#system #evolution


Natural selection is a process which retains at each generation only the fittest of system actors.

Natural selection works through several mechanisms:

While it takes its root in evolution and biology, it is generally applicable to any system with resource constraints and mass effects that span over several generations, such as capitalism, governments, agile, and machine-learning

Natural selection is especially interesting to hypothesize on implicit rules of a system obtained through mutual evolution, based on observed characteristics. For example, a range of venomous insects such bees, hornets and wasps are striped [^3] to advertise that they are venomous. Their stripes are positive advantage: a char This came to be through mutual evolution: colorless insects came to be eaten because animals didn't notice their distinctive patterns, while animals came to die because they attacked bees unaware of the danger. At some point the pattern appeared for a strain of insects through variability, allowing to be noticed by some animals as dangerous. This also allowed their predators to mutually evolve and associate stripes = danger, the pattern spreading through many species[5].

Stripes are a positive advantage: a characteristic acquired by the species at some point which allows it to increase its reproductive success.

Reinforcement learning, and adversarial generative networks are both using natural selection functions to select the fittest models over a large number of generations.



{TODO: link with NEAT when written}