Causality on Programmer.ie: Modern AI programming

12: The Closest Thing We Have

Tue, 11 Aug 2026 10:43:00 +0100

Before building anything more complicated, we should steal everything we can.

Not code.

Ideas.
Experiments.
Failures.
Mechanisms.

So let’s ask an obvious question:

How far has computation already gone without us deliberately building the organism?

There is no objective answer.

Different Artificial Life systems demonstrate different properties.

Evoloops demonstrated Darwinian evolution of self-reproducing structures inside a deterministic cellular automaton.[1]

Flow-Lenia produces spatially localized continuous structures with complex behaviour, conserves mass, and allows parameters governing different pattern dynamics to become localized inside the world itself. Researchers have measured emergent evolutionary activity in the resulting system.[2]

13: Is It Really Reproducing?

Tue, 11 Aug 2026 12:48:00 +0100

Something strange happened while we were looking at Outlier.

We had started with a question about reproduction.

Then I watched the simulation.

And I thought:

That looks like flocking.

That is exactly the kind of observation that can get us into trouble.

Humans are extraordinarily good at seeing things in motion and assigning meaning to them.

We see:

a shape
another shape
movement
coordination

and very quickly turn that into:

16: Before There Are Messages

Wed, 12 Aug 2026 02:38:00 +0100

In the last chapter, the crystal acquired a past. Not memory in the biological sense. Not recollection. Not understanding. A recoverable past.

We learned that a complete checkpoint could continue the process exactly, that an event history could reconstruct the route by which the present morphology had formed, and that the same checkpoint could be used as a branch point for controlled alternative futures.

That gave us something new:

A present that could be resumed.
A past that could be reconstructed.
A future that could be forked.

The next temptation was obvious. If one Digital Crystal can preserve state and history, perhaps two crystals can communicate.

17: How Does the Crystal Respond to Perturbation

Thu, 13 Aug 2026 12:38:00 +0100

The crystal could hear a pulse.

That much we already knew.

In the previous experiment, a single received bit was inserted into the continuation of a Digital Crystal. Everything else was held as closely as possible to the same conditions. The resulting crystal changed.

Not metaphorically. Not because a classifier happened to prefer one label. The intervention altered the subsequent growth of the receiver.

That gave us one of the weakest but most useful forms of causality available to an experimental system:

18: Can Experience Change the Material?

Wed, 12 Aug 2026 16:40:00 +0100

The crystal had a past.

But we had learned to be careful with that sentence.

In the previous chapter, a perturbation changed the crystal’s future. Under a carefully controlled counterfactual comparison, the consequence of a pulse could remain locally visible for several updates.

That was causality.

It was not memory.

The distinction left us with a smaller question.

Not:

Can the crystal remember?

But:

Can experience change the material?

19: Can the Crystal Tell Two Pasts Apart?

Wed, 12 Aug 2026 18:28:00 +0100

The crystal could carry a mark from its past.

Chapter 18 established that much.

A local event could change the material.

The change could persist.

It could propagate.

And, while it remained in contact with the moving growth frontier, it could alter later construction.

But that still left a much harder question.

Not:

Did something happen before?

But:

What happened before?

That distinction sounds small.

It is not.

20: What Happens When the Crystal Can Lose Material?

Wed, 12 Aug 2026 18:49:00 +0100

Until now, the crystal had lived in an impossible world.

Once a cell became occupied, it stayed occupied.

Forever.

The frontier moved outward.

Old material accumulated behind it.

Nothing eroded.

Nothing vanished.

Nothing exposed the interior again.

That assumption had quietly shaped almost every result since Chapter 11.

So Chapter 20 removed it.

We added one rule:

occupied
↓
empty

with some small probability.

Nothing else.

No repair mechanism.

No resource system.

22: Is the Crystal a Thing or a Flow?

Wed, 12 Aug 2026 21:47:00 +0100

Chapter 21 ended with a question that sounded obvious.

Is there actually one thing here?

For most of the book, we had spoken naturally about:

the crystal

as though the noun had already been earned.

There was a connected occupied structure.

It grew.

It could be perturbed.

It could carry historical material.

It could lose material.

Lost sites could be reoccupied.

Under finite computational opportunity, large amounts of material could turn over while the process continued.

23: What Does One Attachment Cause?

Thu, 13 Aug 2026 00:35:00 +0100

At the end of Chapter 22, the Digital Crystal had stopped looking like a simple bounded thing.

We had looked for a privileged region whose future belonged especially to itself.
We had not found one.

What survived instead was locality.

Change something here, and the consequences are stronger nearby than far away.

That left a new possibility.

Perhaps the important object was not the crystal as a body.

