The Power of Understanding - Judgment Is the Art of Seeing With Understanding

Inventor's Mind — feature essay

Judgment Is the Art of Seeing With Understanding

Inventor's Mind — feature essay

I joined an aerospace company at twenty-five, and I came in with my hands already trained. I'd worked on mechanical things most of my life — hands first, thinking second — and then engineering school spent four years reversing that order, teaching me to think more and do less. By the time I graduated, solving a problem meant reaching for a formula, a calculator, and eventually a computer, turning the crank, and writing down the number that came out. We never asked how exact that number was. The number was the answer. That the answer might be more precise than the thing it described was not a question the homework ever posed.

Then I joined the company, and a phrase kept surfacing that I didn't have a category for. Close enough. It came from the fifty-year-old engineers who were training me, and they said it constantly — leaning over my work, glancing at a result I'd carried to six decimal places, and pronouncing it close enough, or not. They never graded me for accuracy. They had something else, something they'd plainly learned on the job and couldn't quite hand over, a kind of gift that told them what a good answer was and what wasn't. And it never ran good, better, best. It ran good, or not right. A binary. The answer was either inside the band that mattered or it wasn't, and they could see the band, and I couldn't yet.

It took me three decades to understand what I'd been watching, and to watch the tools slowly take it away. We told ourselves our instruments made us think better. They didn't. They let us stop thinking about something we'd been doing for free — and we were so pleased with the precision that we never noticed the bill.

The first lie was a decimal place

The smallest version of the whole story is a number you can hold in your hand.

The slide rule, for all its crudeness, was honest about one thing: it gave you three significant figures and not one digit more, because three was all it could show. You ran pi as 3.14, plugged it in, and moved on. The calculator that replaced it handed you 3.1416 and an answer carried to four decimal places — and for any real part you would ever build, it was the same answer. The physical truth hadn't moved an inch. A shaft that was strong enough at three figures was strong enough at four; the extra digits described a precision the world underneath the part did not possess.

But four decimals look like knowledge. Three look like an estimate. And we started trusting the longer number — not because it was truer, but because it was longer. The fourth decimal place is a claim: it asserts you know the answer to one part in ten thousand. You almost never do, because the inputs were never known that well. The material property was good to maybe ten percent; the load was an estimate; the boundary condition was an idealization. Pi to 3.14 was honest about the company it kept. Pi to 3.1416, sitting in an equation next to a number you knew to one significant figure, was theater — a decimal place of pure performance, asserting an exactness that every other term in the equation flatly contradicted.

The slide rule could not tell you more than it knew. The calculator could, and did, and we believed it. Somewhere in those extra, meaningless digits, the band disappeared — the felt sense of how wrong I might be that the old engineers carried in their hands and called close enough. The number got longer and the honesty got shorter, and almost no one noticed the trade, because the trade felt like progress. It looked exactly like getting better at our jobs.

It was the first lie in a staircase of them, and every step up that staircase is the same lie told with more conviction.

Two airplanes

Set two aircraft next to each other and the bill becomes visible.

The F-22 Raptor was the first United States fighter designed and analyzed end to end in modern CAD, FEA, and CFD. It is, by most measures, the finest air superiority fighter ever built. One hundred eighty-seven production airframes were delivered. The line closed in 2011. It entered service in 2005, which means it has been flying for about twenty years, and the aircraft meant to succeed it is not yet in the air.

F-22 Raptor and the B-52 Stratofortress

The B-52 Stratofortress was drawn with pencils and French curves. Its structure was sized with closed-form plate-and-beam equations, the math run on slide rules or worked out by hand down the length of a page. Seven hundred forty-four were built. The last rolled out in 1962. It has been flying since 1955 — seventy years — and the Air Force is paying to give it new engines so it can keep flying after every engineer who drew it is dead.

The airplane built with the cruder tools is the one that lasted. That is not an accident, and it is not nostalgia. It is a fact with a mechanism underneath it, and the mechanism is the whole point.

