The myth of architectural creativity

Part One: Pattern Languages already sketches the standard model of architectural design: a designer internalises all the available information, thinks really hard, and produces a design as a creative act — the greater the genius, the better the result. It also makes the obvious observation that on the evidence of the cities around us, either our geniuses aren’t good enough, or they’re not being given the chance to prove it. This article goes further: creativity, in the sense architecture usually means it, isn’t a method at all, and the people most invested in it know that better than anyone.

Creativity as religion

Even critics who have a genuinely good grasp of what’s wrong with architecture tend to reach for the same fix: if only the right, enlightened people were doing the designing instead of the wrong, ignorant ones, we’d get better cities. That’s not a solution, it’s a restatement of the problem — it just relocates the mysticism from the buildings to the architects, who are now expected to become some kind of ego-less, all-knowing zen master who does the right thing rather than the wrong thing.

People find it very hard to let go of the idea of a ‘creative spark’, without which a thing is supposedly lifeless. But this is a religious concept, not an empirical one. Evolution created every living thing that exists through nothing but trial and error — there was no creation, no genius, no designer, life is not a work of art. And removing the creator from that picture doesn’t leave you with something hollow and dead; life is the standing counterexample. If evolution can produce that without a designer, the claim that a building specifically needs one starts to look less like an argument and more like an article of faith.

The ‘sitting down’ problem

There’s a sharper way to see why the creative-genius model fails, and it comes from watching what happens when you point it at Alexander’s own book. Plenty of people find A Pattern Language obviously true, but complain that it doesn’t fit how design actually happens — nobody sits down, opens the book, and designs a building from it. Alexander’s own reply, more or less, was that the problem is the “sitting down” part, and also the “design” part: pattern languages were never meant to be applied by someone consulting a checklist at a desk.

What Alexander’s actual creative method amounted to was: internalise all the relevant patterns, think very hard, and produce a result — the exact “creative genius” method described above, just performed by someone unusually well-read in his own patterns. And it’s telling that Alexander himself apparently abandoned this as unworkable not long after writing A Pattern Language. Contrast that with something like the École des Beaux-Arts, which really did teach architecture as a series of procedural steps that reliably produced a workable building — a genuine design method, whatever else you think of the results. The lesson isn’t that no design method is possible, it’s that a design method has to be a teachable procedure, not an act of internalised genius, even when the genius in question wrote the book.

Superstition and bravado

None of this is an argument for building an AI architect — that would be a monster even if it could be built, and it isn’t what this project is trying to do. But the case for not trying to duplicate the architect’s design method rests on a stronger claim than “AI can’t do it”: there is no architectural design method to duplicate in the first place. What passes for one is a collection of superstitions and bravado dressed up under the name ‘creativity’, a cover story for the fact that nobody involved really knows what they’re doing. When a project is unusually large or prestigious, the standard response is to hire an especially creative wizard who can supposedly perform creativity at industrial scale — a belief in modern-day witchcraft that the appalling average quality of the results should have discredited long ago. Even the idea that not enough “good design” is happening is, at bottom, a sales pitch for the design profession as a whole, not a diagnosis.

Alexander’s real insight — the one that actually survives scrutiny — is that human imagination works well at a human scale, standing in an actual location and imagining the possibilities immediately around you, combined with iteration, trial and error, and full-size mock-ups. What it doesn’t do is scale up: the idea of one especially gifted person conceiving an entire large project as a single insight is exactly as unlikely coming from a self-styled Alexandrian master as it is coming from a star architect’s atelier.

The eugenics objection, answered in advance

Calling this process ’evolutionary’ invites an obvious, slightly queasy comparison, so it’s worth heading off directly: this is not eugenics, which is specifically the idea that directed evolution should be applied to people. Directed evolution isn’t remotely controversial applied to roses or broccoli, and a house is not alive — a simulated house even less so. The “somebody knows what’s right” role in this system is filled by the Pattern Language itself, which is about as far from an unaccountable authority as a design criterion gets: every pattern has to be justified, and every pattern has to be open to being disproved.

There’s a related worry worth taking equally seriously: that an efficient, systematised design process is just battery-hen architecture with better branding. But “efficient” battery-hen housing is only efficient if you don’t count the whole system, and a fitness process built entirely around human-scale patterns — light, privacy, adaptability, a front door, outdoor space — is structurally incapable of producing that outcome, because none of it scores well against the criteria the process is actually optimising for.

Straight lines aren’t the enemy

It’s worth being precise about what the actual target is here, because it isn’t geometry. Straight, square and flat are not inherently bad at a human scale — plenty of humane, historical buildings are all three. The real problem is design as spectacle: monumentalism, relentless repetition, bold eye-catching shapes, minimal detail. Those aren’t neutral aesthetic choices, they’re the specific attributes of advertising and propaganda, and they’ve quietly colonised the architectural profession’s idea of what looks impressive.

Chasing a science-fiction imagery

A concrete example of that colonisation: architects have spent decades pursuing an imagery of “modernity” that owes more to science fiction than to any actual science, and science-fiction visionaries have an almost perfect track record of being wrong about the future. One case makes the point unusually well — a design tool for a city masterplan in Turkey turned out, on inspection, to be built on software originally developed to simulate flowing hair for computer animation, with the streets traced over the resulting hair-like patterns. Software genuinely capable of designing better-performing streets and buildings already exists. It doesn’t get used for this, because better-performing buildings don’t reliably produce the futuristic look being chased — the software is doing exactly what it was asked to do, it was just asked for the wrong thing.

Two ways out

A Pattern Language is true, in the sense that its individual patterns hold up, but true doesn’t mean usable: it doesn’t supply a design method, and any method built on top of it by hand is necessarily laborious, exactly the “internalise, think hard, produce” grind that even Alexander couldn’t sustain. That leaves two real ways forward. One is to adopt an existing tradition, because traditions that have survived centuries of use already encode a huge amount of hard-won pattern knowledge, cheaply, without anyone having to rediscover it from scratch. The other is to get a computer to do the laborious part — not to be creative, which was never really on offer from anyone, but to do the relentless, repetitive, pattern-by-pattern iteration that a human designer can’t sustain across an entire building. That’s the part this project actually does.

Creative Commons License

Last updated 15 September 2026.