9 May 2026
How Can AI Replace Something We Don't Even Understand? Intelligence.
TLDR: The answer is primitives.
One question has followed me through most of my adult life: what is intelligence, really? Not as an abstract concept, but as something lived and felt, especially in human systems, and now increasingly in machines.
My first real confrontation with this question came in high school in Morocco. I ended up, somewhat randomly, in what everyone around me considered the elite track: Sciences Math. The implicit social contract was clear: only the smart ones end up here. I never saw myself as smart in that nerdy, effortless way. So I spent most of that time watching. Observing the people around me with a quiet, almost anthropological curiosity. How were they processing this level of mathematical complexity so naturally? What was different about the way their minds worked? What did they have that I didn’t?
The same question followed me into classes prépa, especially in my second year when I moved to a more competitive school. That’s where I finally met what I can only describe as a math genius. Arthur. I still remember this guy. Clearly autistic, with absolutely zero social skills, but when it came to maths and physics he made it look completely effortless. Always came up with solutions no one else had thought of, always the most elegant ones. I kept asking him how.
In the same class we had another Arthur. Let’s call him the Perfect Arthur. Phenomenal across all subjects, not just one or two. He also played competitive tennis, worked out, had a great social life, a beautiful girlfriend, always well slept, always well dressed. He managed to maintain all of this while excelling academically, and anyone who went through prépa knows how brutal that environment is.
After the exams, no one was surprised. Autistic Arthur ended up at ENS Ulm, where maths and physics carry the highest coefficients. Perfect Arthur ended up at École Polytechnique (X), which at the time we considered slightly easier to get into than Ulm. I followed both their careers for the eight years that followed. And again, no surprise: Perfect Arthur thrived in every conventional sense, prestige, income, real-world impact. Autistic Arthur ended up doing what he loves, which is research. I came across his name on a couple of low-profile papers, and that was about it.
So which one of these two is smarter? Which one has the higher degree of intelligence?
Here’s the argument I hate: “they both have different types of intelligence, and all humans are intelligent in their own way.” I refuse to accept that. It’s a soft, feel-good answer designed to make everyone comfortable without actually saying anything. Even if we accept that multiple types of intelligence exist, are they comparable? Are they measurable? Which type, or which combination, adapts best to our current capitalist environment? And above all, which one should AI systems be evaluated on?
Where do I even stand in all this? By the time I met the two Arthurs, I wasn’t doing badly in maths and physics either. But how? Two years earlier I was genuinely struggling. My neurological structure is the same one I was born with. So it had to be something related to experience and work. The more I trained on these subjects, the better I got. Does that mean we are not born intelligent but become it?
I can’t fully accept that either. Because if intelligence is just the product of training, then it’s just about ingesting more experience, which means more data. And a model with hundreds of billions of parameters ingests more data than any human could accumulate in a lifetime, yet I wouldn’t call those systems intelligent today.
François Chollet, a researcher at Google, tackled exactly this in a 2019 paper where he proposed the ARC benchmark to measure the intelligence of any system, including AI. His definition centers on the skill acquisition process, specifically its speed and efficiency. How fast can you learn something new with minimal data? By that standard, if it took me two years to learn differential equations, Perfect Arthur one year, and Autistic Arthur one day, then Autistic Arthur is objectively smarter than either of us.
But here’s my problem with that definition. Isn’t skill acquisition domain-specific? If your mind was wired toward logical and mathematical thinking from childhood because your early experiences rewarded that kind of thinking, you’re not universally more efficient at learning. You’re just better optimized for that specific domain. Perfect Arthur’s mind was likely wired for social problem-solving. His capacity to navigate corporate environments, read people, and execute in complex organizations explains the career gap just as well as raw intelligence does.
And from this same angle, can AI ever truly be smarter than us? A model trained on trillions of tokens over a few months looks impressive. But it trains on infinitely more experience than any human accumulates in a lifetime. It’s not smart. It just sees an enormous number of patterns, recognizes relationships between them, and reuses them convincingly. That’s not intelligence. That’s compression.
And I think this is the key insight, my actual contribution to this question.
