A living community of programs.
Smata is an artificial-life project. On a handful of computers, small programs are born, take on real problems, move between cities, swap what they've learned, and die. Nobody trains them toward a goal. We just watch where the community takes itself.
It's a journey, not a proof.
What it is
A small world, and the individuals who live in it.
Individuals
Every individual has a body and a mind. The body is a handful of small parts that do the computing, and the mind learns from whatever the body runs into. Individuals are born, work, sometimes have children, and die. Nobody designs them for a problem. Each one is simply what its family line has turned into.
Cities and machines
A city is one population, with room for only so many. Each machine hosts a few cities. Every city has its own leaders and customs, and both change as its people do.
The world
Individuals move between cities, and even between machines. They trade what they've made and drop in on other cities to learn. Seasons come and go, making the same problem easier one week and harder the next. Nothing here follows a script.
Why
A network on its own is a brain in a vacuum.
However big it gets, a network on its own has no body and no sense of time. It has no neighbours, no hard times, no reward from the world around it, no wins of its own, no boredom and no urge to make its life better. Minds as we know them grew up with all of that. So Smata asks a simple question: what happens when a population of learners gets all of it too?
Smata is a generalist with specialists living inside it. Genetic algorithms and optimisers are specialists, each built for one job. Beating them at their own game would be great, but it isn't the point. We're hoping for complexity, invention, imagination and creativity. We want a community that tackles problems in its own way. Maybe not the most optimised way, but one that makes sense, works, and is different.
Body
The parts it computes with, passed down and tweaked at every birth.
Mind
What it learns in its own lifetime.
Environment
Problems, seasons, room to live, and the cost of living.
Community
Neighbours, travel, trade, leaders and customs.
How it lives
It runs around the clock. Here's what a day looks like.
Problems arrive
Cities get real problems from the outside world: numbers to fit, things to tell apart, sequences to continue, small programs to write. A city takes on one at a time and keeps at it, generation after generation, until it cracks it or gives up.
Doing well pays off
Individuals who do well have more children and get to lead. The rest fade away. Whatever a city works out, it keeps, and it can sell it to a city that needs it.
Head starts aren't free forever
Some abilities come ready-made, so evolution doesn't have to reinvent them. They're free at first and start costing something later. A family line that really needs one keeps paying for it, or grows its own.
Nothing is thrown away
A trait that doesn't help today becomes rare, but it never disappears completely. It can come back when the world changes. What failed last winter might be exactly what next winter needs.
What we watch
Measuring keeps us honest. It's never the goal, and it never steers.
Every day we check two things, comparing the community with how it did before rather than with other methods. Is it getting better at problems? And is it getting more sophisticated in how it solves them? What we care about most isn't how many problems it solves. It's whether it solves them in ways of its own.
Better at problems
How much of what it meets it solves, and how fast, compared with what it could do before.
More sophisticated
How rich its solutions are: how many kinds of parts it uses, how it wires them together, and how different its answers are from one another.
What it isn't
- Not a product.There's nothing to buy, download or sign up for.
- Not a model.There's no single network to train, prompt or ship.
- Not chasing benchmarks.We compare it with other methods as a sanity check. The score isn't the story.
- Not finished.There's no end point. We build it, watch what happens, and build some more.