Feynman Lectures on Computation
How Feynman taught the future
Description
Between 1983 and 1986, Richard Feynman taught a course at Caltech that had no obvious business on his schedule. He was a physicist — the Nobel kind, the one who had drawn diagrams to tame quantum electrodynamics and cracked open the Challenger investigation with a glass of ice water. Computation was not his field. He co-taught the course, at first with the computer scientist John Hopfield and the polymath Carver Mead, then increasingly on his own terms, and he approached the machine the way he approached everything: by refusing to accept that anyone else's explanation was good enough until he could rebuild it himself.
The lectures were never meant to become a book. Feynman lectured; the notes sat around; the material risked evaporating the way brilliant improvised teaching usually does. It fell to Tony Hey, a physicist who had attended, to reconstruct the course years later into the volume published in 1996 as Feynman Lectures on Computation. Guests had passed through the room — Marvin Minsky on artificial intelligence, Charles Bennett on the thermodynamics of computing, John Hopfield on neural networks — but the spine of the thing was Feynman working out, in public, what a computer actually is.
The surprising part is where he lands. He treats the computer not as an abstraction made of logic and symbols, but as a physical object obeying physical law — a thing that costs energy, generates heat, and bumps up against limits set by thermodynamics and quantum mechanics rather than by clever engineering. Written before the word quantum computing meant much of anything, the lectures read now like a map drawn slightly ahead of the territory.
The question we’re asking : What does a Nobel physicist see in a computer that a computer scientist might miss?What we’ll see : A physicist rebuilds computation from the ground up and follows it down to where the laws of physics take over.
Table of contents
01Chapter 1 — A physicist walks into a computer science class
Feynman had a rule he applied to any subject he claimed to understand: he had to be able to derive it himself, from something simpler, without leaning on jargon or authority. The course at Caltech was that rule turned into a syllabus. He did not want to teach students how to program, or how to use the machines that were then spreading across campuses. He wanted to work out what computation is at its root, and he was willing to start almost embarrassingly far back to do it.
So the lectures open not with silicon but with logic. Feynman builds up from the simplest possible pieces — gates that take inputs and produce outputs, the AND, the OR, the NOT — and shows how these primitive elements, wired together, can carry out any calculation you could specify. It is the standard foundation of the field, but he refuses to take it on faith. He constructs adders out of gates, shows how memory can be held, and demonstrates that a handful of trivial operations, repeated enough times, amount to everything a computer does.
02Chapter 2 — What a computer really has to do
Before he can ask what computation costs, Feynman needs a clean idea of what computation is. Here he reaches for the classic abstraction: the Turing machine, Alan Turing's imaginary device that reads and writes symbols on an endless tape according to a small table of rules. It is deliberately, almost comically minimal — a head that moves left or right, a strip of squares, a few states — and yet Turing had shown in the 1930s that this stripped-down contraption can compute anything that any machine can compute.
Feynman treats the Turing machine less as a piece of history than as a working tool. He walks through how it operates, why its simplicity is the point, and how it lets us talk about computation without getting tangled in the specifics of any real hardware. From there he moves to the questions of what is computable at all, and what is computable in a reasonable amount of time. Some problems can be solved quickly; others, though perfectly well-defined, would take longer than the age of the universe as they grow — the distinction that computer scientists dignify with the language of complexity.
03Chapter 3 — The physics that sets the limits
If computation is a physical process, then the limits on computers are physical limits, not merely engineering ones. Feynman spends a good part of the course chasing those limits down to their sources — the thermodynamics of heat, the graininess of quantum mechanics, the sheer amount of energy it takes to change one thing into another. The engineering questions of the day were about making chips smaller and faster. Feynman was asking how small and how fast the universe would let them get.
The most striking result he develops concerns heat. Every computer runs hot; anyone who has felt a laptop knows it. The intuitive assumption is that this waste heat is a symptom of imperfect engineering, something a cleverer design could squeeze away. Feynman, drawing heavily on work by Rolf Landauer and Charles Bennett — Bennett being one of the guests who passed through the course — argues that the story is subtler. There is no fundamental thermodynamic cost to computing as such. The cost appears at a specific moment: when information is erased.
04Chapter 4 — Reversible machines and the cost of forgetting
Take Landauer's insight — that only erasure has an unavoidable energy cost — and follow it where Feynman follows it, and you arrive at a strange conclusion. If forgetting is what makes computing expensive, then a machine that never forgets could, in principle, compute for free. Not free in practice, but free in the sense that no law of thermodynamics stands in the way. This is the idea of reversible computation, and it reframes what a computer fundamentally is.
An ordinary logic gate throws information away. Feed two bits into an AND gate and get one bit out, and you can no longer reconstruct the inputs from the output — the machine has forgotten something, and by Landauer's argument it must pay in heat. Bennett had shown, and Feynman lays out with evident pleasure, that any computation can be rebuilt out of gates that discard nothing, gates whose inputs can always be recovered from their outputs. Such a machine runs its logic forwards and could, if you liked, run it backwards to exactly undo itself.
05Conclusion
The book that carries Feynman's name on computation appeared in 1996, eight years after his death, reconstructed by Tony Hey from notes and memory. It is an odd artifact — a physicist's improvised course, salvaged and set in type, still audible in its own pages as someone thinking out loud rather than delivering settled doctrine. The guests who passed through, Minsky and Bennett and Hopfield among them, left their fingerprints, but the through-line is unmistakably Feynman refusing to accept computation as anyone else had packaged it.