Cart
[Home] [By Thread] [By Date] [Recent Entries]
Roger, IMO, you should go away from the god-like situation to perscribe a set of mutation options, in the first place. Your pickers may evolve in quite unexpected directions if left to themselves - you should introduce a mecahnism for them to do so and wait for several hundred iterations to see the results if you want the maximum gain in knowledge - as they did with E Coli. This is the first stage of freeing your mind from algorithmic thinking patterns, and to develop bottom up, you would need such freedom. I think the vital concept in such case is introducing some context processing mechanism - like the one people adopt when learning languages. If you are not very good in a foreign lanuage, an amazing parallel learning mechanism may occur - when reading about the subject, you grasp new words, and are able to use them to read about the subject in more detail. To get the analogy, your pickers should have "instincts" on what's good for their purposes - in this case survival (intrinsic rules that may mutate as well), and the environment should be left to change too, unpredictably and free of any limitations as long as it obeys some few rules. This way, the pickers can learn by evaluating the environment (e.g. look at the soil and decide in which direction to go basing on their vision results plus the expectation of a better harvest, say, in the east), and develop learning communities/collaborate if this suits them. To be more concrete on how to make your pickers more intelligent, you probably ought to read classical AI, especially neural networks, cognitive science and culturology - I found delight in AI myself some years ago, and can give you a reference to a coherent, easy-to-comprehend cognitive simulation set-up from my country (Bulgaria) called DUAL which I don't follow closely anymore but it may suit you as a starting point for community building architectures [1]. In terms of the DUAL model, you still struggle with the micro-level ;-) while its architecture also allows for meso level (the equivalent of a tribe as far as I can see), and macro level (big formations of agents - whole societies). The DUAL agents are richer than you pickers in that they may belong to different classes (e.g. cups and saucers, or liquid-holders consisting of teapots, cups, glasses and bottles) and have, consequently, more "tags" than the ones you introduced so far in your vineyard, namely properties specific to each class of agent and common properties for all agents, "slots" to express hierarchies, etc. They process knowledge not only single but also in a network, and the knowledge of each agent may be processed by the network (group/society) if this agent is filled with energy and has therefore the properties of visibility and speed. The patterns of activation and activity levels for the existing agents change with each run - just like your stylesheet processing. All in all, given the emerging technology, I would be very curious whether you can (and would want to) re-build this architecture in XML as a continuation of the work you started - I still believe it's worth developing. Best regards, Kremena [1] http://www.nbu.bg/cogs/personal/kokinov/dual_i.html
|
