Artificial Physics


My Artificial Physics (AP) experiments are an attempt to create systems which have some resemblance to physical systems, and in which complex structures emerge. Some day, I hope to create true artificial life with open-ended evolution in such an AP system. So far, i've only managed to produce psychodelic pictures. Enjoy!

The inspiration for these experiments comes from systems called Artificial Chemistries. in particular Organic Builder. Such systems can be used to emulate some of life's chemisty and also have shown complex structure can emerge spontaneously from simple rules. However, true artificial life and open-ended evolution do not seem to emerge from these systems. Something is missing. I believe the problem is that most artificial chemistries simulate only local bonding forces and brownian motion to create a well-mixed reactor. I believe that for the emergence of complexity, a more fundamendal system of forces on objects in motion is needed. Such a force system not only provides for a well mixed reactor, but also provides for the emergence of stucture and the selection of structures, so that some structures survive longer than others.

Life needs more than chemistry, it needs a complete universe that is both life-giving (emergence of complex structures) and life-taking (selection of meta-stable structures) and that, importantly, is also somewhat similar to our own physical universe, so that we can recognise this life. Hence, my attempts to create Artificial Physics systems for Artificial Life to emerge.

The AP systems I've developed come in two flavours: grid based and particle based.

Grid based

My grid based AP systems are cellular automata with continuous (floating point) variables in the cells. The systems are similar to a simulation of the diffusion process. Each cell in the grid represents concentrations of serval different substances. The substances will diffuse to neighbouring cells at some rate, depending on the mixture of substances in the neighbourhood. For each substance, it can specified wheter it is attacted to or repelled from itself and other substances.

Note that grid based AP systems can be extended with chemical reactions, where mixtures of substances will change locally, resulting in a Reaction-diffusion system. I have not implemented this, because I want to let chemistry emerge though local structures. Most of the experiments done so far concentrate on the emergence of semi-stable structures, i.e. with sufficient life-time and mobility and not resulting in static or repetitive end-states.

My grid based AP systems also incorporate a notion of inertia, where the movement of a substance will keep going even in the absence of influences, similar to the inertia of objects with mass in physics. However, this is quite difficult to model in a grid based system and the result is not quite satisfactory. Particle based systems are more suitable for this.

The video above is a visualisation of a system with three substances (red, green and blue, which mix into all the colors that a computer can display). The system features local interactions and (a little bit of) interia.

Another extention of my AP systems over regular diffusion systems is non-locality. Not only neighbouring cells can influence the diffusion rate of concentrations, but cells further away can also influence the rate and direction of diffusion. Non-local interactions are expensive to simulate. Therefore a multi-scale approach is used, where scaled versions of the grid are used to compute interactions at different resolutions. This gives interesting results, but also introduces some undesirable artifacts.

The visualisation above shows a system with three substances (red, green and blue). The system features non-local interactions.

Particle based


Last update: 2014-06-21