: the “when” genuinely shapes the “what do we do about it.”