Why nature needs a brain, and what that can teach AI

October 2026. An essay for a general audience; a technical version will follow.

The most powerful neural networks in the world have read more books than all of humanity combined, and they still lose to a one-year-old. The child does not know a single word, yet already understands that a dropped toy does not vanish. Are we teaching AI the wrong thing?

Today AI learns from internet text. A child learns differently: through the body, play, mistakes and collisions with the world. What if we gave machines the same — not a library, but a world in which they can grow up?

Intelligence without a brain

We are used to thinking that a mind needs a brain. Nature disagrees.

Slime mould. A single-celled organism without a single neuron finds the shortest path through a maze. Japanese scientists placed oat flakes where the cities around Tokyo sit on a map, and the slime mould connected them with a network strikingly similar to the real railway.

Plants. An acacia that starts being eaten releases a gas, and neighbouring trees make their leaves bitter in advance. There is also talk of an underground “internet” of fungi linking trees in a forest, though scientists still argue about it.

Evolution. With no design at all, simply through trial and selection, it “engineered” the eye, the wing and the brain itself.

Conclusion: problems can be solved without a brain. The brain is just one form of intelligence.

Three levels of life in one world

Imagine a simulation inhabited by:

• The “brainless”, analogues of plants. They do not learn; they simply change from generation to generation.
• Animals with simple brains. They act on instincts honed by evolution.
• Learners. Those who also learn from their own experience during their lifetime.

An important caveat: nature never had a separate level of “children”. The capacity for long learning emerged in animals by itself, and the long childhood of humans and corvids is a result of it. So it is more interesting not to create learners up front, but to see whether they appear on their own.

All levels live in one shared world. They do not communicate directly, but are linked through shared resources: light, water, food, air. We do not talk to trees either, yet we breathe their oxygen and eat their fruit, while they change the climate and the soil.

And throughout Earth’s history, animals never ate all the plants. The balance holds not because someone is clever, but by itself: when food is scarce, there are fewer animals, and plants have time to recover. Plants are not passive either — they evolved thorns, poisons and bitter taste. Order emerges from interaction without any design. But it is fragile: mass extinctions have happened too.

One more condition: the world needs different niches — different climates, landscapes and food sources. Otherwise diversity of species has nowhere to come from.

How to build it

• Plants change only across generations, a bit like fashion: successful variants spread, unsuccessful ones disappear.
• Animals get instincts that become fixed in those who survive.
• Learners add experience and curiosity on top: novelty is “interesting” to them, so they explore the world even with no goal in front of them.

There is no need to start with a realistic 3D world; a simple 2D environment will do. Modern GPUs can run millions of creatures in it at once.

What has been tried

• Karl Sims, 1994. Virtual creatures made of blocks “invented” swimming, crawling and jumping on their own.
• Avida. Digital organisms made of code, used by scientists to study how complex functions arise. It was the subject of a paper in Nature.
• OpenAI’s hide-and-seek, 2019. Agents learned to build shelters, and the seekers figured out how to climb boxes and even exploit bugs in the game’s physics.
• DeepMind’s XLand. Agents trained in thousands of generated games and became able to handle tasks they had never seen.

The idea is not crazy: individual pieces already work. But nobody has yet assembled the whole ecosystem.

The main question: when does a brain pay off

A brain is very expensive. Ours consumes about 20% of the body’s energy. So it appears only where it pays for itself.

This probably happens under three conditions:

1. The world changes faster than evolution can keep up, and instincts are no longer enough.
2. There are other intelligent beings around. According to the “social brain” hypothesis, intelligence grew primarily to understand one’s own kind.
3. Life is long enough for experience to pay off.

In a stable environment it is better to stay “brainless”, and plants prove it.

Hence the experiment: turn the world’s volatility like a set of knobs and watch under which conditions brains begin to appear.

God level

When did religion arise among humans, and why? One scientific hypothesis, psychologist Ara Norenzayan’s “big gods” hypothesis, links it to the growth of societies. When groups became so large that people no longer knew each other personally, belief in an all-seeing watcher helped people follow rules and trust strangers. It works as a kind of system of governance and cooperation.

Will something similar emerge among agents in a simulation once their populations become large?

And a twist: in such a simulation we ourselves become gods for its inhabitants. We set the laws of the world, send droughts and decide when it all ends.

Would our world happen again?

A simulation can do what nature cannot: rewind history and run it again. Palaeontologist Stephen Jay Gould called this “replaying the tape of life”. He believed that on every new run evolution would take a completely different path, and humans most likely would not appear at all.

There is an opposite view. Eyes, wings and streamlined bodies arose independently in different species many times. This is called convergent evolution, and it hints that some solutions are almost inevitable.

A simulation lets us test who is right. Run thousands of worlds with the same laws but different chance events, and see how often brains, language, cooperation or something like religion appear. If intelligence emerges in almost every world, it is a law. If only in rare ones, we got very lucky.

And the most intriguing question: can we find conditions that produce a world like ours? If so, it would tell us a lot about how we ourselves came to be.

Problems

1. Compute. Nature needed billions of years and an entire planet; we have only a data centre.
2. Poverty of simulation. A computer world is always simpler than the real one, and a mind grown there may turn out helpless in reality.
3. Cheating. Agents look for loopholes instead of learning “properly”.
4. Alien intelligence. A grown mind may be so different from ours that we will not even understand it.
5. Ethics. If agents become complex enough, do we have the right to subject them to famine and extinction?
6. Collapse of diversity. In simulations, one strategy usually wins. In nature, species diverge thanks to niches, isolation, sexual selection and the endless arms race between predator and prey. Nobody has yet built a simulation that keeps producing novelty indefinitely. This is known as the problem of open-ended evolution.

Conclusion

We are used to thinking of intelligence as a property of the brain. But nature suggests otherwise: intelligence arises as a response to the environment, and the brain is only one way of responding. A tree, a flock and a child solve the same survival problem in different ways.

Perhaps we are building AI from the wrong end: trying to make a brain straight away, instead of creating a world in which a mind becomes necessary.

If intelligence is born of the world, what kind of world are we creating for our machines?