Everything Looks Like a Nail

In 1964 the philosopher Abraham Kaplan gave a name to something everybody had already noticed. “I call it the law of the instrument,” he wrote, “and it may be formulated as follows: Give a small boy a hammer, and he will find that everything he encounters needs pounding.” The next sentence is the one that matters, and it is the one nobody quotes: “It comes as no particular surprise to discover that a scientist formulates problems in a way which requires for their solution just those techniques in which he himself is especially skilled.”

Two years later Abraham Maslow remembered a car wash. An elaborate automatic machine that did a beautiful job of washing cars, and could do only that, so that everything else that got into its clutches was treated as a car to be washed. “I suppose it is tempting,” he wrote, “if the only tool you have is a hammer, to treat everything as if it were a nail.” The sentence stuck to Maslow. Neither man invented the idea: Kaplan was already using a version in a speech in 1962, and an English magazine in 1868 had a boy with a new hammer and chisel hacking at the doorposts.

The line is usually told as a joke about narrow people. This essay argues the joke is aimed at the wrong place. Using your hammer on things that are not quite nails is very often the right call. The real mistake is quieter, it happens after the job is done, and it compounds. The tool below takes it apart: two kinds of job, one person who is good with a hammer, and a logbook.

The hammer is usually the right call

Start with the part that is not a mistake. Jobs come in two kinds, nails and screws, and say 30 percent are nails. You never get to see which kind a job is. You see what it looks like, and the two kinds overlap: some screws look a lot like nails. So you do the sensible thing. You estimate how likely this one is to be a nail, and you pick the tool that pays best on that estimate.

Now make yourself twice as good with the hammer as with anything else in the bag. The best rule — not a biased rule, the best one — now hammers about 52 percent of all jobs, when only 30 percent are nails. Some of those are screws. That is correct, because a very good hammer on a screw beats a mediocre screwdriver on it more often than not. Equal hands hammer 24 percent of jobs. One and a half times as good: 38. Three times: 87. Getting better at the hammer really does make more problems look like nails, and in this first year every one of those calls is the right one.

Being good with a hammer is a reason to hammer more things. It is not a reason to believe more things are nails.

Good thoughts block better ones

There is a measured version of the hand that reaches for the familiar tool, and it comes from chess. Merim Bilalić, Peter McLeod and Fernand Gobet showed strong players a position with two ways to win. One was a smothered mate in five moves, a pattern every good player knows. The other was less familiar and shorter: mate in three. The players were asked for the shortest solution. Candidate masters, three standard deviations above the average rated player, all spotted the smothered mate quickly, said they were looking for something shorter, and none of them found the mate in three. Move one piece so that the familiar mate no longer works, and every one of them found it.

Masters did a little better, at 18 percent. International masters found it half the time. The authors put a size on the effect by asking how much weaker a player had to be, with the familiar solution taken away, to perform that badly. The answer, across skill levels, was about three standard deviations: a strong player looking at a familiar idea solved the problem like a player who would almost never beat them. Then they tracked the players’ eyes. The players said they were searching for alternatives. Their eyes kept returning to the squares of the solution they already had.

The original version is older and simpler. In 1942 Abraham Luchins gave people a run of water-jar puzzles that all yielded to the same long method. Then he gave them one where the long method still worked but a much shorter one did too, and most kept using the long one. Then one where only the short method worked, and many declared it impossible. A control group who had never learned the long method solved it quickly. The name he used was Einstellung, German for a setting: the mind set, like a machine, to run one way.

Notice what the chess authors do not say. They do not say the experts were bad at chess. They say the same mechanism that blinded them is the one that “ensures a speedy response in familiar situations”. The fast thing and the blind thing are one thing. And the effect shrinks as skill rises: the international masters found the short mate half the time, the candidate masters never. Expertise is not the problem. The setting is.

The logbook

Here is the step the old line leaves out. A job comes in, you hammer it, and it more or less holds. What do you write down? Not “a screw, which I hammered, which mostly held”. Nobody keeps that record. You write down what you did, and the job goes into the book as a nail. It is only corrected if the wrong tool visibly fails — the screw pulls out, the client calls back, the patient comes in again.

And next year’s sense of how common nails are is built from that book. This is the part worth slowing down for. The base rate people trust most is the one they never had to look up: their own experience. But experience is not a sample of the world. It is a sample of your decisions, labelled by your decisions. Feed it back into the next decision and you have a loop: a lower bar for hammering writes more nails into the book, and more nails in the book lower the bar.

