Social Embeddedness and Agent Development - Bruce Edmonds
"... If a system contains many agents (many is often interpreted as greater than 10), then the dynamics can become too complex to manage effectively.Now these are sensible warnings for the software engineer, but they are not necessarily relevant for social simulation*1, since the unforeseen behaviour that the engineer is try to prevent is what the social simulator is interested in. For the social simulator, the issue of how society can impact upon individual behaviour is at least as important as how individuals impact on society .
There are several techniques that one can use to try to manage a system in which there are many agents. First, one can place it under central control... Another way... is to severely restrict the way in which agents can interact... one can ensure that there are few channels of communication... [or] by restricting the way in which agents interact. Thus very simple cooperation protocols are preferable over richer ones, with "one-shot" protocols... being both adequate and desirable for many applications."
The point is that it is possible that one can not engineer truly social agents because a critical aspect of their sociality might come from their social embeddedness. In other words you might get a different result if you design agents and later put them together to interact, than if the agents learn significant parts of their `cognitive content' in the context of each others learning. If this is the case then if one wants truly social agents then there may be no choice but to allow such agents to embed themselves into a society by a such a process of co-development - it would be something that we could not do for them.
Generated with CERN WebMaker