Signs of AGI

It has arrived when we say it has arrived

Welcome to this update from New World Same Humans, a newsletter on trends, technology, and society by David Mattin.

If you’re reading this and haven’t yet subscribed, join 28,000+ curious souls on a journey to build a better future 🚀🔮

---

OpenAI released its frontier reasoning model, o3, inside ChatGPT last week. The superstar economist Tyler Cowen responded with this:

Just how much smarter was I expecting AGI to be? It’s a valid question.

Others are pointing out that as these AI models get more intelligent, we continuously move the goalposts. This description of AGI is from a book published in 2019:

Okay, o3 can’t quite bake you a cake. But it can give you a great recipe. And it has everything else in that list covered.

Every time a new model arrives, I find myself asking it a set of questions. Most are playful philosophical puzzles; they’re intended to give me a sense of the model’s cognitive flexibility, and understanding of context and nuance. Here is one of my favourites:

And here is the response from o3:

It’s clear that o3 can articulate a full explanation of the joke and its meta-linguistic nature. What surprised me was the fluency, the almost poetic nature, of its response: ‘…the sign tells you to treat the sign as a sign. It’s a playful recursion, like a Möbius strip in sentence form.’

It’s never occurred to me to compare this sign to a Möbius strip. But it’s a wonderful comparison. This response is, I think, the first time I’ve had a back and forth with an AI and felt myself to be in the presence of another entity that can write imaginatively.

o3 is also able accurately to deploy a deep knowledge of semiotics, drawing on Charles Peirce and Derrida, to deepen its explanation.

Overall, my guess is that o3’s explanation is better than anything 90% of philosophy graduates could deliver given the same prompt. I’m sure all of them would understand the joke. But I bet most wouldn’t write such an accomplished, nuanced take.

This, by the way, is what GPT-4 makes of the same puzzle:

It understands the nature of the joke; a loop of meaning. But only just. And the whole articulation is just bland and lifeless in comparison to o3’s response.

This does feel like a crossing over moment. Perhaps Tyler Cowan is right. Perhaps AGI arrived in the world on the 16 April 2025.

Others say that o3 does not constitute AGI; some argue the threshold is still decades away. The argument itself is instructive. It helps us to remember that AGI has no clear, quantitative, or universally agreed definition. Rather, it is essentially a social consensus term; AGI will have arrived when enough of us say that is has.

We deal with human intelligence in the same way. We talk to one another, and form opinions on who is intelligent and who is not. Over time those views tend to coalesce around a rough consensus: Fred is smart, but his brother George is not. But there can still be valid disagreement on these matters: actually, I think George is smarter than Fred.

Sure, there are IQ tests. But as Stephen Hawking famously pointed out, ‘people who boast about their IQ are losers’. I like to think that his intention, here, was to remind us that intelligence is too intangible, too multi-faceted, to be captured by a single number. And once we we get more than four or five standard deviations about the human average — which by definition is 100 — then the measure probably breaks down into incoherence. All this talk of AIs with an IQ of 4,000 is, I suspect, pretty meaningless.

So we’ll deal with AGI in the way we deal with one another. At some point most people will accept that AGI has arrived. Some will still say it has not. Some people will probably never accept that AGI has arrived, just the way some people will never accept that Fred is smarter than George.

Meanwhile, the world and our place in it will be transformed.

As for me, I’m still making up my mind. Part of me thinks that if o3 isn’t AGI, it’s pretty damn close. Another part thinks that for us to talk about true artificial general intelligence, we have to have a model that is embodied — so, inside a robot — and able to do an extremely wide array of physical tasks, too. The ability to figure out how load the dishwasher, drive a car, or draw a picture of the oak tree that is in front of you now: all this is also intelligence.

But if you were to approach me and argue with deep conviction that o3 is AGI, I wouldn’t put up that much of an argument. In this way, consensus is forming.

I’ll be back next week. Until then, be well,

David.

This was #14 in the series Postcards from the New World, from NWSH.