Re: The A.I. Project II
- Posted by Christian.CUVIER at agriculture.gouv.fr Nov 12, 2002
- 475 views
I'd be glad to join in this global exercise, which could show the real deep strength of Euphoria, in spite of the various inconveniences. First of all, I can't imagine doing any serious work on this before Feb. 2003. Just in case this is an issue. The Eu Thing should be able to react to a partly unpredictible environment. By this I mean an environment formed of fixed patterns, some slowly moving patterns and unexpected things once in a while. So, the Eu thing can't be alone. It should have clones or variations it may meet and observe, and also other entities some of which may be real dumb - learning by error is so impressively efficient -. Thus, Yhe Eu Thing should have several senses, sending contradictory input in some rare cases. First, TET accesses raw input - color patches, textures, or whatever more simple we can come up with -. It should have a basic evaluation function - sort of energy level - and an evaluation of this function - initially very low -. The idea is that TET could grow more sophisticated with time. For instance, it will learn than cooperating with like-monded entities is better for its energy, thus inducing it into adding friendship into its evaluation function, and inducing it to learn how to make efficient use of this newly discovered skill. It would soon learn than being friendly without discernment may not be advantageous, etc etc So, all entities should be able to dynamically develop - procedures to extract information from raw input - procedures to evaluate the information - procedures to evaluate the evaluatoon function - procedures to change the ones above We may need a lot of inline ASM here, since this seems to be the easiest way to execute dynamuc code. I think the 3d stuff is completely external and could even be written in another language if needed. Have a nice day! CChris