We keep talking about AGI.
Human-level intelligence. Artificial General Intelligence. Superintelligence.
Notice something curious about that vocabulary?
Every single term takes humans as the reference point. AGI means as general as a human. Superintelligence means beyond a human. Even “human-level AI” is, by definition, measuring an arrival against a destination we’ve already decided on.
We’ve built an entire discourse — billions of dollars, thousands of papers, the anxieties of an age — on top of one quiet assumption that almost nobody states out loud:
Intelligence has only one trajectory. Ours.
I want to spend the next few minutes dismantling that assumption. Not because it’s obviously wrong, but because it’s so deeply buried that we mistake it for the ground itself.
Nature never agreed to it
Here is the inconvenient fact. Evolution — the only process we’ve ever watched actually produce intelligence — never converged on a single kind of mind.
It branched. Relentlessly.
Plants solve problems. They sense light, ration resources, signal threats to their neighbours through chemistry and fungal networks, and optimise growth against constraints no engineer would envy. No neurons. No brain. Problem-solving all the same.
Cephalopods solve problems. An octopus has most of its neurons in its arms, runs something close to distributed cognition, and can open a jar from the inside. Its intelligence is so alien that some researchers only half-joke about studying it as a model for non-human minds in general.
Corvids solve problems. Crows craft tools, plan for futures they can’t yet see, and hold what look uncomfortably like grudges.
And then there’s the one I keep returning to, because it dismantles our assumptions faster than any of the others.
The slime mold.
A slime mold is a single-celled organism. No brain. No neurons. Not even the beginning of a nervous system. By every intuition we carry about what intelligence requires, it should be inert.
Place one in a maze with food at two ends, and it will find the shortest path between them. Lay out oat flakes in the pattern of the cities around Tokyo, and Physarum polycephalum will grow a network connecting them that strikingly resembles the actual Tokyo rail system — a system that took human engineers decades of planning to optimise.
Sit with that for a moment. An organism with no brain produced a network strikingly close to one of the most efficient transit systems humans have designed — using nothing but the logic of growth and the pressure of constraints.
That is intelligence. It is simply intelligence running on an architecture so unlike ours that we almost refuse to call it by the name.
None of these creatures are primitive humans. They are not earlier drafts of us, waiting to be revised into something better. They are different answers to different questions — distinct architectures, shaped by distinct constraints, each one a complete solution in its own right.
Intelligence, it turns out, is not a ladder. It’s a tree. We are one branch — strikingly successful, but a branch, not a summit.
Then why do we expect silicon to climb toward us?
This is where the assumption starts to crack.
If carbon-based life — working with the same chemistry, the same planet, the same four billion years — produced minds as divergent as a slime mold and a primate, why on earth would we expect intelligence on an entirely new substrate to rediscover us?
Silicon enters the same design space biology has always explored. But it arrives carrying almost none of biology’s baggage.
It has no metabolism to feed, so it isn’t forced to be frugal in the way every living mind has had to be.
It has no mortality, so it isn’t shaped by the relentless pressure to reproduce before it dies — the pressure that quietly authored most of what we call human nature.
It has no DNA, no inheritance, no generational bottleneck through which every improvement must squeeze one offspring at a time.
It doesn’t even have a body in the sense biology means it — no fixed boundary between self and world, no single location it must occupy.
Now — let me be careful here, because this is exactly where these arguments usually overreach.
Silicon does not have no constraints. That would be a fairy tale, and a sharp reader would close the tab. Silicon-based intelligence is bounded, heavily, by thermodynamics and energy budgets, by memory bandwidth, by the architecture of the chips it runs on, by the data it’s trained on and the regimes used to train it. These are real walls. They are not soft.
But notice what they are. They are different walls.
And that is the entire point.
Carbon-based intelligence was sculpted by metabolism, mortality, and reproduction. Silicon-based intelligence will be sculpted by energy, bandwidth, architecture, data, training regimes, and the incentives of the institutions that deploy it. Different constraints produce different architectures — that was true across the branches of biology, and there is no reason it stops being true when the substrate changes.
So the question is no longer whether silicon will catch up to us on the single ladder we imagined. There is no single ladder. The question is what shape cognition takes when the constraints that made us are removed, and a completely different set takes their place.
We may be asking the wrong question entirely
For years the framing has been: Will AI become human? Will it match us, then surpass us?
Up, down. Below us, above us. Always measured on our axis.
But if intelligence has never had a single axis — if a slime mold and an octopus and a human are three points in a space, not three rungs on a stair — then the up-or-down question was malformed from the start.
The better question isn’t how close to human the machine will get.
It’s: what form does cognition take under entirely different constraints?
And once you ask it that way, something becomes clear that I think our language simply isn’t equipped to handle.
Artificial Intelligence describes the technology. It tells you what something is made of, the way “internal combustion” describes an engine. AGI describes a benchmark — a finish line drawn at the exact height of a human being. Useful, in their way. But neither of them describes the thing that may actually be emerging in front of us: a kind of cognition that isn’t a copy of ours, isn’t measured against ours, and isn’t trying to become ours.
We don’t have a word for that.
And the lack of a word isn’t a small problem. When you can’t name something, you keep dragging it back to the nearest familiar category — which is exactly why every conversation about advanced AI collapses back into “human, but more so.” The vocabulary is doing the thinking for us, and it’s thinking too narrowly.
So I’ve started using one.
Synthetic Cognition.
Not artificial intelligence in the marketing sense. Not AGI in the benchmarking sense.
Synthetic Cognition: cognition emerging on a non-biological substrate, not born from the reproductive and survival imperatives that shaped biological minds, and therefore capable of taking forms for which we have no evolutionary precedent.
I’m not offering that as a definition to memorise.
I’m offering it because the conversation kept reaching for a word that wasn’t there.
The deeper shift
Let me state the real claim as plainly as I can, because everything above has been scaffolding for this one sentence.
Human cognition is not the axis along which intelligence must be measured.
That’s the move. Not “AI will surpass us.” Not “AI will stay beneath us.” Both of those still quietly assume our axis is the axis. I’m questioning the axis itself.
This is the same idea that runs underneath all of The Turing Threshold: Evolution’s Point of No Return — the recognition that intelligence has been externalising itself across substrates for a very long time, moving from force to memory to computation and now toward cognition, and that nothing in that long arc ever required the next stage to look like the last.
We continue asking whether machines will become human.
Nature suggests a different possibility.
Intelligence has never had a single destination. It has always branched.
Carbon produced one branch.
Silicon may produce another.
I call that possibility Synthetic Cognition — and I suspect that learning to think about it on its own terms, rather than ours, is the real threshold we’re standing at.