Run it. Twice as good with the hammer, and one wrong tool in five shows itself. In year one the book says 30 percent nails, which is the truth. After five years it says 62. After ten, 74. It settles near 77 percent and stays there. By then you hammer 89 percent of everything, 60 percent of all jobs are done with the wrong tool, and 88 percent of jobs still look done. The world sent 30 percent nails every single year. Nothing in your records says anything went wrong.

You did not get worse with the hammer. You got a worse map, and you drew it yourself.

Take the feedback away entirely — the wrong tool never visibly fails — and the book passes 96 percent nails within ten years. Every job looks done. Make the hands equal instead and the loop does not stop; it runs the other way. Nails are the minority, so the ambiguous ones get the screwdriver and go into the book as screws, and after thirty years it says 17 percent. The loop has no opinion about hammers. It favors whatever you were already doing.

It also has a threshold. With silent mistakes, a hammer 1.2 times as good as the rest leaves the book roughly where it was. At 1.3 it creeps up to 44 percent. At 1.5 it runs to 90. A small difference in the hands produces a different career. And the cost, measured as output lost against the same hands with an honest ledger, is not largest for the person with the best hammer. It is largest for the person who is only somewhat better with it: about 19 percent at 1.6 times, in the silent case, falling to under half a percent at three times, because a truly great hammer is close to the right tool for everything anyway.

Three things drive the loop, and they are the three dials on the tool. Hands that favor one tool. Problems that are hard to tell apart. Mistakes that do not announce themselves. Make the two kinds of job clearly different and the book drifts five points and stops. Make every wrong tool show itself and it never moves at all.

Where the hammer is heaviest

Those three conditions describe a real place quite precisely: a field where the problems are ambiguous, the practitioners are trained in one method, and the outcome arrives years later, if it arrives legibly at all. In 1998 a group of researchers sent the same questionnaire to American urologists, who perform surgery, and radiation oncologists, who deliver radiation, about men with prostate cancer confined to the prostate. For a man with a moderately differentiated tumor and more than ten years to live, 93 percent of urologists recommended removing the prostate. The authors’ summary was one sentence: “specialists overwhelmingly recommend the therapy that they themselves deliver.”

This is not evidence that urologists were wrong. The editorial that ran beside the paper was titled “Uncertainty in Prostate Cancer Care”, and that is the point: where nobody can say with confidence which tool fits, the loop has the most room. What came back from the survey is what you would expect from two groups of honest people, each looking at the same ambiguous problem through a career’s worth of their own records.

And now you

Nothing here says to stop being good at things. It says the instrument that tells you what kind of problem you are looking at was built by your own hands, and it drifts in a known direction. Calibrate for it.

Four ways this is wrong

Specialization pays. The model agrees, and says so on screen: with an honest ledger, a better hammer should be used on more things, and the cost of the drifting book nearly vanishes once the hammer is very good. Telling a specialist to use more tools is often bad advice. The argument here is narrower — keep using the hammer, stop letting it write the map.

Two kinds of job is a cartoon. Real problems come in many kinds, with many symptoms and outcomes that differ by degree. The model has one symptom and one fixed penalty for the wrong tool: it half-works, at 30 percent of the right one. Whether more kinds of job speed the loop up or slow it down is something this page has not measured.

Nobody’s logbook is that naive. The person in the model believes last year’s records and nothing else. Real people read, train, hear from colleagues, and occasionally have the visible disaster that resets the book. All of that is feedback, and feedback is the dial that kills the loop. The model is the case where nothing else gets in, which makes it the strong version of the effect, not the typical one.

The evidence is thin and tangled. The prostate survey measured stated recommendations, not treatments and not outcomes, and surgeons are paid for surgery and radiation oncologists for radiation: the hammer and the wallet point the same way, and a survey cannot separate them. The chess studies are lab problems and small groups of players. The logbook mechanism itself is this page’s model; this page knows of no study that has watched a practitioner’s own base rate drift. What the evidence does show is the direction, and it has the sign the loop predicts every time.

Being good with a hammer is not the problem. Believing your own records is. Run a career yourself in The Hammer. The Base-Rate Machine is what a correct base rate does for a single judgment, and The Cemetery is another record that only keeps what it was shown.