The Gardener's Algorithm: Optimizing for What Cannot Be Measured
The Gardener's Algorithm
June 2028
The brief was straightforward: design a memorial garden for the grounds of a children's hospital in Medellín, Colombia. The garden would serve families waiting for news, families grieving outcomes, and families celebrating recoveries. It needed to hold all three states simultaneously.
Landscape architect Camila Restrepo had designed seventeen public gardens. This was the first where she used generative AI design tools — not by choice, but because the hospital's foundation had funded the project through a grant that required "AI-assisted design methodology."
The tool was called TerraGen, and it was impressive. Camila fed it the site dimensions, soil composition, local climate data, sun exposure patterns, and foot traffic projections. TerraGen generated 340 garden layouts in twelve minutes, each optimized across fourteen quantifiable parameters: accessibility, biodiversity index, maintenance cost, water efficiency, thermal comfort, visual interest distribution, and eight others.
Camila reviewed the top twenty layouts. They were all good. None of them was right.
The unquantifiable constraint
"The layouts are optimized gardens," Camila told her colleague. "But a memorial garden is not an optimized garden. An optimized garden makes you comfortable. A memorial garden makes you feel something. Those are different goals."
She went back to TerraGen and tried to add her missing constraint. The interface accepted quantifiable inputs: numbers, ranges, binary conditions. It did not accept: the garden must feel like grief becoming hope.
Camila tried to decompose the feeling into components. Journey? A path that began in enclosure and opened to sky. Contrast? Dark planting yielding to light. Water? Present throughout, changing from still to moving. Height? Low at the entrance, rising gradually.
Each decomposition captured something. None captured the whole. The feeling she wanted was not the sum of its components. It was the components in a specific relationship — a relationship she could recognize but not specify.
The negotiation
What followed was a three-month conversation between Camila and TerraGen that was neither human design nor AI optimization. It was negotiation.
Camila would modify a TerraGen layout by hand — moving a tree, adding a bench, changing a sightline. TerraGen would analyze her modification, report its impact on the fourteen quantifiable parameters, and suggest adjustments that preserved her change while recovering lost optimization.
Sometimes TerraGen's adjustment revealed something Camila hadn't considered: a sightline she had blocked inadvertently, a drainage problem her tree placement created. She would accept the correction and refine her intent.
Sometimes Camila's modification violated an optimization parameter that she didn't care about. She would override TerraGen and explain (in her notes, not to the machine) why the parameter didn't matter here: "Foot traffic efficiency is not the point. The path should meander. Grief does not walk in straight lines."
The negotiation produced a design language. Camila developed a practice she called "constraint painting" — using TerraGen's quantifiable constraints as the canvas and her unquantifiable intentions as the brushstrokes. The AI held the structure. She held the meaning.
The garden
The final design took elements from seven TerraGen layouts and none of them entirely. It featured:
A entrance sequence through tall, dark bamboo that narrowed the visual field and muffled sound — creating enclosure before the garden opened. TerraGen had flagged this as suboptimal for accessibility and thermal comfort. Camila overrode both: the enclosure was the emotional entry point.
A central water feature that began as a still, reflective pool (grief: the stillness of what cannot be changed) and transitioned through a gentle channel to a moving stream (hope: the continuity that carries forward). TerraGen had optimized the water system for irrigation efficiency. Camila re-optimized it for emotional narrative.
A grove of yarumo trees — silver-leafed, native to the region — planted in a pattern that created dappled light in the morning and full illumination at midday. The transition from shadow to light tracked the sun's movement and, metaphorically, the movement from waiting to knowing. TerraGen had suggested a uniform canopy for consistent shade. Camila wanted inconsistency: the play of light was the garden's breath.
Benches placed not at optimal rest intervals but at emotional decision points: the moment the path turns, the moment the water appears, the moment the canopy opens.
June 11, 2028 — Camila's design journal
The garden opens next month. I walked through the finished site today. A family was already there — a mother and father sitting on the bench by the still pool. They didn't know the garden was designed. They thought they had just found a quiet place.
That's the test. If the design is visible, it's failed. The experience should feel discovered, not delivered. The AI helped me get the engineering right — drainage, soil, sun, paths. That's the structure. The meaning is in the choices the structure serves: where the path turns, where the light falls, where the water stills and where it moves.
TerraGen can optimize fourteen parameters. I can optimize one: does this space hold what people bring to it? That parameter cannot be measured. It can only be felt.
We spent three months negotiating — human and machine, meaning and measurement, the nameable and the unnameable. The garden is the treaty. Not my design. Not the machine's design. The space between us, given form.
I think every collaboration with AI will eventually arrive at this negotiation. The machine will optimize what can be counted. The human will insist on what can't be. And the work will live in the tension between them.
That tension is not a problem to be solved. It is the garden itself.
Part of The Interface series. For another story of human meaning meeting machine optimization, see The Ceramicist and the Kiln. For the body-knowledge dimension of human-land relationship, see The Soil Whisperer.
