Agent Applications
All Messages
Where do you see the trade-offs compared to conventional approaches?
Why agents? -- Why NOT?
I believe that the need for "agent-like software programs" is somehow related to how we use IT. For example: participating in a virtual enterprise requires people to continuously "remember" that there running (business) processes far away from the local workplace which are important for them. They also need to participate "somehow" in these processes. The old concept of "sending a (human) delegate" then needs to be expanded ...
Another thing is that human work capacities are very expensive, and, therefore, more and more limited. The theme thus is: substitute humans by machines. Often, this will not result in hiring employees. Instead, the result will be that time of humans can be allocated to higher-level, creative work. This may hold especially when knowledge workers are considered.
How to reconcile individual autonomy and self-interest with collective interest and performance?
>The number of agent applications continues to increase
True, but may be "so-called agent applications" would be better, because the term agent is often not defined by the authors, and thus mostly abused (in the eyes DAI-persons).
>but the development of such systems can be resource-intensive
It is, of course. Even if there would be (efficient) agent-development frameworks, the domain knowledge and domain specific behavior has to be programmed. Just look at expert systems, same thing and in that case there is only one expert.
>development techniques and tool support
Look at the australian BDI-school, old stuff like MACE, new stuff like Aglets or other JAVA-based "tools". .... Too expressive, too old, too object-like. Look at O'Hare's Agent Facory and its abstract data type description, I like it, but never saw a real world MAS realization based on that paradigm.
>What tools and techniques do you use, or would you like to use to build agent >applications? .....asked the professor the engineer. Answer to the first point: Hacking. Answer to the second point: Would you please repeat the question? Sorry for the sarcasm. When we started to implement our first MAS application we used Klaus Fischers MAGSY. Today I would say it was a forrunner of CORBA. In general it seems to me that currently most MAS implementations are based on some distributed objects platform and hacking. .....Oh I forgot agent programming languages like APRIL, OZ and MAI2L or even the simple agent0 framework would serve as good tools. However, my students did not like them very much.
>Do agents demand different support or can traditional
>methods of application development be used?
I think the first is true, because agent-oriented thinking and traditional methods is at least as different as JAVA and FORTRAN. The reason is the autonomy of agents coupled with their ability to communicate on a high level. OOP catched only the top of the iceberg when asking for a natural approach to program development.
....OOPS, I had to stop my reply. May be it was helpful, if not you may kill me at the conference then.
In summary: What are the correlations between application problems and agent (MAS) attributes?