Researchers at the Santa Fe Institute have developed new machine learning methods to study conflict. Their work appears in PLOS Computational Biology on May 13. Quantitative¹ studies（定量研究） of behavior traditionally rely on game theory to investigate the logic² of conflict. Game theory seeks to identify normative（规范的，标准的） strategies that maximize payoffs（支付，报酬） for individuals in the face of uncertainty³.

Although game theory has been very useful for determining which of a predefined（预先定义的） set of strategies – for example, "tit for tat（以牙还牙，针锋相对） " – will be stable given certain assumptions, its has not proven to be very useful for determining what the natural strategy set is, or which strategies individuals are using out of equilibrium⁴（均衡，平静） . Game theoretic models are also not practical for studying strategies when interactions involve multiple players interacting simultaneously⁵. This is the case in many complex animal and human systems.

Santa Fe Institute Omidyar Postdoctoral Fellow Simon DeDeo and Institute faculty⁶ members Jessica Flack and David Krakauer developed the new method, which they call Inductive Game Theory, and applied⁷ it to a time series of fights gathered from detailed⁸ observations of an animal society model system.

"With these approaches, we can identify those strategies likely to generate periods of intense conflict," DeDeo says.

"Fights are not explained by 'rogue⁹（凶猛的） actors,' or single aggressive individuals, but by complex interactions among groups of three or higher, and the decision to fight is very much dependent on memory for what happened in previous conflicts," says Krakauer.

"These results suggest that individual agency has been over-emphasized in social evolution," says Flack. "We need to re-examine the idea that a single individual or nation can cause turbulent（骚乱的，吵闹的） periods in history and consider the possibility that what predicts long periods of conflict is how we respond to the actions of our friends and enemies in their conflicts."

"This new empirically-grounded approach to conflict is a crucial step towards designing better methods for prediction, management and control," she says.