May 31, 2026
For some reason I’ve been thinking a lot about the nature of intelligence. (Maybe it’s watching the development of my infant sons; maybe it’s the obnoxious AI discourse; God, I hope it’s not the latter.) I’ve come to think of intelligence as a sort of universal simulator, capable of modeling more and more abstract, distant, and fuzzy things. This sounds sort of obvious on its face, but when you think through it, it’s startling how much it encompasses.
The first thing intelligence models is the physical world. It’s obvious, watching a toddler move around, that they’re learning about physics in real time. They stand, tumble, fall, throw, thrash, and roll; with each action and reaction, they learn something about gravity, momentum, or substance. Even without a vocabulary for these things, in time they start modeling the world so well they can move through it with grace and precision, like animals do.
(Note that this notion of world-modeling is well in line with current neuroscience. The theory of “predictive processing” holds that the brain’s whole job is to constantly predict the behavior the world around us, refining the model when something surprising happens, like a blurry patchwork of color resolving into the camouflage pattern of a predator. What you perceive isn’t the world; it’s the brain’s prediction of the world. Perception, writes the neuroscientist Chris Frith, is “a fantasy that coincides with reality”—more or less well depending on how well the brain has modeled the world around it.)
If the physical world were the extent of it, we would be intellectually outmatched by most animals. But we have another way of modeling the world: with abstract tools like physics, mathematics, textbooks, and calculators. This abstract method isn’t necessarily deeper—robots have a long way to go before they can run on muddy trails! But it’s certainly broader, capable of simulating things the thinker has never touched or seen, like outer space or the deep past. And that leap makes possible unbelievable, species-differentiating achievements, like putting a man on the moon. Yet the toddler and the physicist are doing basically the same thing: making and testing predictions about the physical world. Across species, across centuries, across disciplines, this is the shared base activity of intelligence.
Among humans, we tend not to specify what we mean by “intelligent”: Quick-thinking? Highly educated? The notion of accurate simulations—accurate world modeling—unifies both. Someone gifted with an extraordinarily capable brain can accomplish many things. Someone born with a normal brain, who studies a topic and picks up new tools, can accomplish similar things. (Consider two computers: the first far more powerful than the second, but the second given access to specialized tools and datasets, such that their performance on a given task is equivalent.) If these two people can, leaning on everything at their disposal, model equivalently complex things, we might as well call them equivalently intelligent. That is, the distinction between raw IQ and acquired knowledge isn’t meaningful; both are implementation details behind the general capability of simulation.
(Another distinction that proves meaningless: book learning versus common sense. While these are distinct things, they’re valuable in exactly the same way: insofar as they help you model the world. Certainly you need both.)
Now it gets interesting. One of the other common distinctions you hear about is IQ versus EQ: emotional intelligence. These, too, don’t seem to me to be different things—EQ is nothing more than the extension of intelligence (the universal simulator) to the point where it’s able to simulate other people. So much of emotional intelligence boils down to predicting how things will come across, where other people are coming from, what other people are really looking for. These all hinge on the ability to model other people at a deep level. IQ without EQ strikes me as an profoundly limited form of intelligence. They’re the same thing—more or less accurate predictions about the world—applied to different things in it.
If you can model other people, how about modeling that all-important person: yourself? Understanding one’s strengths and weaknesses, one’s foibles, one’s blind spots, one’s mistakes, one’s deepest desires and fears—this sort of self-knowledge is usually thought of as something like maturity, not mere intelligence. Again, though, we see the same pattern as EQ: a simulator is more capable if it can accurately simulate itself. The traits associated with self-knowledge reflect intelligence that’s capable of accurately modeling the thinker.
We’ve extended our notion of intelligence beyond modeling the physical world, beyond the dichotomy of inherent intelligence versus acquired education, beyond the dichotomy of IQ versus EQ, beyond self-knowledge. How much farther does it go?
Well, if EQ is a sort of meta-intelligence (intelligence simulating other intelligence), and if self-knowledge is a sort of meta-meta-intelligence (intelligence simulating itself), then here is meta-meta-meta-intelligence: art. Great art in every medium is itself a simulation of the world; it can model things as varied and subtle as the human experience, the tragedy and beauty of the universe, the irony and strangeness of it all. This extended vocabulary represents another step-change in breadth, in simulation capacity, just like abstract over embodied physics. But the really interesting thing about art is that it lets us trigger these experiences in others. It’s like software that doesn’t just simulate the most varied and subtle aspects of life, but programs other computers (other people) to simulate it themselves. This sort of world-modeling goes far beyond the theory of mind implied by basic EQ. It doesn’t just predict the actions and feelings of other people; it models others so thoroughly that it can then give them information that, rather than communicating mere facts about the world, instead prompts them to simulate the world in a way that can’t be communicated directly. Intelligence simulating other intelligence simulating the world.
