Gradient is a 2D grid world in which artificial "lifeforms" live. These are simulated agents (virtual robots) which can perceive their surroundings, perform computations using their simulated neural network "brains", and act on these. The agents require energy to live and so they need to eat virtual plants and drink virtual water every so often in order to survive. They can also reproduce. This is what makes them a form of artificial life.

The agents in the gradient world initially start with randomly connected brains. These are recurrent neural networks The agents perceive the world as a series of boolean values indicating the contents of the grid cells around them. The grid world is updated iteratively. At each iteration, the agents perceive their environment, perform computations, and get to pick an action to perform (e.g.: turn left, turn right, move forward, consume food or water, produce offspring).

Gradient implements a kind of genetic algorithm. The world is initially seeded with randomly created agents. If the agents do not eat food and drink water, they run out of energy and die. Inevitably, some of the agents will die of starvation. If some agents manage to eat some food and drink water, they gain energy, which allows them to survive longer. If the number of agents in the world dips below a minimum number, more random (seed) agents will be created to maintain a minimum population size.

If an agent accumulates enough energy, it will be allowed to choose to voluntarily reproduce. This has an associated energy cost, and produced a new agent based on a mutated version of it's parent's neural network. Because of natural selection, the agents eventually develop enough "intelligence" to survive. That is, the agents with brains that allow them to survive more effectively become the most prevalent.

The hope is that eventually, agents that are capable of seeking food and surviving long enough to reproduce will prevail, and the population will maintain itself on its own. The world then no longer needs to be seeded with random agents. Note that there is no upper limit on the population size, it is purely limited by the availability of food in the world.

If you let the simulation run in "fast mode", it will run as fast as your computer will allow. The amount of time required for the agents to exhibit some degree of "intelligent" behavior (e.g.: looking for food and water) can vary. In some instances, the simulation does not seem to converge at all, and may be restarted by refreshing the webpage. I recommend you let it run for over 10 minutes in fast mode and see what happens.

I, Maxime Chevalier-Boisvert, programmed Gradient in my spare time a few years ago. It was originally written in C++ and only ran on Linux under GTK. I extended Gradient in 2009 to test some ideas for a neuroscience class project. In May 2011, I decided to port this program to JavaScript so that more people can see it and experiment with it. I believe it's also a potentially interesting and demanding JavaScript benchmark.

Yes, all the JavaScript code for Gradient is licensed under the GPLv3. You are free to download the code, modify it and experiment with it, so long as you respect the terms of the GPL license.

Gradient has been tested on the latest Google Chrome and Firefox. It requires support for the HTML5 canvas element to run, and is rather demanding in terms of JavaScript performance. At the time of this writing, the latest Internet Explorer does not support the canvas element and so will not be able to run Gradient. We recommend Google Chrome for best performance.

Most definitely. I recommend doing a Google search for artificial life. One of the most interesting examples I have found is Polyworld, which is actually quite similar to Gradient.