No Intelligence Without Illusion

Not a member? Read this story for free here.

(This image was generated using AI technology.)

The same cognitive architecture that produces human intelligence also produces human delusion. We are now scaling both.

A while ago I came across a statement by Yuval Noah Harari that stayed with me. I’ve followed his work closely over the years and I find him consistently precise on the things that matter. This one stopped me:

“It is very important to understand intelligence is not wisdom. What we know from history is the most intelligent entities on the planet are also the most delusional entities on the planet. We humans are by far more intelligent than chimpanzees or elephants or dogs, and we believe in delusions that no chimpanzee or dog will ever believe. I come from a region of the world where people are killing each other by the thousands believing, for example, that if you kill people from the other tribe, after you die, you will go to heaven and receive infinite joy for ever and ever. No chimpanzee will ever believe such a thing. But humans, many humans, many powerful humans, many intelligent humans believe that. And as we are now creating superintelligence, we should take into account they could be super delusional.”

— Yuval Noah Harari

What kept pulling me back was the structural claim: that intelligence and delusion are not opposites. That more of one tends to produce more of the other. That this is a feature of how intelligence works, not a failure of it.

Intelligence Produces Delusion

The chimpanzee doesn’t believe in paradise. Abstract thought lies outside its range. It sees what’s there: the fruit, the threat, the social hierarchy in front of it.

We see further. We also believe further, in things that can’t be touched, verified, or pointed to. Religion, money, national identity, legal personhood. None of these exist as physical objects. All of them organize the behavior of billions of people. A corporation has no body. A national border has no smell. These fictions shape nearly every significant decision humans make.

Harari identified this as the key to human dominance: the capacity for shared fiction. Large-scale cooperation depends on it. No band of chimpanzees ever organized a market, a military campaign, or a central bank.

The same capacity enables shared delusion. Fiction and delusion share one source.

The Architecture Is the Same

Delusion would be more manageable if it were a malfunction. Something introduced by bad actors, correctable with better education, solvable with sufficient rationality. The historical record says otherwise.

Intelligence and illusion run on the same cognitive apparatus.

Language makes truth possible. It makes lying possible through exactly the same mechanism. There is no grammar that blocks false sentences. The rules that generate “the experiment was replicated” also generate “the experiment was fabricated.” The system has no internal way to distinguish them.

Abstraction enables science. It enables ideology through the same process. The capacity to form a model of the world, to simplify, categorize, and predict, is also the capacity to form a model that is wrong, self-consistent, and resistant to correction. The more sophisticated the model, the more it can elaborate its own errors into something that looks like a coherent worldview.

Delusion is the shadow cast by abstraction itself. Higher intelligence increases the elaborateness of what can be believed.

Language Models Are Not an Exception

LLMs hallucinate. That observation has become routine. The explanation is stated clearly less often.

The hallucination follows from the competence. The model that constructs a coherent argument is the same model that constructs a coherent fabrication, with identical weights and the same optimization target. I examined the mechanics of this in an earlier piece on hallucination. More training can reduce this. Whether it can eliminate it structurally is genuinely open.

The system learns to produce text that is contextually plausible and internally coherent. Truth correlates with those properties. It is not defined by them. The model has no way to feel that difference from the inside.

Neither, incidentally, do we.

What “Super Delusional” Actually Means

Harari’s phrase could be read as a sci-fi warning. I think it’s more useful as a structural description.

Human delusion has natural limits.

It travels at the speed of language. It requires social uptake. It meets resistance from competing narratives, from people who saw something different, from institutions built specifically to push back. These friction points are imperfect and often too slow. But they exist.

Superintelligent systems remove them.

A sufficiently capable model produces narratives that are more internally consistent than most true accounts, emotionally calibrated to specific audiences, and generated faster than any correction mechanism can follow. The delusion doesn’t announce itself as delusion. It arrives fluent, confident, and tailored.

This is already visible at the current scale of AI. What scales alongside capability is not just coherence and speed. It’s persuasiveness. The gap between a convincing true claim and a convincing false one narrows as the system improves.

We’re Building It Anyway

Harari said this clearly. We heard it. And we are building superintelligence anyway.

That’s worth sitting with, not as a moral failure but as a recognizable pattern. We have done this with every sufficiently powerful tool. Fire, the printing press, nuclear fission. The risks were sensed in rough outline beforehand, sometimes loudly. The specific consequences — which fears materialized, which didn’t, which ones nobody saw coming — only became clear afterward. None of these technologies were stopped by the warnings.

What tends to happen instead is that the tool gets built, the damage becomes specific, and then the correction systems follow. Slowly, incompletely, but persistently. The same species that has organized mass violence around shared fictions also invented the institutions that prosecute it, document it, and argue about its causes for centuries afterward.

Harari’s deeper point, the one I keep returning to, is that intelligence and the capacity for self-correction come from the same source.

We are the species most prone to elaborate delusion and the species most capable of recognizing and correcting it afterward.

Those two facts about us are not in tension.

We are now building something that will inherit both.

I write about AI for leaders and professionals who want to understand what’s actually changing. Follow along or explore my full catalog in the **AI Writing Compass**.

This story is published on Generative AI. Connect with us on LinkedIn and follow Zeniteq to stay in the loop with the latest AI stories.

Subscribe to our newsletter and YouTube channel to stay updated with the latest news and updates on generative AI. Let’s shape the future of AI together!