03 · The Heuristic Age

THE SHORTEST PATH — A History of the Art and Science of Problem-Solving

Coutances, June 1880

For a fortnight, Henri Poincaré had been failing on purpose. Twenty-six years old and already restless with talent, he sat at his desk in Caen for an hour or two each day trying to prove a negative: that a certain kind of function — later named, in his honor and against his protest, Fuchsian functions — could not exist. He tried combination after combination. Nothing held. Then one evening he broke his own routine and drank black coffee before bed. He could not sleep. "Ideas rose in crowds," he wrote; he could feel them colliding, pairing off, until they locked into stable combinations. By morning he had done the opposite of what he set out to do: he had established that a whole class of these functions did exist, built from the hypergeometric series. He needed only a few hours to write out the verification (Poincaré, "Mathematical Creation," 1910).

That should have been the strange part of the story. It was not. A few weeks later, Poincaré left Caen on a geology field trip for the École des Mines, glad to think about rock strata instead of differential equations. At Coutances the party changed to a horse-drawn omnibus. "At the moment when I put my foot on the step," he recalled, the idea arrived, unbidden and complete: the transformations he had used to define his new functions were identical to the ones he had encountered, in an entirely different investigation, in non-Euclidean geometry. He did not check it on the spot — he had a conversation to get back to — but he felt, in his own word, a "perfect certainty." He verified it at leisure once he was home (Poincaré, "Mathematical Creation," 1910; the fullest secondhand retelling is in Shaw, "Henri Poincaré as an Investigator," Popular Science Monthly, 1913).

Nothing about mathematics required him to be riding a bus. He had not been thinking about Fuchsian functions in the previous hour, or so he believed. And yet the answer to a problem he had spent months on arrived in the second it took to lift one foot. Poincaré is the first major scientist to take that fact seriously rather than treat it as a charming anecdote for after-dinner speeches. If the mind can solve a problem while, by its owner's own testimony, not working on it, then whatever "solving" is, it is not simply conscious search performed faster. Something else is going on beneath the desk, and it was time somebody studied it the way one would study any other natural process — by collecting cases, naming the stages, and looking for the pattern that repeats.

Four Stages, Named

That naming job fell to Graham Wallas, a British social psychologist who had never proved a theorem in his life but had read Poincaré and Hermann von Helmholtz — another scientist who had left notes on the odd, sourceless arrival of his own good ideas — closely enough to see the shape common to both. In his 1926 book The Art of Thought, Wallas proposed that creative problem-solving runs through four distinguishable stages: preparation, in which the problem is consciously studied and the relevant facts assembled; incubation, a period — sometimes minutes, sometimes years — in which the problem is set aside and, apparently, worked on without supervision; illumination, the sudden arrival of the idea, often at a moment (a bus step, a half-sleep, a walk) that has nothing to do with the problem's content; and verification, the deliberate, effortful checking of what illumination handed over, because illumination is a hunch-generator, not a proof-generator (Wallas, The Art of Thought, 1926; overview in Wikipedia, "Graham Wallas").

Notice what this model does and does not claim. It does not say incubation is magic, and it does not say illumination can be summoned on demand. It says that the four stages are functionally distinct — that a mind engaged in incubation is doing something different from a mind engaged in preparation, even though the thinker experiences the first as "doing nothing." Wallas had, in effect, taken Poincaré's private confession and turned it into a public, testable claim about how any mind solves any hard problem, whether the problem is a theorem, a poem, or a stuck hinge. It was the first attempt at a science of the solver's own head — not of the problem, but of the person facing it.

The claim needed more than two data points, however brilliant those points were. It got them from a mathematician with a habit even more useful than genius: Jacques Hadamard kept a filing cabinet's worth of curiosity about his own mental habits, and when he could not satisfy it alone, he wrote letters. His 1945 book The Psychology of Invention in the Mathematical Field is built from decades of his own introspection cross-checked against reports he solicited from working mathematicians and physicists, George Pólya and Albert Einstein among them (Hadamard, An Essay on the Psychology of Invention in the Mathematical Field, 1945). Hadamard confirmed Wallas's four stages independently and added a detail that startled his correspondents when they compared notes: several of them, Einstein included, reported that their best mathematical thinking did not use words at all. It arrived as a felt sense of relations between objects — visual, muscular, almost architectural — which then had to be translated into ordinary language and symbols afterward, sometimes with real effort and loss. Preparation and verification are the parts of discovery that happen in a language you could dictate to a stenographer. Incubation and illumination, on Hadamard's evidence, mostly do not.

