The comprehension gap
System complexity has outrun the human mind by four orders of magnitude. Scroll to watch the wedge open.
For fifty years, the systems we build have grown faster than our capacity to hold them in mind.
We have answered that divergence the only way we knew how — by enlarging the engineer. More abstraction. More tooling. More documentation laid over an artefact that keeps getting larger underneath it.
This feature makes one chart argue one claim: the gap is now structural, and it will not be closed by making engineers bigger.
A system you could hold in one head
The first decade. System complexity and human comprehension start at the same point. The lines are together.
An order of magnitude, already
Twenty years in, complexity is 32× the baseline. Comprehension has barely moved off 1×. The wedge begins.
A thousand to one
The gap is no longer a matter of effort. No amount of reading closes three orders of magnitude.
Sixteen-thousand to one
The ceiling becomes visible. Independent re-evaluations collapse the headline capability numbers — but the complexity curve does not bend.
The bridge, or the gap is permanent
The substrate is either the bridge by 2030, or the divergence is a permanent feature of how we build.
We don't enlarge the engineer. We reshape the substrate.
There is a second curve that makes the first one urgent: energy.
Every closed gap has a price measured in joules. The substrate that finally attends must also be the substrate that can afford to — and on that axis, today's machine intelligence sits six orders of magnitude above the brain it imitates.
One nanojoule per event
Transformer inference on GPUs. The reference point — and the most expensive way we have to think.
Two orders better
Loihi and TrueNorth-class silicon. State of the art — and still far from the only existence proof that matters.
The real benchmark
Mammalian cortex, at roughly 100 femtojoules. Not the next increment — the target.
Below biology
A self-organising memristive substrate, under 10 fJ. Five orders of magnitude from where we started.
Close the gap not by enlarging the engineer, but by reshaping the substrate they navigate.