EvoQuantum Computing

Solve a problem by growing the organism that can survive it: the problem to be solved is the final environment

The lab's board comes to life: the biological sieve on the left, the runtime engine on the right. In a vent's iron-sulfur labyrinth the soup copies itself until it turns green, then grows along light waveguides through three survive-or-die sieves. It is copied into two sandboxes, one property each, and the two are pushed into a membrane-less chamber where they swap parts and fuse. Its side branches fold in, and one organism reaches the final environment. The camera follows it inside, where it swims and feeds.

What this is

A way to compute an answer that is grown rather than written. The run starts from the same primordial soup every time: a wall-less gel in an alkaline vent's iron-sulfur labyrinth, before anything has been selected. A trail of environments then steers it, with light waveguides, fuel funnels and survive-or-die sieves, but never touches the soup itself. To get two properties at once, the mass is split into two sandboxes, each grown toward one, and then pushed back together in a chamber where neither half can live alone, so they fuse. The last environment is the problem itself. Whatever survives it is the answer. A runtime engine runs all of it at warp speed from outside while, inside, time stays slow and natural.

The technical shape

Principle: environment-only steering · the soup is never opened, edited or scored · each environment is a sieve: the arrangement survives it or it doesn't
Env0: identical for every run, whatever the problem: an alkaline hydrothermal vent, iron-sulfur labyrinths, pH 9.0 vent fluid against pH 6.0 ocean water, a proton gradient for power, templates copying without enzymes
Trail: Env1 → Env2 → Env3 in series · analog waveguides (structured light) carve the path · fuel funnels (chemotactic gradients) pull · binary sieves prune
Split: the mass forks into isolated sandboxes: the Main Branch selects Property 1, Sub-Branch A Property 2 · no cross-contamination
Fusion: one membrane-less chamber with the baseline fuel gone and one gradient for each property · a cross-catalytic light bridge between them · the branches swap parts and fuse into one hybrid
Answer: the problem to be solved is the final environment · the organism that survives it is the solution
Runtime: a time-dilation sandbox: warp speed outside, slow and analog inside · analog steering knobs and passive valves · SieveEngine over a 64-dimension probability mass
Stage: concept specification · the architecture and a first SieveEngine sketch drafted

The deep end · full technical outline

Outline v2 · expanding

The problem is the place

Most computing writes down the steps to an answer. This turns that around. The problem is set as an environment, a set of conditions only the right structure can live in, and the system grows something until it survives there. Nothing in the run is told what the answer looks like. The environments say only what dies. The same idea is described for a qubit-coherence chamber, a massive data graph and a structural-biology map: in each case the target is a place, and the organism that lives in it is the answer.

Env0, the same every time

Every run begins in one invariant environment, whatever the final problem is. It is modelled on where life had not yet started: an alkaline hydrothermal vent with iron-sulfur labyrinths, warm pH 9.0 vent fluid rising into cold pH 6.0 ocean water, and the proton flux across the mineral walls as the power supply. Short strands copy themselves on templates without enzymes. Nothing is steered in Env0. Selecting for the problem this early would shrink the space before it had grown, so Env0 runs at maximum permutation density until the soup is self-sustaining: the green state.

Wall-less, so nothing is lost

The soup is a continuous probability mass in a fluid gel, not a crowd of separate cells. With no walls, any part can pass to any other, so a useful piece found in one place is available everywhere. The system never simulates single agents, lines of code or lineages. It holds the whole mass as one high-dimensional field and watches the numbers change as each environment acts on it.

The micro-steering trail

From Env0 the mass moves through a series of small environments, Env1 to Env2 to Env3, each a slight shift from the one before so nothing is shocked. Three kinds of instrument do the steering. Analog waveguides, geometries of structured light, carve the paths it can grow along. Fuel funnels, chemotactic nutrient gradients, draw it forward. Binary sieves are absolute survive-or-die boundaries. What can't hold up there is left behind, at warp speed.

Split, then fuse

An answer often needs two properties that pull in different directions. Grown together, one would crowd out the other. So the mass is copied into two isolated sandboxes. The Main Branch is steered toward Property 1, for example stable electron-transport corridors, and Sub-Branch A toward Property 2, for example spatial-hashing memory structures. Each then enters the final-final environment. It has no walls, so nothing is rejected. The baseline fuel is gone, and each of its two gradients can be processed by only one of the branches. A cross-catalytic light bridge runs between them, a corridor that needs both. Neither half can live alone, so they swap parts and become one organism with both properties.

The runtime engine

The engine is the outside of the experiment. A time-dilation sandbox runs the external clock at warp speed while, from the soup's point of view, every change is slow and analog. Analog steering knobs and passive valves set the environments, and the engine reads only global signals: entropy expelled, energy throughput. It never opens the soup to look inside, which would contaminate it. When it senses a hard bottleneck it slows the clock to give the permutations time. The first sketch, SieveEngine, holds a 64-dimension probability mass and runs each shift until the mass either fits or collapses below a structural threshold.

What is still open

This is an architecture imagined in October 2026, not a working machine. The SieveEngine is a numerical sketch, not chemistry. No one has yet steered a pre-cellular soup through designed environments into a working structure, nor mapped a real problem, such as a qubit-coherence target, onto a physical environment an organism could be grown to survive. Open questions include which physical attributes the 64 dimensions should stand for, and how a dashboard can show the mass deforming at each sieve without looking inside.

Related in the lab

Environment steering

Restaurant Quality Food — At Home →: Shares environment steering — getting the organism you want by choosing only the environments it must survive, never the organism itself.

Built at this bench

The machine, the instrument, the film that explains it — all designed and built in-house. The same bench works for hire.

Services →