Re: Thesis project... long winded.

[John Eikenberry [MSAI]]

On Thu, 27 Feb 1997, Nelson Minar wrote:

> And IBM aglets (Java) http://www.trl.ibm.co.jp/aglets/

This is an interesting site. Thanks for pointing it out. BTW, do you know
if it works with the linux jdk. I didn't think it looked like it did,
since they (blackdown) haven't release 1.1 or the RMI for linux. Oh, I've
read some of the telescript stuff. You're right, good ideas, bad
marketing. 

> The metaphor for lots of software agents I came to the Media Lab with
> was artificial life. I tell people that I used to work on simulating
> alife and now I want to create real artificial life on the Internet.
> There's a lot of value in ecological metaphor, especially when you
> think about hostile agent interactions (predators, viruses, etc).

I got into both alife and 'mobile agents' around the same time. So I
decided to combine the two (seems a natual combo). I like the ecological
metaphor as well. I like to be specific though (at least when it comes to
implementing something), so I choose specific critters to emulate (ie. the
wolf and the bee). 

>   http://www.agorics.com/agorpapers.html

Thanks, I'll check these out.

> As for simulating systems of thousands of mobile agents, well, I have
> to wonder if that's a bit premature. 

I think thousands of agents is a bit premature as well. This thought just 
got me interested. I think I'll start small and see how many make sense 
as I go along (hopefully at least a hundred or so).

> What do you want the agents to do? 

I have a few specific behaviours in mind. Things I think would be useful 
things for agents to do.

1. A courier (agentmail) - for this behaviour the agent will simply find 
the person (most likely using an address gained from a database) and 
deliver a file of some sort to them. The agent then will wait for a reply 
or confirmation as required.

2. Information foraging (to use Holly's term) - this behaviour will 
entail hunting down a piece of info. This info would be something in an 
archive or at an ftp site.

3. Information retreival - similar to the above, except this would entail 
the agent knowing where to find the info (such as a database on some 
system) and going and getting it (and dealing with database).

4. Information negotiator - this would be when the agent goes to a site 
where it will have to negotiate with another agent to attain some info. 
The negotiations would be along the lines of security checking and 
instructions as to what the agent is after.

5. Data watcher - the other half of the above combo. The agent that 
babysits some info and gives it (possibly after getting it froma 
database) to those agents whom deserve it (pass the security check).

6. scheduler - the agent goes and meets with another agent to arrange a 
time when their corresponding clients/owners can meet.

Now I'm obviously not going to implement all 6 right off. I plan on 
starting with the easier seeming ones and progressively adding the others 
(or perhaps different ones if I get the ideas). The first will be the 
courier, then the forager, then probably the retriever, then whatever 
makes sense (I haven't really thought that far ahead in the implentation 
schedule).

> What is the content of their interactions? 

I'm not exactely sure what you have in mind here. If you mean 
communication, that is something that I plan on introducing after I get 
the basics working. I plan on starting things very simple in this matter, 
but I believe this to be very important (if not necessary). I have a bit 
more reading to do before I decide on anything. Any suggestons? 

> Those are my first questions. Once you answer those, then actually 
> building a system (either a Swarm simulation or a deployed thing) is 
> the next step.

This is about where I am (or so I believe). 

John Eikenberry
address@hidden
______________________________________________________________
"A society that will trade a little liberty for a little order
 will deserve neither and lose both."
                                         --B. Franklin