Thinking too logically can actually hold you back | Dan Shipper

ELI5/TLDR

For 2,000 years the West has believed you only truly know something if you can spell it out in clear rules and definitions. That belief built physics, computers, and vaccines, but it quietly taught us to distrust our gut. Dan Shipper argues that intuition, the stuff you can’t put into words, is actually the foundation everything else sits on. And the proof showed up in an odd place: AI only started working once we stopped trying to write down the rules and built machines that learn by feel instead.

The Full Story

The idea that runs the modern world

Shipper starts with a word most people have absorbed without ever naming: rationalism. His definition is simple.

Rationalism is really the idea that in order to truly know something, you have to be able to describe it explicitly.

Think of it like this. If you can’t reduce something to “if X is true, then Y will happen,” then, on this view, you don’t really know it. You just have a hunch. This way of seeing the world, he says, is “built into the way that you see the world” whether you’ve heard the term or not. Computers, weather forecasts, the way therapy tries to talk you through your feelings, even pop business books promising “The Five Laws of Power” are all downstream of it.

Where it came from: Socrates picks a fight

The origin story runs through a Plato dialogue called Protagoras. Picture a debate. On one side, Protagoras, a sophist (the root of “sophistry,” meaning someone who sounds convincing but is, in Shipper’s words, “actually full of shit”). On the other, Socrates. The question on the table: can excellence be taught?

Protagoras answers with a sprawling myth, full of stories about how humans gained the capacity to be good. Socrates wants none of it. He wants a definition: tell me exactly what excellence is, what it isn’t, and its parts. Protagoras can’t produce one without contradicting himself, and the implied verdict is brutal: if you can’t define it, you don’t know it.

That moment, Shipper argues, set Western thought on a 2,000-year path. Descartes carried it into philosophy, Newton and Galileo into science, all reinforcing the same rule: real knowledge is explicit, ideally written in mathematics. And it worked spectacularly, giving us smartphones, rockets, and electricity.

Where it stalls

The trouble shows up the moment you point this method at messier territory. Psychology, economics, and neuroscience all tried to copy physics, reducing human behavior to universal laws. Newton found laws that held everywhere. Psychology, after a century of trying, is instead stuck in a “gigantic replication crisis,” where findings don’t hold up when retested. Yet, Shipper notes, the field can’t quit the approach “because we have no better alternative.”

The clearest example: AI

The sharpest demonstration is artificial intelligence. The original 1950s plan, now called symbolic AI, was pure rationalism: reduce thinking to symbols and explicit rules, and chain them together. Pioneers Herbert Simon and Allen Newell built a “general problem solver,” at first running it by hand on paper (computers were too expensive), even roping in family to play the role of the machine. It cracked simple puzzles beautifully, then collapsed as problems grew, because the number of possible solutions to search exploded.

Shipper’s example is the email inbox. Suppose you write a rule: “emergency” in the subject means top of the inbox. Spammers learn to type “emergency.” So you add: only from coworkers. But the computer doesn’t know what a coworker is, so you define that too. Then a pushy colleague abuses it, so you add another exception. And on it goes.

If you wanna, for example, define what an important email is, you have to define pretty much everything about the world. You have to create a world full of definitions.

That project, making the whole world explicit, simply broke. Too brittle, too many exceptions.

The fix: learning by example instead of by rule

The alternative is the neural network, loosely inspired by the brain: layers of artificial neurons that learn patterns from thousands of examples rather than from hand-written rules. Show it enough emails, let it guess, correct its wrong guesses, and it gradually learns to sort them, without anyone ever defining “important.”

The catch, and the whole point, is that the rules it learns are inexplicit. You can’t open it up and read the list, any more than you could find, under a microscope, the rule your brain uses to recognize a cat. Language models are this same trick aimed at text: fed the internet, asked endlessly to predict the next word, they absorb “many thousands of partially fitting rules” that live nowhere you can point to.

The full circle

This, Shipper says, looks a lot like human intuition, also built from thousands of hours of direct experience rather than stated rules. He notes we tend to model the mind on whatever tool is dominant: Freud reached for the steam engine; the 20th century reached for the computer and decided we should be logical machines. That metaphor made our own intuition invisible to us.

And so we land back at Protagoras, who was right all along. He taught through stories and lived experience, a real way of knowing that Socrates dismissed because it couldn’t be defined. Neural networks, Shipper argues, are the first machines that vindicate him.

There’s many different ways of knowing things… that doesn’t mean that we don’t understand them. It just means that we understand them in a different way than we might be used to.

His closing example: spend enough time with ChatGPT and you develop a feel for when it’s reliable and when it’s making things up, the same way you sense when a friend is lying. Not a rulebook. A trained intuition. And in this era, he says, the useful skill is being comfortable working with things that stay a little mysterious.

Key Takeaways

Rationalism = the belief that real knowledge must be statable as explicit rules or definitions; if you can’t define it, you supposedly don’t know it.
It traces from Socrates (in Plato’s Protagoras) through Descartes, Newton, and Galileo, and underpins essentially all modern technology.
It works brilliantly in physics but stalls in psychology, economics, and neuroscience, where universal laws don’t replicate.
Symbolic AI (1950s) applied rationalism to thinking: reduce intelligence to symbols and explicit rules. It solved toy problems but broke on complex ones as the search space exploded.
The rule-writing approach always drowns in exceptions, defining “an important email” requires defining nearly everything about the world.
Neural networks learn patterns from many examples instead of rules; the knowledge they gain is inexplicit and can’t be read off as a list, much like the brain.
Neural networks resemble human intuition, which is likewise trained by accumulated experience, not stated rules.
We tend to model the mind on our dominant tool (Freud: steam engine; modern era: computer), and the computer metaphor hid the role of intuition.
The practical takeaway: build a feel for tools like AI, and stay comfortable working with things you can’t fully reduce to rules.

Claude’s Take

This is a clean, genuinely useful framing of a real intellectual history, and Shipper is good at making a 2,000-year arc feel like one continuous argument. The symbolic-AI-to-neural-net pivot is a well-documented turning point, and the spam-filter example is the best one-paragraph explanation of why rule-based systems break that you’ll hear.

Where it gets slippery is the central claim that neural networks “vindicate” intuition and prove Protagoras right. That is a rhetorically satisfying loop, but it leans on a metaphor doing heavy lifting. Neural nets are loosely brain-inspired; they are not evidence about how human intuition works, and treating their success as a philosophical verdict on Socrates is a stretch he glides over. There’s also a faint false-binary running through it: rationalism vs intuition, as if you must pick. To his credit, he partly defuses this himself, saying rationality “emerges out of” intuition and “you sort of need both,” which is the more honest position and slightly undercuts the provocative title.

The replication-crisis point is real but used as a rhetorical cudgel; plenty of psychology and economics findings do replicate, and the crisis is more about incentives and statistics than about rationalism per se. Score of 7: smart, well-told, intellectually nutritious, and a good on-ramp to a deep topic, docked a little for letting an elegant analogy stand in for an argument.