One of the problems with peer-to-peer systems is that searching them can be a big problem. Pure p2p systems have scaling problems as the search space widens, centrally managed (like napster) have scaling problems handling the number of queries.
Agent (automaton) based queries can bassically reduce to pure broadcast searches is the agent is allowed to replicate and depth/breadth first searches otherwise. None of these options is efficient enough to handle really huge,huge (like sum total of human knowledge huge) datasets.
So, agents get a credit card. Each agent has a certain amount of credit in it's account. If it has enough credit it can repliate itself. Credit is gained by answering questions from other agents (which takes credit from those agents). Even if the answer is NO. Some places cost more, and if you're percentage of correct answers is high you can charge whatever you like (you can actully always charge whatever you like, but no one will talk to you unless you've got a good average).
All agents have ownership, no agent can not have an owner (and ownerless agents are always rejected).
So, a little gedanken to help verstahen --
agent get's kicked off. Queires the first (highest score) in the its list
of server's who know stuff. Agent makes query of server. Sever checks
to see if agent has enough credit to get question answered. Anwers in the negative cost
less (server can always lie) If server answeres in the negative, itmay provide a list
of other sserver it thinks might know. List providing is very cheap. If server knows answer, it first tells the agent how much the answer will cost. The agent can decide if it's willing to pay
and if it is, it takes the "payment" gives the answer and that's that.
Servers may also gossip. Gossip is in the form of reporitng ratings for other servers. This one is fast, this one is slow or this one cheats or that one isinexpensive but never knows tha answer.
The system does not have a global accounting system set up. The credit is between server, and the agents and servers record their own transactions. This works because a cheating owner begins to be ignored. A cheeting owner that keeps changing his name cannot generate enough credit to be abusive. A web of trust get's built automatically and enforced by transactional costs.
Agent owners can also set costs for their friends (or for anyone really) to be whatever they want. If you build
a web of people you know and you let your servers gossip to their agents you get reasonable redistribution of informationwithin peer groups. You can also do things like have the agents able to access your information (like your schedule or the report you've been working on).
I'll point out here that I am being deliberately ambiguous about exactly where the agents are running. This is something to be figured out later.
There is no reason why people should design bridges. In fact, structual skeletons of any kind should not be designed by
humans. I should define what I mean by designed.
A human architect should describe the shape of the thing. However, the actuall structure should not be laid out by the
the Architect. Chip manufacturing has long used automated processes to lay out the internals of chips and cirucuits to
maximize function. Bridge builders should do the same.
The program should take basic measurements like width, max load and GIS information about the span to cross
(including geological information about the surface, soil and shelf) and the general shape contrainst of the bridge
and it should generate a bridge that meets these requirements for the least costs in labor and materials.
GA + computing time = bridges
This would work for any construction project, really. add in a smattering of constructor bots and skyscrapers could
be assembled in days for little cost. Houses : no problem. In fact, you could even just use transposed cellular automata
to create the design for you (though that would be harder computationally).
This is my programing blog. This will host software type stuff. Questions I have for myself or anyone else reading. Answers I've found. Software I'm trying to publish. You know, the normal stuff.
Think of it like a pre-abandoned SourceForge site.