(Part of what’s cool about this is the way it challenges the old trope of STEM vs. the arts, where STEM is usually thought of as more demanding and rigorous. Yes, the direct and measurable world-modeling of STEM is key to our species’ accomplishments, but the simulation involved in the arts strikes me as more sophisticated and uniquely human. At any rate, both are squarely within the realm of intelligence as I’m defining it here: broad and accurate simulation.)
Let’s continue extending the notion. A simulator’s job is to model the world, not judge it—you would consider it a flaw of a simulation engine to resent the world, to hold grudges, to engage in magical thinking, to wish things were otherwise. An accurate simulation shouldn’t just include as much detail as possible; it should exclude anything that’s not part of the world. And “should be” is not part of the world. Therefore, this version of intelligence accounts for something like equanimity, wisdom, or stoicism. Yes, we all sometimes think this way—my model of human consciousness does not wish we were robots! But there’s a reason so many wisdom traditions emphasize acceptance, grace, and nonjudgment, and I think it’s correct to locate these lessons in the realm of accurate world-modeling, and therefore intelligence.
We’re farther and farther from crude IQ tests, and closer and closer to something that feels like character. But we can go one step farther: this model of intelligence goes so far as to encompass much of ethics. If part of simulation is modeling others, it follows that a stronger simulator can model more and more distinct others. It can build a rich understanding not just of people who are very much like the thinker, but of people who are very much unlike the thinker—or who are perhaps not people at all, or who are very distant in space and time. Peter Singer talks about expanding our moral “circle of concern”: from self to family, to city, to nation, to mankind, to all sentient life, to future generations. Or: Rawls asks us to design societies as if, behind a “veil of ignorance,” we couldn’t know our circumstances (gender, status, ethnicity, etc.) until we were spawned within the society we’d designed. A huge part of philosophy and ethics hinges on either simulating other subjects (as in Singer’s circle of concern) or simulating other circumstances (as in Rawls’s veil of ignorance). Therefore, ethics, too, falls under the umbrella of intelligence as universal simulator. I’m not just making the obvious point that some sort of intelligence is a precondition for ethics—I’m proposing that we are intelligent only insofar as we can model other beings, and that a focus on ethics necessarily follows. It’s no coincidence that so many religions locate in their god both infinite intelligence and infinite compassion; they’re the same thing.
What does all this give us? It gives us an improvement on the schoolteacher’s lukewarm line about “many kinds of intelligence”: no, there’s just one kind, applied to many things in the world. It gives us a clarified understanding of that cliché about “education being a life’s work”: In academia, where we go deep in so many topics that we never revisit after graduation, that may not ring true. But if we think of intelligence as encompassing not just textbooks but social skills, empathy, maturity, art, equanimity, ethics—becoming smart, becoming good, becoming wise—well, clearly there’s enough there to keep growing and growing for a lifetime, with whatever faculties we have and in whatever circumstances we find ourselves. There’s something empowering about that. And (AI was bound to come up eventually) it gives us reason to doubt that LLMs will ever produce truly flexible, open-ended, human-style intelligence. Trained on words, they can certainly simulate more words, and things made of words—but I can’t see how they would evolve to simulate embodied experience, or the subjective experience of others, or the intangible experience of art, or all the other things in the world that aren’t made of words. Our bodies, our senses, and our selves give us—for now at least—far more breadth as simulators.
One last thought. It would be easy to read this and conclude that I’m describing a life purely of the mind, predicting everything and everyone with such perfection that the world is abstracted away. The opposite is true. If the point is accurate world-modeling, it would be inaccurate say “these representations are all that matters.” No—real accuracy means simulating the world, and realizing that it’s all real, in addition to and beyond our model of it. Understanding, as a child, that other people are real. That your parents are real people. That strangers are real, as real as you. That future people are real, that the past is real. That animals (and all their pleasure or suffering) are real. I’m not sure if this is part of simulation or the limit of simulation, but it’s a sort of humility—an insistence on seeing the map as map and not territory. The job of the universal simulator is to make the highest-fidelity map possible, but—critically—to retain the capacity to make contact with the real world wherever we can, and tell the difference. (Coming to terms with the reality of everything turns out to be an unintended and emergent theme of this blog.)