This is all my opinion. I'm putting it out here to make clear what assumptions I'm working from. I'm planning on adding some examples later to make it clear what I mean by each point.
Social behavior/algorithms exist because they achieve the following goals for groups of entities:
Social behavior accomplishes this by:
Social groups accomplish these things by providing rules for interaction between agents. Theoretically, agents could re-negotiate all these things as they come up, but practically that would waste a lot of time. A social group pre-negotiates all these things, and provides an enforcement mechanism against people who violate these negotiations.
Note that working out a set of rules for a social group is a non-trivial problem. I will probably start by assuming a set of social group rules, and working out how agents interact with them.
So - what are the properties of social groups? Social groups exist to more effectively coordinate the activities of their members. What types of coordination occur? I think a clear example is the distribution of resources. A common problem encountered by cooperating groups is how to divide the spoils of their cooperation. Social groups use hierarchies to help accomplish this, but that's not the whole story. I'm babbling - I'll try to put down a more detailed outline.
Members of social groups possess and transfer knowledge about the preferences and capabilities of other members. Thus, when one member becomes aware of new information, a new oppourtunity, or a new problem, they will know who in the group will be most capable of dealing with it. (I think actual human behavior is more complicated than this - f'rinstance, intra-group competition occurs, which means that when a member discovers a new oppourtunity, they may choose not to disclose it to another member of the group, so that they can keep all the credit for themselves.) Note that this knowledge about capabilities could be viewed as a hierarchy, over a specific domain.
So, when a group collectively receives a windfall, the windfall needs to be distributed among the group. Open conflict is usually more costly than a peaceful distribution. Knowledge about the capabilities of other members can defuse open conflict (example). But it goes further than that. Payoff functions in the real world are usually not linear: when you're hungry, 1000 hamburgers are not 500 times better than 2 hamburgers. So when a group receives a windfall, it doesn't make sense for it all to go to one member. Social groups can also perform a smoothing function for resource acquisition. Consider this (highly contrived) example: A group of 4 hunters - each hunter stands about a 1/3 chance of killing an elephant in a month - a hunter will starve to death if he does not eat for a month - after 1 month, the meat from a dead elephant will be too rotten to eat. Now, clearly, one hunter on his own will starve to death in fairly short order. However, a group of 4 of them should do fairly well, catching about 4/3 of an elephant a month. So logically, they should form a social group to share the proceeds of their hunts - so that when they get unlucky, they don't starve. However, the disadvantage of this system is parasitism. Suppose one hunter reasons that if he doesn't hunt, the other 3 will still catch an average of 1 elephant a month, so he can afford to go do something he prefers. If all hunters reason like this, they all starve. So there needs to be some mechanism to keep them working. More on this later.
To go back to our hunter example. Suppose that when they go hunting, they have to pick a direction to go to look for elephants, (N,S,E,W). If more than 1 hunter goes in a particular direction, then the entire group of hunters that went that direction only has a 1/3 chance of catching an elephant (i.e. more people doesn't help, although it doesn't hurt). To maximize payout for the group, one hunter should go in each direction. A social group encourages this, by allowing for communication between them. Again we see a use for a hierarchy though. Consider what happens if some directions are better than others. Suppose that if you go north you only have a 1/5 chance of getting an elephant, but if you go south, you have a 1/2 chance of getting an elephant. It's desirable to go south, but if everyone goes south, they will all eventually starve. I think that hierarchies are also useful in this circumstance - indeed, you can treat the oppourtunity to go hunting in the south as a resource to be divided.
I believe I covered this in an earlier point. If people exchange knowledge about their capabilities in conflict, conflict can be avoided.
Ok, so I've said a lot about how neat heirarchies are and social groups are for helping people. The real question is in deciding how neat. So - consider a computer agent given an oppourtunity to join a social group. Suppose he's given complete information about the members of the group, and the rules they use to interact (this is a huge assumption - in reality, joining social groups is much more complicated - another topic to get into later - for now I'll just say that you have to deal with the possibility of a false offer without exposing any of your vulnerabilities (which will become exposed if you join the group).) So given this information, how do you decide if it's worth your while to join the group. Conversely, how does the group decide if you would make a useful addition? I think the SharedPlans might have some bearing here. Maybe a generalization to allow for non-specific plans, or ongoing interactions?
The views and opinions expressed in this page are strictly those of the page author.
The contents of this page have not been reviewed or approved by the University of Minnesota.