Creatures on a 90×60 grid, each with a brain of 176 numbers. Nobody tells them what is good. The code describes a body, what it can notice and what it can physically do. Behaviour is selected by death alone.
At birth the numbers are random, behaviour makes no sense, and almost everyone dies. Those whose numbers happened to work leave children with nearly the same numbers. After four hundred generations the world holds habits nobody wrote.
| what moves the weights | who sets the goal | |
|---|---|---|
| supervised learning | the error against a right answer | a person, with the right answers |
| reinforcement learning | a reward for the outcome | a person, with the reward formula |
| unsupervised learning | structure in the data | a person, by choosing what counts as structure |
| this stand | random mutation and death | nobody |
Predation appears on its own, but only after hundreds of generations. Mobility is a stronger knob than abundance. At a step of 0.75 cells per tick, 1–5% live a life without taking from anyone. At 0.10 it is 91–100%, even with food cut until the population falls sixfold. Settled creatures meet more often, so taking needs movement for something else: to choose whom to approach.
Good reputation appears and bad does not. Creatures share with those who fed them before, 18–36% of the time against a background of 0–2%. Selection has found no revenge.
It is simpler than a bacterium. The nearest relatives are teaching models of artificial life such as Sugarscape and Tierra. On the same base it can grow towards signals between creatures, choosing a partner, caring for offspring and learning within a life. Language and planning will not come from it.