Can you name the mutation that changed a world?

A deep old question

Artificial-life researchers have been building evolving digital organisms for decades. Tierra and Avida are important precedents. I built this small executable world because I wanted to inspect its histories and test these distinctions for myself. The note records what happened in my system; it makes no claim to a new principle of evolution.

THE NOTE

At tick 993, the two histories are identical.

At tick 994, organism 2231 is born. It inherits its parent's genome with two changes: a lower threshold for reproduction and a different program deciding when reproduction should happen.

threshold 198 → 192

program 37 → 33

In one history, I leave that birth as it happened. In the other, I undo only the program change. The lower threshold stays.

Then I run the world forward.

If a world can be replayed, evolution can become an experiment.

Cell80 is a small artificial ecology. Organisms eat, move, reproduce and die. They inherit small executable programs and numerical settings. Mutation can alter either, and the consequences play out among other organisms competing for resources.

Watching such a world raises a familiar temptation. Something spreads, so we assume it helped. A population changes, so we tell a story about why.

I wanted to be able to test that story at the point where it began.

The same world, twice

Before I could undo a mutation, I needed to know that the world would otherwise repeat itself.

The first experiment tested exactly that. Repeating the same starting conditions produced identical histories. The parallel GPU implementation also agreed with the slower CPU reference at every tested tick.

Randomness is part of the world, but each draw is tied to its seed, tick, organism and purpose. Running organisms in a different dispatch order does not give them different luck.

This makes a controlled replay possible. Keep the setup and rules fixed. Change one inherited field at one birth. Check that the histories agree before the intervention, then measure what follows.

The change that became common

The lineage experiment found a sustained shift in which reproduction program was most common.

At tick 1080, program 33 had become the plurality choice. It accounted for 35.3% of the population, reaching 41.6% during the recorded sustain window. The detector sampled every twenty ticks and required the new leader to persist for a further five samples.

That was enough to identify an event worth explaining. It was still a population containing many competing programs: the new leader did not occupy a majority, let alone replace everything else.

Tracing backward from organisms carrying the program at the event led to a single origin: organism 2231, born to organism 2059 at tick 994.

The ancestry record also exposed something easy to miss. That birth had changed two fields together.

If I undid both, I would lose the ability to say which change mattered.

Undo one thing

The comparison keeps the offspring's reproduction threshold at 192 in both histories. It changes only which reproduction program the offspring inherits.

Every tick before the birth remains identical.

With the program change undone, the detected plurality shift no longer occurs.

That gives a precise answer: this inherited program change mattered to this event in this world. The claim rests on an intervention as well as an ancestry trace.

It does not establish that the program is better everywhere, that every population shift has a single cause, or that a different lineage could never reach the same result. The report searched for a suitable event and demonstrated one end to end.

But I can point to the birth, show the complete change, and show what happens when one part is removed.

A test that can disappoint you

The replay machinery soon became useful for a less satisfying result.

Another experiment introduced compositions of existing programs as movement genes. Some appeared in the population. Their presence made it tempting to describe the ecology as exploiting new code.

I tested fifteen births where an organism acquired a composed movement gene. Each replay gave that same organism its parent's movement gene instead, then compared its direct offspring count.

None of the fifteen compositions increased that count. Twelve tied. Three reduced it.

The later, broader experiment would find useful compositions under a different pool and sampling procedure. But this early sample did not support the claim I had wanted to make.

That is part of the value of the replayable world. The instrument can challenge the interpretation suggested by the animation.

A living-looking system gives us something to watch. A controlled intervention gives us something to ask.