This is not the whole story of what this book will call reframing — Wallas and Hadamard describe a process happening to the solver, not a deliberate choice the solver makes. But it is the first hard evidence that incubation is real, that a mind can restructure a problem's representation without being asked to and without the owner watching it happen. Chapters ahead will show reframing being done on purpose, as a discipline. This chapter is about what a mind does when reframing happens to it — and about the tools people built once they noticed the pattern and tried to make it happen more often.

What Apes and Candles Taught the Gestaltists

While Poincaré's confessions were circulating among mathematicians, an entirely separate community was reaching the same conclusion from the opposite direction: not by asking brilliant humans to introspect, but by watching animals fail and then, suddenly, not fail.

In 1913, the Prussian Academy of Sciences opened an anthropoid research station on Tenerife, in the Canary Islands, and installed as its director a young Gestalt psychologist named Wolfgang Köhler (Wikipedia, "Wolfgang Köhler"). Over the next several years Köhler set his chimpanzees problems that had a food reward just out of reach — bananas outside the bars, requiring a stick to pull them closer; bananas hanging from the ceiling, requiring two crates stacked to make a step. The prevailing theory of animal learning at the time, championed by Edward Thorndike, held that all such learning was trial-and-error: the animal flails, something works by accident, and the successful action is stamped in by reward. Köhler's chimpanzees did something that did not fit that story. They would sit — visibly, almost theatrically, to Köhler's eye — doing nothing, surveying the crates and the bananas, and then rise and execute the entire stacking sequence in one continuous, confident motion, with no visible trial before the error-free success. Köhler called this insight, and published his findings in 1917 as Intelligenzprüfungen an Menschenaffen, translated into English in 1925 as The Mentality of Apes (Wikipedia, "Wolfgang Köhler"). Whatever this was, it looked much less like accumulated luck and much more like a sudden reorganization of the animal's picture of the situation — the crate stopped being "a thing chimpanzees sit near" and became "a thing chimpanzees stand on."

Köhler's student Karl Duncker pushed the same question into the human laboratory and gave it a name that survives every introductory psychology course: functional fixedness. In the task now known as the candle problem, a subject is handed a candle, a box of thumbtacks, and a book of matches, and asked to mount the lit candle on a wall so no wax drips on the table below. Most people try to tack the candle straight to the wall, or melt wax onto the wall as glue — both fail. Very few think to empty the tack box, tack the box to the wall, and stand the candle inside it, because the box has already been cast, before the subject even starts reasoning, in the role of "container for tacks." When Duncker instead handed subjects the tacks loose, next to an empty box, the solution rate rose sharply — the box, no longer costumed as a container, became visible as a shelf (Duncker, "On Problem Solving," Psychological Monographs, 1945; Wikipedia, "Candle problem"). Functional fixedness is the mind pre-loading a representation — assigning an object one role and refusing, without help, to try another — and it is the same failure that trapped every petitioner at Gordium: a state space silently narrowed by habit rather than by the actual rules of the problem.

Norman Maier, working at the same moment on the opposite side of the Atlantic, went further and asked a harder question: does insight need to be conscious to work? His two-string problem hangs two cords from the ceiling, too far apart for a subject to hold one and reach the other; a pair of pliers lying nearby can be tied to one cord and swung as a pendulum, letting the subject catch it while holding the first. Subjects who stalled were given a hint disguised as an accident: Maier would walk past and "accidentally" brush one of the strings, setting it swinging. Most who then solved the problem could not say why the answer had occurred to them — in Maier's post-experiment interviews, only around a third of those helped this way recognized they had been cued at all (Maier, "Reasoning in Humans II," Journal of Comparative Psychology, 1931; replication and discussion in Landrum, "Maier's (1931) Two-String Problem Revisited," 1990). The unconscious, in other words, can accept a piece of evidence, use it to restructure a solution, and hide the receipt. Wallas's incubation was not a metaphor. It was a mechanism, and Maier had just shown it operating below the floor of self-report.

Taken together, Köhler's chimpanzees, Duncker's candle, and Maier's swinging string describe the same discovery three ways: solving is not always, or even usually, a matter of trying harder inside a fixed picture of the problem. Often the picture itself is the obstacle, and the picture can change — sometimes all at once, sometimes with help the solver never notices — without the solver doing anything you could call "more search."

Pólya's Toolbox

All of this — Poincaré's testimony, Wallas's stages, Hadamard's survey, the Gestalt animal and human labs — described what happens to a solver. None of it told a stuck student what to do on a Tuesday afternoon with a geometry problem due Wednesday. That gap was closed, decisively and for the first time in a form ordinary people could use, by a Hungarian mathematician who had spent years watching himself and his students think.

