The useful thing here is that Curtis is already circling a problem very close to the one we were reaching in “The Machine That Wanted Nature to Balance”: not simply that machines simplify reality, but that we increasingly select realities which behave well inside machine descriptions.
- The RDB front
My read of Curtis’s actual proposition is narrower—and more interesting—than “computers are bad”:
Human beings repeatedly invent models of reality which promise freedom from uncertainty, hierarchy and political responsibility. Then, when those models become infrastructure, we mistake the behaviour of the model for the behaviour of reality.
The machine does not have to enslave anyone.
It only has to become the thing through which enslavement, freedom, nature, individuality, society and even the self become intelligible.
That is the connective tissue running through all three films.
But underneath, Curtis is following one object:
the fantasy of a self-regulating system.
So:
economic system → ecological system → biological system → human subject
is not four unrelated stories.
It is one recursive operation:
Convert uncertainty into a system. Convert the system into components. Convert the components into variables. Then claim that the variables explain the whole.
- The really important word is “balance”
This is where Curtis is considerably stronger than the usual anti-tech reading.
The machine does not merely want control.
It wants balance.
That matters because control implies an agent.
Balance implies that nobody is responsible.
A balanced system supposedly regulates itself.
The market regulates itself.
The ecosystem regulates itself.
The network regulates itself.
The genes regulate behaviour.
The individual follows incentives.
The feedback loop corrects deviation.
And suddenly nobody has to say:
“I chose this.”
Curtis explicitly frames the dream as systems that stabilise themselves through feedback, whether in nature, society or capitalism.
This is exactly where I would extend him.
Curtis says:
Humans project machine fantasies onto nature.
I would push it one stage further:
Machines reward humans for projecting only those aspects of nature that can be represented as machine-stable.
That's a much nastier loop.
Because now the process is recursive:
world → model → infrastructure → altered world → model appears correct
At that point the machine isn't merely describing reality.
It's participating in its decomposition.
“Decomposition” ordinarily sounds like destruction.
But analytically it can mean something more precise:
a whole is broken into components so that relationships between components can be represented.
That's exactly what systems thinking does.
And that's exactly what computation needs.
The danger isn't decomposition itself.
Science cannot function without decomposition.
The danger is the silent transformation:
“This has been decomposed for analysis”
into
“This is actually composed this way.”
And once you've done that, something incredibly important has happened:
the relationships that do not survive decomposition disappear from the ontology.
Not necessarily because they're false.
Because they are not machine-legible in the chosen representation.
That's a substantially deeper criticism than “computers simplify things.”
Curtis's documentaries often contain an odd tension:
he is extremely good at identifying conceptual machinery, but his own films are also machines.
In other words:
Curtis is analysing machine logic with a cinematic machine.
And therefore All Watched Over performs the problem it diagnoses.
That's not a flaw.
It is part of why the film survives.
But it means we shouldn't treat Curtis as standing outside the machine looking in.
He's inside it, building another model.
And his model has its own equilibrium.
Analysis—Isshin