Hamming's Glib Line: Two Kinds of Not Knowing

The Note That Asked For A Quote

On the 5th of July 2026, a note went into a capture inbox that was not an idea, not a task, and not a reminder. It was a request for accuracy.

The note said, roughly, that there is a quote about science and engineering, that it goes something like: in science, if you know what you are doing, you shouldn't do it, and in engineering, if you don't know what you are doing, you shouldn't do it. And then the instruction: let's get that transcribed properly.

The transcription had garbled it slightly, which is fitting for a line about not knowing what you are doing. It is filed under the podcast category, which means the person who recorded it thought it might make an episode. It has sat there for 8 weeks.

Here is the line, properly. It is by Richard Hamming, from a book called The Art of Doing Science and Engineering, subtitled Learning to Learn, and it appears on page 5, introduced by Hamming himself with the phrase "put glibly", which is a warning he attached to his own aphorism and which almost everyone quoting it since has dropped.

In science, if you know what you are doing, you should not be doing it.

And then the mirror image for engineering: if you do not know what you are doing, you should not be doing it either. Hamming immediately adds a qualification that gets quoted even less often, which is that you almost never see either pure state.

That qualification is the whole thing. The line is not a wall between 2 disciplines. It is a description of what mixing them costs.

The Man Who Ran The Machines

Richard Hamming was born in 1915 in Chicago and took his doctorate in mathematics from the University of Illinois in 1942, and then the war took him where it took everybody with a mathematics degree.

He ended up at Los Alamos, and his job there was not designing anything. His job was running the computing machines that other people's calculations went through. He described his own role afterwards with a striking lack of vanity: he was, he said, in effect a janitor of the machines. The physicists thought, and he made sure the machines that did their arithmetic kept working.

That is a peculiar seat to occupy at a historical moment, and it did something to him. He was close enough to the largest scientific project of the century to see exactly how it worked, and far enough from the centre to be permanently aware that he was not the one thinking the important thoughts. The question of what makes work important, as opposed to merely competent, seems to have entered him there and never left.

After the war he went to Bell Telephone Laboratories, and stayed for 30 years. The building he worked in at that time contained, at various points and in nearby offices, Claude Shannon inventing information theory and John Tukey doing much of the statistics the modern world runs on. Hamming later said, quite plainly, that being surrounded by people like that changed what he thought was possible.

The Weekend The Machine Stopped

The thing his name is attached to came out of frustration, which is where a surprising amount of good engineering comes from.

The relay computers at Bell Labs in the late 1940s ran unattended over weekends. They had error detection: if a machine noticed something had gone wrong in its calculation, it would stop, drop the job, and move on to the next one. Hamming would submit a job on a Friday, come in on a Monday, and find that his run had been thrown away because of a single bad bit somewhere in the middle.

The frustration produced a question, and it is the question rather than the answer that is worth carrying around. The machine can tell that something is wrong. Why can it only tell? If it knows an error occurred, why can it not work out where the error is and put it right?

That question is close to absurd on first hearing. The machine has no idea what the correct answer was supposed to be. But Hamming worked out that if you add extra bits in the right arrangement, each one checking an overlapping subset of the data bits, then the pattern of which checks fail points directly at which bit is wrong. And once you know which bit is wrong, correcting it is trivial, because a bit has only 2 possible values.

Error-correcting codes came out of that, published in 1950, and their descendants are now everywhere. In memory chips, in storage, in every wireless connection, in deep space communication where retransmission takes hours. The idea that a message can carry within itself enough structure to repair its own damage is one of the genuinely load-bearing ideas of the century, and it started with a man being annoyed on a Monday morning.

He received the Turing Award in 1968. There is an award named after him now, which is a strange kind of immortality for someone who spent his early career as the janitor of the machines.

What The Aphorism Actually Says

Return to the line, because the popular reading of it is too flattering to be useful.

The popular reading is that science is brave and engineering is careful, and that the quote is a compliment to both. That reading makes it a poster.

The harder reading is that the line describes 2 completely different relationships to uncertainty, and that the failure mode of each is doing the other one's job by accident.

Science, in Hamming's sense, is the activity where not knowing the answer is the entire point. If you can predict the result, running the experiment tells you nothing except that you were right, which you already suspected. He made this argument in an earlier book too, in 1973, using the word research rather than science, and adding the sharpest part of it: when the answer turns out exactly as expected, you have learned nothing new, though your confidence may have gone up a little. Confidence is not knowledge. It is a feeling about knowledge.

