Computer agents: Testing theories of social interaction

The complex behaviour of primates can be understood using artificially-intelligent computer 'agents' that mimic their actions, shows new research published in a special edition of Philosophical Transactions of the Royal Society B and presented at the BA Festival of Science in York.

Scientists using agents programmed with simple instructions to work out why some primate groups are 'despotic' whilst others are 'egalitarian' - overturning previous theories developed by primatologists.

They have also found support for an existing theory of how dominant macaques make it to the safer positions at the middle of their troop without seeming to be pre-occupied with getting there.

Using agents programmed with two rules -- stay in a group for safety and pester subordinates until they move away -- scientists found that their more dominant agents would make their way to the centre of the group.

This is one of the few times I've seen agents being used to test theories. Oftentimes, agents are criticized for not being predictive. That is, you can run agent models, see interesting and thought-provoking patterns, but then ask: How does that help me to set a course of action?

While that question is unanswered, for now, the idea of mimicking social behavior and comparing the agent model to actual creatures, is quite fascinating and a marvelous application of this science.