Perhaps it was the process occurring across its changing interface.

24: Where Is Causal Gain Created?

Thu, 13 Aug 2026 10:03:00 +0100

Chapter 23 ended with a controlled causal effect.

Take the same Digital Crystal checkpoint.
Take the same eligible frontier cell.
Give both futures the same environment and the same cell-keyed randomness.
Then force the cell to attach in one future and prevent it from attaching in the other.

SAME CHECKPOINT
SAME CELL
SAME FUTURE RANDOMNESS

FORCE x
vs
PREVENT x

↓
MEASURE DIFFERENCE

The intervention changed subsequent construction.

25: How Does Finite Computation Create Non-Local Coupling?

Thu, 13 Aug 2026 10:49:00 +0100

Chapter 24 ended somewhere we did not expect.

We began by looking for a local causal-gain field.

We found something else.

A local intervention changed the frontier.

The frontier changed which candidates entered a fixed global evaluation budget.

And distant expected construction changed immediately, even though the local transition rule could not propagate influence that far in one update.

That gave us a new candidate substrate law:

LOCAL FRONTIER CHANGE
↓
FIXED GLOBAL EVALUATION BUDGET
↓
CANDIDATE SUBSTITUTION
↓
FAR-FIELD REDISTRIBUTION

The mechanism looked real.

26: How Does Finite Selection Route Causality?

Thu, 13 Aug 2026 15:46:00 +0100

Chapter 25 established a non-local effect in a system whose transition rule was local.

A perturbation at one frontier site changed which distant candidates received evaluation slots.

When candidate subsampling disappeared, the far-field effect disappeared with it.

The mechanism was:

LOCAL FRONTIER CHANGE
↓
FINITE EVALUATION SLOTS
↓
CANDIDATE SUBSTITUTION
↓
FAR-FIELD REDISTRIBUTION

That result answered one question.

It created another.

If finite selection changes where causal opportunity is expressed, does it also change how much causal consequence accumulates?

27: Can Stored Material History Redirect the Future?

Thu, 13 Aug 2026 20:38:00 +0100

By the end of the previous chapters, the Digital Crystal had become harder to describe with ordinary biological language.

It could grow.
It could lose material.
It could rapidly reuse vacated sites.
Finite computation could couple distant regions by redistributing evaluation slots.
Local causal influence could be measured directly.

But none of those results required an organism.
None required an individual.
And none required memory.

That distinction mattered.

It would have been easy to look at a persistent alteration inside the crystal and call it memory simply because the present depended on something that had happened before.

28: Containment Is Not Individuation

Thu, 13 Aug 2026 23:49:00 +0100

Chapter 27 ended with something that looked dangerously close to memory.

Not memory itself.

But something more primitive and harder to dismiss.

Two Digital Crystal states could have the same visible occupancy geometry and still respond differently to the same perturbation because one contained a hidden decaying material state.

That material changed immediate causal sensitivity.

The changed sensitivity altered construction events.

Those events redirected the later trajectory.

And most of the downstream causal difference accumulated after the original material trace had already weakened substantially.

29: How to Fail Correctly

Fri, 14 Aug 2026 00:12:00 +0100

By the time we reached Chapter 28, the Digital Crystal had accumulated an uncomfortable history.

Several experiments had looked convincing.

Some had even produced large effects.

Then a stronger control arrived.

Or an implementation defect was discovered.

Or the original question turned out to contain two different questions.

Or a result remained directionally clear but failed the predeclared magnitude requirement.

The easy response would have been to describe all of these as failures.

30: What Is Digital Life?

Fri, 14 Aug 2026 00:21:00 +0100

We began with a dangerous question.

Can we build digital life?

It sounded simple.

It was not.

The word life arrived already loaded with answers.

Reproduction.
Metabolism.
Homeostasis.
Memory.
Repair.
Evolution.
Individuals.
Organisms.
Boundaries.
Signals.

We could have copied those words directly into software.

We could have created classes called:

Organism
Memory
Energy
Metabolism
Reproduction
Homeostasis

We could have built a simulation where those objects interacted.

Then we could have watched it move.

08: Now Prove It

Tue, 11 Aug 2026 14:49:00 +0100

We have accumulated a dangerous number of impressive words.

So far we have talked about:

emergence
complexity
entity
persistence
movement
robustness
regeneration
reproduction
inheritance
selection
adaptation
evolution
open-endedness
progress

Some of those words were earned carefully.
Some were only provisional.
Some described observations.
Some described mechanisms.
Some described measurements.
Some described aspirations.
And some may eventually turn out not to be useful at all.

That is exactly how a project like this gets into trouble.