The chase cam and neural networks demystified

The "chase-cam" view in Genesaver might look like a random array of blinking lights and fluctuating gauges, but everything on the screen does mean something - you could figure it out by careful observation, but I'll give a quick rundown to speed things along.

At the top center we have a view of the creature that we're following. We see as much of the world as the creature might be able to see; the radius we're looking at exactly corresponds to its maximum visual range. Note that the creature can't actually see things behind it, but for our convenience we've got a 360-degree view.

To either side of the limited world view we have energy gauges. When the one on the right maxes out, the creature splits. When the one on the left disappears, the creature dies. The shifting colors are just for prettiness value.

The bottom is the real reason this view exists - so you can see the creature's neural network in action. Each flashing dot is a neuron; blue indicates a positive value, red indicates a negative value. The lines connecting neurons are "synapses". Cyan synapses are positively weighted, yellows are negatively weighted. The neurons along the top are inputs, and the neurons along the bottom are outputs.

Above is a slightly doctored screenshot of a simple green carnivore chasing a potential blue meal. I'll be walking you through the Rube Goldberg contraption that is this creature's brain.

The left blue input A is on because the creature's right eye sees something blue very close to it. You might notice that both of the inputs for seeing red, to the left, are on, but not quite as brightly; this is because of that red creature up ahead. Input A is connected to neuron C down below by a positive weight, so when A lights up, so does C. C is in turn connected to D by a negative weight - it's tough to see in a compressed picture, so just trust me, that synapse is very yellow. D then connects to output F by another negative weight, yielding a postive output. F happens to be the output that governs turning, and a positive value results in turning left.

So when this creature sees a meal on its left, it turns left. Coincidence? No, millions of generations of evolution. As it turns out, the connections to the right eye result in turning right if the blue thing is on this side - you can convince yourself of this pretty easily by tracing it out.

What tells the creature to go forward? Even simpler. Input B is the one that represents the creature's current energy level, and it feeds into output E, which is the output that governs forward movement, through a series of positive weights. We can surmise from this that the creature will run quickly when it's got lots of energy, and slow down as it tires. The other input that feeds E is the one that indicates overcrowding behavior; it looks like when green gets overcrowded, this carnivore will tend to slow down so as not to conflict with its brethren quite as often.

Almost all of the intelligence of the creature is contained within those synapse weights, as you've probably noticed. Unlike most real-life creatures, the critters in Genesaver don't modify their synapse weights during their lifetimes, so no individual really learns anything. However, since the weights are encoded within their genes, the species as a whole is able to learn - more successful weight combinations are passed down to offspring, which sometimes improve on them through random mutation. The end result is intelligent behavior.

Back to the main Genesaver page