I'm trying to understand Artificial Life (e.g. here for a simple background) in Computational Evolution.

I understand that in this set of methods, you set up a dynamic environment (e.g. the ecology of the environment) and then you set a series of rules; e.g.:

  • You need energy to reproduce.
  • You intake energy from food sources.
  • For nourishment, you can eat plants, animals, or steal food.
  • You must stay alive until you reproduce.
  • Every action consumes energy.
  • When you have no energy left, you die.

I think I need a set of rules that govern the survival of an artificial life. You run the environment and see what persists (there's a set of rules instead of a fitness score), and the individuals that survive are said to be successful.

I can imagine a scenario where a successful organism in this environment consumes a lot of food, reproduces, but possibly runs out of energy and dies. I'm wondering if there's ever a situation where an organism does very little (or nothing), and still be successful? I'm not sure if this question makes sense, please let me know if clarification is needed. Given the specified environment, I want to know if the most active organism will always be the most successful. The most active organism would be the one that obtains the most food/energy/reproduces the most. Or is it possible to not be the most active organism and still be successful?

  • 2
    $\begingroup$ beautiful thing about a complex system which you are simulating is its unpredictable ... given sufficient opportunity ( time and variability ) you will see a wide range of surprises ... essence of evolution is (1) genetic inheritance (2) diversity of offspring (3) individual organism exposure to the vagaries of the environment which determines survival or death ... its very good to have real time plots showing several different dynamics as well as instrumentation to record when predicted events happen ... have fun $\endgroup$ – Scott Stensland Dec 24 '20 at 14:09

It is entirely possible!

You see, the agents will perform whatever actions are available to them, and if the evolutionary algorithm is setup correctly, whatever set of actions provides them with a bigger survival rate will be the one that gets explored and reproduced the most.

Here is a very interesting list of "Specification Gaming" in AI, where the agents happened to "game" the rules to reach their goals (metric optimization) without actually doing what the creators intended: https://docs.google.com/spreadsheets/u/1/d/e/2PACX-1vRPiprOaC3HsCf5Tuum8bRfzYUiKLRqJmbOoC-32JorNdfyTiRRsR7Ea5eWtvsWzuxo8bjOxCG84dAg/pubhtml (second link)


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.