On Sun, May 31, 2009 at 10:22:20AM -0600, Nicholas Thompson wrote: > > The reason that I raise all of this is that it seems to relate to the > little dust-up that we had vis-a-vis epstein a few months back. What are > models for? What does "verisimilitude" do for a model? Do we put skirts, > trousers, and hats on our turtles or is it better not to? And WHEN is a > robot something more than a turtle with trousers. I assume that people on > this list have firm opinions on this subject. >
In an evolving or learning system, environmental complexity is a big factor. If you want to build a walking robot, then evolving a controller in a simulated physical environment is not good enough, although it makes an excellent starting point. There is the suspicion that simulated environments are not good enough to evolve intelligence (the "embedded AI hypothesis"). Similarly, there are some that think that open-ended evolution is impossible in a simulation setting. However much sympathy I have for these views, extending it to the realm of models is just going too far, so I agree with you that adding trousers to the turtles isn't worth it. A physical instantiation of a model is only going to add additional confounding factors, which are already bad enough. Cheers -- ---------------------------------------------------------------------------- Prof Russell Standish Phone 0425 253119 (mobile) Mathematics UNSW SYDNEY 2052 [email protected] Australia http://www.hpcoders.com.au ---------------------------------------------------------------------------- ============================================================ FRIAM Applied Complexity Group listserv Meets Fridays 9a-11:30 at cafe at St. John's College lectures, archives, unsubscribe, maps at http://www.friam.org
