Agents, Organisations, and Social Sciences

All Messages

Organisation forms
Joerg Mueller -- Tuesday, 10 February 1998, at 9:30 a.m.

Which models of organisation theory are most relevant (and useful) for the design of multiagent systems? Social models were hardly developed with the idea of turning them into computational models in mind? Where do you see problems, ways, and trade-offs? How does your work fit into this?

Re: Organisation forms
Stefan Kirn -- Sunday, 8 March 1998, at 10:00 p.m.

There are a lot of very different organization theories available today. Each of them addresses a particular pespective, is intended to describe/address/solve a particular problem, etc.

I believe that MAS theory can learn a lot of org. theory, and vice versa. I thus are considering two areas of great interest to me: (1) how can we intgrate MAS / MAS theory into the body of traditional organizational theory (in order to establish HCCW-teams), and (2) to evaluate how exactly traditional org theory can help solving different problems when designing (and using) MAS.

In my work presented in the paper, the control of individual behavior in a MAS is addressed. Out of the numerous different org. theories I thus selected an instrumental one. This approach defines organization as the set of rules and constraints /instruments) by which individual agents can be controlled (maybe, towards a global goal). The perspective I have chosen is a life cycle perspective. That means the life cycle of "formally organized social systems". I personally feel that this approach is quite helpful.

Possible problems: depend on the approach you choose. In any case, organizational models are more or less pre-formalized. For example, the instrumental approach requires rational agents in order to develop expectations about how they will behave in a particular situation.

For me, a very important point is to open (break up) the barriers traditional org theory has established against th idea of "artificial org members". If this fails it will be very difficult (possibly impossible) to find appropriate agent applications in real life environments.

Tradeoff: to learn about what generations of org. researchers and practioners all around the world have worked about how more or less autonomous individuals can be fruitfully work together ...

---

explicit representations vs. emergence
Cristiano Castelfranchi -- Thursday, 19 March 1998, at 4:41 p.m.

Are cognitive agents endowed with explicit mental representations (beliefs, desires, intentions, expectations, etc.) useful or necessary for simulating social and organisational phenomena, and why? How is this reconciliable with an emergentist view of macro-social cooperation?

Re: explicit representations vs. emergence
Kerstin Dautenhahn -- Monday, 23 March 1998, at 7:53 p.m.

In my view the distinction between usefulness and necessity is central to the debate of (mental) representations and models of social phenomena. Often discussions seem to argue for one or the other "right" approach, but I don't think that there is one "solution" to the problem of producing "social behaviour", it simply depends on the purpose of the particular research. In my paper I take a particular "emergence-oriented" attitude and discuss that in order to produce a particular (social) behaviour of agents it might not be necessary to take a "model-based" approach. However, explicit representations and models can be highly useful, e.g. when they help to EXPLAIN a particular phenomena, or when an agent product can be implemented on the basis of such models. Emergence is a central concept in Artificial Life research, and as e.g. Langton pointed out, emergent phenomena cannot be DESIGNED in the same way as people engineer the desired behaviour of a software program. Some people in software enginnering say that software engineering is still a kind of art form, however, there are textbooks and guidelines precise enough to build a text editor with a desired performance. I'm glad that the behaviour of my text editor is (more or less) predictable, and that slight changes of my interaction with it (e.g. speed of typing) does (usually) not result in chaotic behaviour of the program. The same idea can be applied to "social agents". Thus, emergence versus modeling of social (mental) representations is not a matter of one or the other, or finding the right recipe to combine the two. If somebody is interested in how on the basis of local interactions and dynamic agent-environment couplings interesting social phenomena can result, then "emergence" is the focus of interest. Both approaches are in my view justified and serve particular purposes, in particular research fields. (very brief summary of a long story).