Geb is an alife simulation that as far as I know passes all of the tests we have tried to come up with in defining open endedness. However, when you actually run the code, the behavioral complexity certainly increases, but the physical bodies of the creatures never changes (and cannot change), and Geb bodies are the only thing present in the world.

Tool development, or at least developing new physical capabilities and new “actions” seems to be a crucial part of what makes evolution open ended. I think the problem with Geb is that the evolution and progress all takes place in the networks and network production rules, which are systems outside the physical world. They are external systems that take in data from the world and output actions for the agents. So while this rich complexity and innovation is occurring, it’s not integrated with the agents actions and physical bodies.

This leads to a simple question: is there an alife system that passes all the same tests Geb does, but is “fully embedded” in its world? In the sense that any mechanism agents use to make actions must be part of the physical world of those agents, and subject to the same rules as the bodies of the agents themselves.

What I’m saying here is loose, you could come up with plenty of weird edge cases that meet what I ask exactly without meeting my intent. And perhaps being fully embodied isn’t necessary, just being more embodied would be enough. What my intent is is to ask if we have any systems that pass the open endedness tests Geb have passed, but have the innovation occurring in a way that leads to emergent growth of “actions” and emergent growth of “bodies” because to evolve and do better those aspects must be improved as well.

  • $\begingroup$ Geb software available from channon.net/alastair/#Software - I'm not sure how/where to link that $\endgroup$ Jun 25 '20 at 9:25
  • $\begingroup$ It's not an alife system, but are you aware of the work of Karl Sims? karlsims.com/evolved-virtual-creatures.html There are some examples of creatures set to compete directly with each other. The control system complexity is not embodied or part of the fitness criteria, but available "actions" do appear spontaneously $\endgroup$ Jun 25 '20 at 9:45
  • $\begingroup$ @NeilSlater yes! You are right that the action set seems to be pretty emergent there, especially when they compete for the object. It’s an example where just moving more (but not all) innovation to the physical world still does quite a bit. Has anyone ran the open endedness tests on some variant of sims work? $\endgroup$
    – Phylliida
    Jun 25 '20 at 19:35

Your Answer

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

Browse other questions tagged or ask your own question.