John McCarthy and Symbolic AI
John McCarthy was an American computer scientist who coined the term "Artificial Intelligence" in 1955 and pioneered the idea that human reasoning could be represented as formal symbols manipulated by machines — a school of thought now called Symbolic AI or "Good Old-Fashioned AI" (GOFAI). His work established the foundational belief that intelligence is, at its core, about logic, language, and rules — not neurons or statistics — and shaped the entire first half of AI's history. The counterintuitive fact: McCarthy was deeply skeptical of neural networks his entire career, yet the modern AI systems he doubted most have eclipsed the symbolic approach he championed — while simultaneously borrowing many of his core ideas about knowledge and reasoning.
Symbolic AI is built on one bold claim: thinking is symbol manipulation. Just as algebra lets you solve problems by rearranging letters according to rules, McCarthy believed a machine could "think" by storing knowledge as symbols — words, categories, logical statements — and applying rules to rearrange them into new truths. The "aha moment" is realizing that this turns intelligence into something you can read, inspect, and argue with: unlike a neural network's billions of inscrutable numbers, a symbolic AI system reasons in plain statements like "IF the patient has a fever AND a rash THEN consider measles." That transparency is both its greatest strength and, ultimately, its greatest limitation.
How It Works
Symbolic AI works roughly the way a very organized library works. Imagine every fact the system knows is written on an index card — "birds have wings," "penguins are birds," "things with wings can fly." These cards are stored in a knowledge base. Alongside the knowledge base sits an inference engine — a set of rules for combining cards to produce new conclusions. If the engine knows "penguins are birds" and "birds have wings," it can derive "penguins have wings" automatically, without anyone telling it directly. This process — moving from known facts to new conclusions using rules — is called logical inference, and it is the heartbeat of every symbolic AI system McCarthy championed. The rules are written in formal languages borrowed from mathematics and philosophy, particularly a branch called predicate logic, which lets you say things like "for every X, if X is a bird, then X has wings" with perfect precision. McCarthy's biggest personal invention in this space was a programming language called LISP, created in 1958. LISP was revolutionary because it treated program code and data as the same kind of thing — both were lists of symbols — which made it perfect for building systems that could manipulate and reason about their own knowledge. For the next thirty years, virtually every AI lab in the world ran on LISP. Beyond LISP, McCarthy developed the concept of situation calculus — a formal language for describing how the world changes when actions are taken. Think of it as a way of writing down "before this action, the world looked like THIS; after the action, it looks like THAT." This was the intellectual foundation for everything from robot planning to automated scheduling systems.
McCarthy grew up in Boston and Los Angeles, the son of Irish and Lithuanian Jewish immigrants, and showed a prodigious talent for mathematics from childhood — he taught himself university-level calculus before finishing high school. He organized the 1956 Dartmouth Conference, the ten-person summer workshop that gave the field of AI its name and its founding ambition, and spent most of his career at Stanford where he led the Stanford AI Laboratory (SAIL) for decades. A surprising personal fact: McCarthy was a passionate advocate for space colonization and nuclear power, believing technological optimism was a moral obligation — he once wrote that pessimism about technology was "a form of cowardice."
Newell studied mathematics at Stanford and then shifted to cognitive science, driven by a conviction that understanding human thought and building intelligent machines were the same project viewed from different angles. Together with Herbert Simon at Carnegie Mellon, he built the Logic Theorist in 1956 — arguably the first true AI program, which proved 38 of the 52 theorems in Whitehead and Russell's Principia Mathematica — and later the General Problem Solver, a system designed to mimic human problem-solving strategies universally. The surprising fact: Newell and Simon were so confident their General Problem Solver represented genuine human cognition that they published predictions in 1958 that a computer would be the world's chess champion within ten years — a prediction that was off by about forty years.
Intuition
Imagine you are a new detective joining a police department. On your first day, your senior colleagues hand you a thick rulebook — thousands of pages describing how crimes are committed, what clues mean, how suspects behave. They also give you access to a filing cabinet stuffed with case notes about every crime in the city's history. When a new crime arrives on your desk, you work through it the same way every time: you look up the facts in the filing cabinet, you apply the rules from the rulebook, and you step logically from one conclusion to the next until you either name a suspect or hit a dead end. You never guess. You never have a gut feeling. You follow the rules, always. This is exactly how a symbolic AI system works. The filing cabinet is the knowledge base. The rulebook is the inference engine. The conclusions you reach are guaranteed to be logically correct — if the rules and facts you started with were correct. Now here is why this approach eventually ran into trouble: imagine the rulebook has to cover not just crime, but everything in human experience — how weather affects mood, what sarcasm sounds like, why a child might lie to protect a friend. Nobody could write that book. The real world is too vast, too messy, too full of exceptions. Symbolic AI was a perfect detective in a world where the rulebook already existed. The hard part was writing the rulebook in the first place.
Historical Development
Origins · 1 minhow this was discovered
Why It Matters
Prerequisites
What This Leads To
See Also
Alan Turing and the Foundations of Computing · Marvin Minsky and Perceptrons · Expert Systems and MYCIN · The Dartmouth Conference 1956 · Neurosymbolic AI