A confession

I should be honest before I go further, because the easy version of this essay is a lie.

The easy version says the old tools were wiser and the new tools made us reckless. I don't believe that, and I have no standing to pretend I do, because I preferred the new tools. I still do. The first time finite-element analysis let me shave a section that the old hand calculation would have left thick — thick "to be safe," thick because the cruder method couldn't see precisely enough to do otherwise — I felt the clean satisfaction of precision. I trimmed it. And I did not, in that moment, notice that "to be safe" had been doing real work. I thought I was removing fat. I was removing a margin I hadn't been asked to keep and didn't yet understand I'd been receiving.

That margin was close enough made physical. The fifty-year-olds couldn't carry six decimals in their heads, so they carried a band instead — a sense of how wrong they might be, built straight into the part as extra metal. I had a tool that erased the band, and I called the erasing progress.

So this is not a story about good engineers and bad engineers. It is a story about something that used to be free becoming something you now have to buy on purpose — and about how few of us noticed the price tag had appeared.

The tool was never the thing

Here is the correction, and it is the spine of everything that follows.

The tool is neutral. CAD, FEA, CFD — they compute what you point them at, with terrifying competence, and they care about nothing. Point them at minimum weight and they will give you minimum weight. Point them at the exact edge of the requirement and they will walk you right up to it and stop, precisely, on the line.

The slide rule was neutral too. It didn't think. What the slide rule did — and this is the part we lost without recording the loss — was force the understanding into your hand. You could not get an answer out of it without running the mechanism yourself. You had to know the equation, know what each term meant, know roughly where the answer should land before you got there, because the tool gave you no help in noticing if you'd gone wrong. The crudeness wasn't wisdom. It was a kind of friction, and the friction kept your hand on the mechanism every single time.

Which lets me say what judgment actually is, because the slide rule and the workstation together draw the line around it precisely.

Judgment is not a rule. It is not a procedure, and it is not a virtue you can resolve to have. Judgment is the art of seeing an issue with the understanding already inside the seeing. A novice and an expert look at the same drawing. Only one of them sees the failure — not because the expert reasons faster afterward, but because the expert is looking with the mechanism fused into the perception. A radiologist sees a tumor where you and I see gray. A structural engineer sees a fatigue crack propagating from a stress riser where the layman sees a line. The understanding isn't applied after the seeing. It is the seeing. That fusion — understanding gone fast enough to arrive as sight — is the whole of engineering judgment.

And it is exactly the thing the clean answer dissolves.

How precision strips the seeing

Walk through what the cruder method was actually doing.

Closed-form plate-and-beam equations could not model the real stress field in a complicated part. They idealized it — treated the messy three-dimensional truth as a beam, a plate, a ring — and because the idealization was always a little wrong, the honest engineer carried margin to cover the gap between his model and the world. He over-built. And over-built things last, because the margin he carried for modeling ignorance turned out, decades later, to also cover loads he never imagined, missions the airframe was never drawn for, the slow surprises of a long life.

FEA closed that gap. It models the real stress field, near enough, so the reason for the margin evaporates — and the engineer, trusting the model because the model now deserves trust, cuts to the edge it can finally see. Nothing in this is a mistake. Each step is correct. The part is lighter, cheaper, and exactly strong enough for every load case anyone specified.

The trouble is the loads nobody specified. FEA answers the question you asked with a precision that feels like completeness, and completeness is the illusion — the same illusion as the calculator's fourth decimal, one floor up. The decimal place lied about how exactly you knew the answer. The clean stress plot lies about how completely you asked the question. The number tells you the part survives the load case. It says nothing about the load case the world will hand you in 2040, because you didn't ask, and the tool only answers what it's asked. The slide rule couldn't give you false completeness — its roughness announced itself, and the announcement kept you looking. The workstation gives you an answer so clean there's nothing rough to catch your eye on, and the eye, finding nothing to snag, stops looking.

