A small world, running live

Agency from gradients

Stars burn energy fast, yet most of it spills into empty space. This world asks what happens when creatures are never told what to want, only that those who run out of energy die and those with spare energy have children.

Hover over the world to inspect a square.
Grass. Everyday sunshine energy. Brighter means more.
Creatures. Redder means hungrier. They start with random instincts.
Small deposits. Leftover energy that stays shut until a creature spends energy to open it.
Big deposits. Rich, but only several creatures working together can open them fully.
Ring. A creature is spending energy to open a deposit right now.
Glow. A faint scent of deposits that creatures can sense from afar.
Day 0
Starting a new world.

    Sunshine used or wasted

    Wasted means energy that grew or leaked away without anyone using it. Leftovers are the point of this world.

    Where creatures get their energy

    Creatures alive

    Energy used, %

    From deposits, %

    Instincts, averaged over all living creatures

    At the start these hover around zero because every creature is random. Bars to the right mean creatures are drawn toward it.

    Try something

    Rules of this world

    Changing a rule starts a new world.

    Who has children?

    This is the only pressure in the world.

    What is going on

    Sunshine falls on the ground as grass. Whatever is not eaten leaks away. Creatures eat, spend energy to move, and die when they run out. A creature with spare energy has a child that copies its instincts with small random changes. Over many generations, instincts that help creatures survive become common.

    The leftovers

    Deposits hold energy that grass cannot provide. They need effort to open, and no creature is told to bother. The question is whether creatures that happen to bother end up with more children, and whether the habit spreads.

    Why it matters for AI

    Most AI systems are given a goal. Living things were not. The idea tested here is that agency may need the same starting conditions as life: scarcity, inheritance, real consequences for failing, and a ladder of harder energy sources to climb.

    What this is not. It is a toy world, tuned until selection could act in it, and it does not show how life began or that AI will become agentic. The real experiments, with 30 random worlds per condition, preregistered hypotheses and every design change listed, are in the project repository. This page runs a JavaScript port of the same rules, so runs look like the Python ones but are not identical.