Engineering is the opposite relationship. Engineering is when something must work, for other people, in conditions you will not be present for. If you do not know what you are doing there, you are not being brave. You are transferring your uncertainty onto whoever has to live with the result.

The failure that the line is actually warning about is doing one while believing you are doing the other.

Building a production system while secretly running an experiment is the common version. You have not decided how the thing should work, so you find out by building it, and the finding-out gets shipped along with the result. That is science wearing engineering's clothes, and the bill arrives later in maintenance.

The rarer version is the opposite and just as costly. Running an experiment with the caution of an engineer means hedging, keeping the old thing running alongside, testing what you already believe, and never actually reaching the place where you would be surprised. That is engineering wearing science's clothes, and its cost is that nothing is learned, slowly, at great expense, with excellent documentation.

Neither of those is a lack of skill. Both are category errors about which activity you are currently in.

Two Doors

Hamming's most repeated piece of work is not the codes. It is a talk he gave at Bell Labs in 1986, called You and Your Research, which is really a talk about why some people do important work and most people do not.

The observation from it that has aged best is about doors.

He noticed that colleagues who worked with their office doors closed got more done day to day. They were less interrupted, they finished more, their output was tidier. And he noticed that, over 10 and 20 years, it was consistently the people who kept their doors open who did the work that mattered.

His explanation was not that interruptions are good. It was that the open door let in the questions that were not yours yet. Somebody wanders past with a problem from an entirely different field, and 2 years later that problem turns out to be the thing you were unknowingly equipped to solve. The closed door optimises the work you already have. The open door changes which work you have.

The cost is real and he did not pretend otherwise. The open-door people were less efficient, and they knew it. He was describing a trade, not a free lunch: short-term throughput exchanged for the chance of relevance.

The Courage Part

There is a harder claim in that talk, and it is the one people quote least because it is uncomfortable.

Hamming argued that the distinguishing feature of the people who did great work was not intelligence. Everybody at Bell Labs was intelligent. It was courage, by which he meant something quite specific: the willingness to keep going into a problem after it has become clear that it is hard, that you might fail publicly, and that a safer problem is available.

His example of what happens without it is one of the sharpest things in the talk, and it is about Shannon. Once you are famous, Hamming said, small problems become hard to work on, because after inventing information theory, what do you do next. So the great ones stop planting the small acorns from which the large trees grow. They try to go straight to the big thing, and that is not how any of it works. Early recognition, he thought, has a sterilising effect.

That is a warning aimed directly at anyone who has ever had one thing go well and then found the next thing strangely difficult to start.

He also had a line about problem selection that is worth carrying: doing the right problem badly beats doing the wrong problem beautifully. The craft is in the choosing, and most people spend their care on the execution instead, because execution is the part you know how to be good at.

Compound Interest

The last piece of the argument is arithmetic, and Hamming used it deliberately because he was a mathematician talking to people who would find a number more persuasive than an exhortation.

Knowledge and productivity, he argued, behave like compound interest. Two people of equal ability, where one puts in 10 percent more effort in the right direction, do not end up 10 percent apart. The gap widens continuously, because each increment sits on top of everything accumulated before it, and the difference over a career is not a margin but a different category of outcome.

The clause everybody drops from that argument is "in the right direction". Compound interest works just as reliably on effort spent on the wrong problem. Thirty years of diligent work on something that does not matter compounds into thirty years of diligent work on something that does not matter, and it feels, from the inside, exactly like the other thing.

Which is why the door and the courage and the problem selection are not separate pieces of advice. They are all the same piece: the compounding is automatic, so the only decision that carries real weight is what you point it at.

Where It Lands

Hamming left Bell Labs in 1976 and went to teach at the Naval Postgraduate School in Monterey, which he did for over 20 years, until shortly before his death in 1998. The book with the glib line in it came out of those lectures. He said he had turned to coaching, on the theory that in his own field the great things tend to get done young, and the useful thing an older person can do is help somebody else do them.

The line itself, restored to its full form, is not really about 2 professions. It is a question you can ask about any afternoon: which of these 2 activities am I actually doing right now, and am I applying the right standard to it.

If the honest answer is that you are finding out whether something works, then the standards of engineering are the wrong standards, and the thing should be allowed to be ugly, temporary, and thrown away without ceremony.

If the honest answer is that this thing is going to be relied upon, by anyone, including a version of yourself who will have forgotten everything about it in 6 months, then curiosity is no longer the governing value, and the parts you do not understand are not interesting. They are debt.

The trouble, and the reason the note sat in an inbox for 8 weeks, is that almost nothing announces which one it is. It gets decided retroactively, by whether the thing survives.