That is the mechanism. The tool returns the what — the result — without the why — the mechanism underneath it — and judgment is precisely the seeing of the why inside the what. Sever those, and you still have eyes. You just don't have judgment anymore. You have a competent person looking at a correct number, mistaking looking for seeing.

None of which means the tools were wrong to build. This has to be said plainly, because the F-22 is the proof. Its envelope — stealth, supercruise, the coupled behavior of air and heat and structure all at once — is simply not reachable by hand. Closed-form methods cannot get you there. The tools didn't merely permit that airplane; they were the only road to it. The expanded reach is real and it is a genuine gain. The loss is separate, and quieter: somewhere in the gain we stopped running an objective that valued the life of the thing, and only valued its performance on the day it was delivered. The margin that made the B-52 immortal was free in 1952. By 1997 it had to be deliberately bought — and no one wrote it into the requirement.

Where judgment comes from, and how it dies

If judgment is understanding fused into seeing, the obvious question is how anyone gets it. The obvious answer is the dangerous one: experience. You fail, you pay, the failure burns a mark on the dial, and over a career the marks become a feeling — this is off, and badly — that you can't quite put numbers to but can absolutely trust.

That answer is half right, and the wrong half is a trap.

It's true that judgment is calibrated by failure. But your own failures are a poor dataset. They are too few, and they are survivor-biased by construction: you only carry the ones you lived through, you never see the near misses that should have terrified you, and the failures that would have taught you the most are precisely the ones you didn't survive to remember. An engineer who trusts only his own scars is trusting the testimony of a witness who wasn't allowed in the room for the worst of it.

There is a far better dataset, and it is free: everyone else's failures. The recorded ones. The programs that died, the structures that fell, the molecules that washed out of trials, the Raptor truncated to a hundred and eighty-seven. That record is a graveyard with every headstone legible, and unlike your own experience, it includes the bodies — the failures are documented precisely because they failed. You are allowed to walk that graveyard and read every stone. Doing so is the only form of calibration that isn't survivor-biased, because the dead are right there in the data.

But — and this is the discipline that separates judgment from superstition — you have to read for the why, not the what. The what of a failure is local: this specific coupling, this specific power ceiling, this specific thermal path. Learn the what and you learn a superstition — don't do the exact thing that company did — calibrated to a circumstance that will never recur. The why is portable: they optimized to a spec whose mission outlived the airframe; they coupled a layer that needed to stay separate; they trusted a clean answer and stopped looking. The why recurs across every domain that ever shipped a thing too optimized to last. Extract the why and you have bought a mark on your dial without having to bleed for it. That, precisely, is what forensic engineering is — the deliberate extraction of transferable mechanisms from other people's catastrophes — and it is the one way judgment scales beyond a single nervous system.

Now the way it dies, because it does die, and it dies quietly.

The same fusion that makes judgment real — understanding gone fast enough to feel like sight — is also what makes it impossible to audit from the inside. The expert who sees the failure and, when you press him, can only say "I just know" — that man's judgment may be alive, or it may have calcified into a bias he can no longer detect, and from where he stands the two feel identical. A feeling calibrated by past failures reads the past. When the regime changes underneath it — when the new problem only resembles the old danger on the surface while being causally different underneath — the most experienced engineer in the room becomes the most confidently wrong, and his confidence is strongest exactly where his calibration is most stale. The radiologist who sees real tumors also sees tumors that aren't there, and cannot introspect the difference.

So the live form of judgment is not the fast seeing alone. It is fast seeing that stays reversible — that can, on demand, unfold back into the mechanism that earned it. The engineer who sees the failure and can still walk you down to the causal why has judgment that's alive. The one who sees it and can only say "I just know" has judgment that has fused so hard it lost contact with its own foundation, and from the outside that is indistinguishable from prejudice. The test is the unfolding. Can you put the why on the record? A forensic engineer has to — that's the job, you see the failure and you show your work to an attorney — and that obligation to show the work is not a burden on judgment. It is the thing that keeps judgment honest.