George Pólya's How to Solve It, published in 1945, opens with a plan of embarrassing simplicity: understand the problem, devise a plan, carry out the plan, look back. Understanding sounds too obvious to need saying, and Pólya's own point is that this is exactly why students skip it and then wonder why they are stuck; his prompts — what is the unknown, what are the data, can you restate this in your own words — exist because "understand the problem" is a stage, not a formality (How to Solve It (1945), summarized at Wikipedia). But the four phases were only the skeleton. The body of the book is a dictionary of some sixty-odd named heuristics — practical, unguaranteed methods for finding the plan in step two — and it is these that made the book a fixture on desks far from any mathematics department.

A few of Pólya's heuristics deserve their names remembered, because they recur, disguised, throughout the rest of this book. Analogy: if you cannot solve the problem in front of you, have you solved one that resembles it? The auxiliary problem: can you invent an easier, related problem, solve that one, and use the solution as a ladder? Working backward: start from the goal and ask what state would have to precede it, then what would have to precede that — a heuristic Pólya credits with turning "find a path" into "find a path," read right to left, which is sometimes the shorter search. Specialization and generalization: if the problem is too abstract to get a grip on, try a small concrete case; if it is too narrow to see the structure, ask what larger family it belongs to. None of these procedures guarantees an answer. All of them narrow the region of the problem space worth searching, by betting — on the strength of experience with other problems — that the path is more likely to run through analogy, or backward from the goal, or through a simpler cousin, than through blind trial.

The book's afterlife is its real argument. Marvin Minsky, writing in 1961 at the dawn of artificial intelligence research, told his readers flatly that "everyone should know the work of George Pólya on how to solve problems" (quoted in Wikipedia, "How to Solve It"), and Allen Newell — who with Herbert Simon would go on to build the problem-space formalism this book leans on throughout (chapter 00) — later wrote an entire paper examining exactly how much of Pólya's heuristic vocabulary a computer program could be made to use, and how much resisted translation into code (Newell, "The Heuristic of George Polya and Its Relation to Artificial Intelligence," 1981). Newell's conclusion was double-edged: Pólya's heuristics were recognized by everyone in the field as the right kind of idea, and yet decades of effort proved how hard it was to turn "notice an analogy" into an algorithm. A heuristic that a human applies with judgment is not automatically a heuristic a machine can apply at all — a tension this book will meet again when the search moves from human heads into silicon (chapters 04 through 06, and again in chapter 10).

Forty Thousand Patents and One Idea

Half a world away, and unaware for years that Pólya's book even existed, a Soviet inventor was solving the identical problem from the opposite trade. Genrich Altshuller worked reviewing patent applications for the Soviet Navy, and it needled him that engineers kept "inventing" solutions that had already been invented, in different industries, under different jargon. He began, starting in 1946, to read patents not for their content but for their shape: what kind of difficulty had the inventor faced, and what kind of move had resolved it. Building the classical contradiction matrix, Altshuller and his collaborators examined on the order of forty thousand patents drawn from a much larger pool — one contemporary account puts the source pool at roughly four hundred thousand — screening for the subset that represented genuine breakthroughs rather than routine refinements (Savransky, "A Few Words About the Altshuller's Contradiction Matrix," The TRIZ Journal, 1997). Exact patent counts vary across the movement's own retellings — later organizers cite larger cumulative totals as the research continued for decades after 1946 — but the order of magnitude, tens of thousands sifted by hand and eye, is not in dispute (Wikipedia, "TRIZ").

What Altshuller extracted from that reading was TRIZ — Teoriya Resheniya Izobretatelskikh Zadatch, the Theory of Inventive Problem Solving — built on a single organizing unit he called the technical contradiction: improving one parameter of a system (make it lighter) degrades another (it becomes weaker). Altshuller's claim, radical for engineering in 1946, was that a genuine invention is not a compromise between the two — thinner and therefore weaker — but a move that escapes the trade-off altogether, and that the moves which escape trade-offs repeat across industries, so that the solution to a chemical-plant problem might be structurally identical to a solution already patented in bridge-building. He and his colleagues eventually distilled this into forty named inventive principles and a matrix for finding, given any pair of colliding parameters, which of the forty was likeliest to apply (Wikipedia, "TRIZ").

The convergence with Pólya is not coincidence dressed up as destiny — it is what you would expect if both men had, independently, put their hands on the same underlying fact: across wildly different domains, the number of genuinely different move-types is far smaller than the number of problems they solve. Pólya found perhaps sixty for mathematics. Altshuller found forty for engineering. Neither number is sacred, and this book will not pretend the two lists are secretly the same list. But both men had discovered that a stuck solver rarely needs a new idea invented from nothing; far more often what is needed is a shift into a known kind of move, correctly matched to the kind of obstacle in front of them.