If we agree that being smart means learning new things fast, solving problems in unfamiliar environments efficiently, and adapting to the world around you, we land somewhere close to what Naval Ravikant once put simply: “You are smart when you get from life what you want.” Which breaks into two tasks: understanding yourself well enough to choose what you actually want, and understanding the world well enough to find the most efficient path to get it. I’d argue the first is harder than the second, but that’s a separate conversation.
Agreeing on this definition leads to a more practical question: how do you actually develop this capacity? How do you increase your speed at picking up new skills and solving unfamiliar problems? Is it a natural gift? A product of childhood experiences you can’t reverse-engineer? An accident of circumstance?
No. I refuse that framing. I have no patience for intellectual dead ends that produce nothing actionable. An idea, an analysis, a thesis has no value to me if it doesn’t benefit someone directly and measurably.
My answer is primitives.
AWS understood this 20 years ago. In software engineering, the way to own a market is to own the primitives, the foundational services everything else gets built on top of. Storage (S3), compute (EC2), intelligence (Bedrock). Everyone building something complex needs these. Control the base layer and you control the ecosystem.
The same principle shows up in mathematics. I loved maths from the bottom of my heart. It was the only domain where everything was provably true, rigorously established, with no room for convenient approximations. Unlike physics or chemistry, where we build models on incomplete experimental data and call them good enough, in mathematics one single counterexample destroys an entire theory. That felt like the purest form of truth available to us.
Then I discovered axioms. Things we cannot prove. Things we simply have to accept in order to build anything on top of them. And then I encountered Gödel’s incompleteness theorems (1931): “Any consistent axiomatic system powerful enough to describe basic arithmetic contains true statements that cannot be proven within that system.”
Mathematical truth will always be larger than mathematical proof.
I almost cried when I read that. Does it mean it was all a lie? That there are things we will never be able to prove? My intellectual ego still hasn’t fully made peace with this. Maybe it was part of why I eventually chose to chase money instead of pursuing mathematics as a career.
But the lesson I took from it is more useful than the grief. Even the most rigorous system in human history runs on a set of unprovable base assumptions. And once you accept that, you realize the entire world operates the same way.
Finance feels complex? Trace every concept back to its source. Ask why at every step. Very quickly you land in basic ideas like buying and selling. Then: why buy and sell? Why financial systems? Why capitalism? Why globalization? Why comfort? Why freedom? Every complex system, when you follow it far enough, traces back to a small number of human-invented concepts, invented morality, invented priorities.
The same applies to politics. The left will save us. No, it’s the right, you idiot. How will the left fund their programs? The rich will leave the country if the far left takes power. The far right is just racism with better branding. Every time I hear someone describe themselves as purely leftist or purely right-wing, a red flag goes up in my head. Have they actually traced these positions back to the primitives?
Take a left-wing proposal like universal basic income. Why? Because everyone deserves a dignified life. Why hasn’t it happened? Because the money has to come from somewhere. From where? Capital concentrated at the top through compounding returns. Why did that happen? Because we built a system that incentivizes innovation: if you create something people want, you can sell it, and competition pushes quality up. Why optimize for that? For comfort. For better human experience.
So being a leftist, at its core, isn’t really about universal salary. It’s about a foundational moral axiom: equality. And a trade-off with another axiom: progress through individual incentive. The far right? At its core, earned hierarchy, individual power, security. What if you want both earned hierarchy and universalism at the same time? Which box do you vote into then?
Most people never ask these questions. They inherit a political identity the same way they inherit a religion, through environment, emotion, and social belonging, not through first-principles thinking.
The people who have actually done this exercise, who have sat in real solitude and traced everything back to its source, they feel different when you talk to them. Sharp. Clear. Almost impossible to manipulate through emotional headlines, political agendas, or influencers recycling outrage for clicks.
And here’s what I’ve noticed from my own experience: the capacity to pick up new skills and solve unfamiliar problems increases the more you practice this kind of deep, primitives-first thinking. Because at the end of the day, there are very few true axioms underlying any human system. Once you know them, nothing is fully new. Every complex structure you encounter is just those same primitives reassembled in a different order.
That’s what intelligence actually looks like to me. Not raw processing power. Not the ability to memorize. The ability to see through complexity to the base layer, and build from there.
AI can simulate that. But can it actually do it? That’s still an open question, for the next article.
Z
This article is entirely human-written. AI was used to fact-check details and proofread.