What comes next

The staircase has three steps so far, and they climb in the same direction. The calculator inflated the digits — it lied about how exactly you knew the answer. CAD and FEA inflated the completeness — they lied about how fully you'd asked the question. The next tools will inflate the judgment itself, and that is a longer fall than either.

AI design tools are arriving that don't just return a clean answer — they propose the approach, generate the options, and present a result with no visible seam where a human would have had to understand anything. They remove the last forced understanding, which was our slowness, our need to deliberate, the plodding that used to keep our hand on the mechanism whether we wanted it there or not. The slide rule forced understanding into your hand by being too crude to do otherwise; these tools will hand you the seeing and the confidence both, requiring neither understanding nor the flinch that understanding produces.

I can tell you what the failure of that era will look like, because it is the F-22 failure run forward. It will not look like obvious error. It will look like confident, beautifully analyzed, exhaustively modeled designs that are wrong in ways nobody flinched at — because there was no rough edge anywhere in the process to snag a human eye, and the one faculty that could have caught it was never engaged. The Raptor is the first data point. There will be more, and they will be harder to see, because they will arrive cleaner.

Lessons learned

The clean answer is the dangerous one. Whether it comes from a solver or from your own gut, an answer that arrives smooth and certain is the moment the looking stops. Distrust precision most exactly when it feels earned.

Judgment is seeing with the mechanism inside the seeing — not a flinch that reasoning checks afterward. The expert and the novice look at the same drawing; only one sees the failure, because only one looks with the understanding fused into the perception.

Keep the seeing reversible. Live judgment can drop from sight back to mechanism on demand. The forensic test is simple: can you put the why on the record? If the only answer is "I just know," the judgment has already calcified into bias you can no longer detect.

Calibrate on the graveyard, not on your own scars. Your failures are too few and survivor-biased; you never see your near misses. The recorded failures of everyone else include the bodies. Walk among them.

Extract the why, not the what — and distrust how clean the why came out. The local circumstance teaches superstition; the mechanism travels. But the most teachable explanation of a failure is usually the survivor's myth, not its cause, so audit your own appetite for a tidy lesson.

The tool has neither judgment nor understanding. Both are yours to supply — on purpose, to a machine that will increasingly feel as though it has supplied them for you. It hasn't. It has been competent in a narrow window and let you mistake the competence for both.

Close

The slide rule was never smarter than the workstation. It was cruder, and the crudeness kept a hand on the mechanism — and we mistook the crudeness for our own diligence, right up until the tools got smooth enough to take the diligence away and we discovered how little of it we'd been doing on purpose.

The B-52 is flying because the men who drew it could not compute their way to the edge, and the margin their ignorance left behind became seventy years of absorbing missions they never imagined. The F-22 is the finest fighter ever built, and it will be retired younger than some of those bombers' engines, because we could finally see the edge and could not resist standing on it.

I understand now what the fifty-year-olds were doing when they looked at my six decimals and said close enough, or not right. They weren't grading my arithmetic. They were reading the band — the gap between the number and the thing — and telling me whether my answer lived inside it. Good, or not right. It was never about precision. It was about whether I had seen the problem or only computed it. That was the gift, and it was the one thing none of our tools, then or now, could hand to me. I had to earn it the way they did: by understanding the mechanism well enough that I could finally see it, and by keeping that seeing honest enough that I could always unfold it back into the why.

I don't want the old tools back. I want the thing the old tools forced on us and the new tools let us skip: the hand kept on the mechanism, the seeing kept reversible into the understanding that earned it, the refusal to mistake a clean answer for a finished one. That was never the slide rule's gift, and it was never the computer's to take away. It was always ours — the one part of the work that no tool, however precise, was ever doing for us.

The answer arriving clean is the most dangerous thing that can happen to judgment. It is also, now, the most common. We will have to learn on purpose what our tools' limitations used to teach us for free — and we will have to teach the next ones, somehow, to look at a clean and beautiful answer and still ask whether it is good, or only not yet caught being wrong.