Invert, Always Invert

One of those move-types is old enough to predate the whole psychological turn, and simple enough to state in five words: when stuck, try the opposite direction. The German mathematician Carl Gustav Jacob Jacobi is widely credited with the maxim man muss immer umkehren — "one must always invert" — reportedly drilled into his students as a standing instruction whenever a differential equation resisted a direct attack: stop asking what the integral of this expression is, and ask instead what expression this could be the integral of. The attribution deserves the same caution this book gave the sword and the pin at Gordium: the phrase survives only as something Jacobi "is said to have" told his students, quoted at several removes and popularized outside mathematics only much later, so the wording — and possibly the entire anecdote — should be read as folklore with a mathematical core, not a verified transcript (Mathematical Association of America, "Quotations in Context: Jacobi"). What is not folklore is the technique's power: Jacobi and Niels Henrik Abel both made their names by inverting elliptic integrals — asking not "what value does this integral take" but "what limits produce this value" — and finding, on the far side of the inversion, a whole new family of well-behaved functions waiting to be named.

Inversion belongs on Pólya's list in spirit if not by his exact label — it is a cousin of working backward — and it is the same move Alexander made at Gordium in chapter 00, whether by sword or by pin: refuse the direction the problem was handed to you in, and search the reverse instead. It will resurface, formalized into an algorithm rather than a maxim, when this book reaches dynamic programming's habit of solving a problem from its last step backward (chapter 04).

The Searchlight, Not the Guarantee

Step back from the individual names and a single shape emerges from this whole chapter. Wallas's incubation, Hadamard's wordless thought, Köhler's insight, Duncker's fixedness, Maier's hidden hint, Pólya's dictionary, Altshuller's contradictions, Jacobi's inversion — none of them is a procedure that finds the shortest path by exhausting the space and comparing every route. Every one of them is a bet, made in advance of full information, about where in the space the short paths are statistically likely to hide. Analogy bets that a problem's structure has already been solved once, somewhere else, in different clothing. Working backward bets that the goal state constrains its predecessors more tightly than the start state constrains its successors. Altshuller's contradiction matrix bets that whatever principle broke a trade-off in aircraft design will also break one in chocolate packaging.

This is why the heuristic age is not the end of this book's argument but its opening move. A heuristic does not change the problem space — it does not throw away states, redefine the goal, or invent new operators, the way Euler's abstraction of Königsberg did in chapter 02's aftermath, or the way Snow's re-definition of a "case" will in chapter 07. A heuristic simply shines a searchlight into a fixed space and says: look here first. That is real, useful, and teachable — Pólya's book has stayed in print continuously since 1945 for exactly that reason (Wikipedia, "How to Solve It") — but it is a smaller claim than reframing, and honesty requires saying so before the next four chapters show what happens when solvers stop shining lights into the given space and start rebuilding the space itself.

The Move

Move 03 — Carry named tools.

A heuristic is not a guarantee. It is a bet, made by someone before you, about where in a space of possibilities the short paths tend to cluster — and every bet has a failure mode you should know before you place it.

• Keep a short deck, not a long one. Pólya's dictionary runs past sixty entries and Altshuller's matrix has forty principles, but almost no solver holds more than a handful in working memory under pressure. Choose a working set — analogy, working backward, the auxiliary problem, inversion are a defensible starting four — and know them cold rather than knowing a hundred shallowly.
• Name the failure mode with the tool. Analogy fails when the surface resembles a solved problem but the structure does not. Working backward fails when the goal state under-constrains its predecessors. Functional fixedness is not a flaw in the solver, Duncker and Maier showed — it is a predictable cost of any heuristic that pre-assigns roles to objects, including the useful ones.
• Protect incubation on purpose. Poincaré did not summon the omnibus insight; he set the problem down and let a fortnight of grinding preparation finish itself off-screen. If a plan has failed after honest effort, walking away is not surrender — it is deploying the one stage of discovery, per Wallas and Hadamard, that specifically requires you to stop looking.
• Treat a heuristic's silence as information. When your whole deck comes up empty against a problem, that is evidence the space itself is wrong, not that you need a sixty-first heuristic. That is the cue to stop searching within the given frame and start asking what would happen if you changed it — which is where this book goes next.

Next — Chapter 04 · War and the Science of Search: heuristics compressed the search inside one mind. The Second World War would industrialize it — turning the shortest path into a discipline with a budget, a uniform, and a body count.