Rigidity and Flexibility of Social Systems-Further Research

政府和企业等等级制度如何应对变化？此类系统是否存在对其环境反应迟钝的趋势？决定此类系统刚性的因素有哪些？为了解决这些问题，该项目将一个社会系统表示为一组有序的级别。每个级别都有一群代理，并且与代理相关的是行为规则。在每个时期，智能体观察随机环境的实现并根据其规则选择行动。两种动态决定了系统中代理部署的规则。首先，选择（或晋升）动态决定哪些代理晋升到层次结构中的更高级别，从而影响高级别规则的总体组合。其次，社会学习动态决定了主体进入系统时采用的规则。新特工试图通过尝试从历史记录中推断出他们使用的规则来模仿高级（即成功的）特工。这两种动态描述了一个在行为规则空间上定义的反馈系统：高级智能体的历史决定了新智能体采用的规则，然后选择对这个群体进行操作，生成一组新的高级智能体，这些高级智能体会影响下一代新智能体。在 NSF 拨款 9708910 下进行的分析描述了一组动态稳定的结果，并探讨了它们对内部结构（例如层次结构中的级别数）和外部结构（例如环境的波动性）特征的依赖性。 最初，进化发生的行为规则空间包括对环境反应不同的规则。在当前项目中，空间被重新定义，以允许规则响应不同的内容。一条规则 - 个人学习规则 - 让代理根据他们自己的（嘈杂的）环境信号选择最好的行动。第二条规则——社会学习规则——让代理人选择大多数代理人在前一时期选择的行动。当大多数代理人使用社会学习规则时，组织惰性可能会出现。对先前工作的第二个修改是内生化“表现良好”的含义。在公司的背景下，代理人的晋升不是基于某些固定的、外生的绩效概念（例如利润），而是基于对更高级别代理人的主观评价。如果不同的智能体根据其特征以不同的方式解释绩效，那么绩效本身就是内生的，因为高级智能体的特征是过去选择的产物。这项研究体现了选择过程的这一特征，为等级社会内的动态提供了更丰富的描述。

How do hierarchical systems, such as governments and corporations, respond to change? Is there a tendency for such systems to become unresponsive to their envi-ronment? What are the factors determining the rigidity of such systems? To address these questions, this project represents a social system as a set of ordered levels. At each level is a population of agents and associated with an agent is a behavioral rule. In each period, agents observe the realization of a stochastic environment and choose an ac-tion according to their rule. Two dynamics determine the rules deployed by agents in the system. First, a selection (or promotion) dynamic determines which agents advance to higher levels in the hierarchy and thus affects the population mix of rules at high levels. Second, a social learning dynamic determines the rules that agents adopt when they enter the system. New agents seek to imitate high-ranking (that is, successful) agents by trying to infer the rule they used from their historical record. These two dynamics describe a feedback system defined on the space of behavioral rules: the history of high-ranking agents determines the rules adopted by new agents, selection then operates on this cohort to generate a new set of high-ranking agents who influences the next generation of new agents. Analysis conducted under NSF grant 9708910 characterized the set of dynamically stable outcomes and explored their dependence on features of the internal structure (for example, the number of levels in the hierarchy) and the external structure (for example, the volatility of the environment). Initially, the space of behavioral rules over which evolution took place included rules that differed in their responsiveness to the environment. In the current project, the space is redefined to allow rules to differ in what they are responsive to. One rule - the individual learning rule - has an agent choose that action which is best based upon their own (noisy) signal of the environment. A second rule - the social learning rule - has an agent choose that action that most agents chose in the previous period. Organizational inertia might then emerge when most agents use a social learning rule.A second modification of previous work is to endogenize what it means to "perform well." In the context of a corporation, agents do not advance based on some fixed and exogenous notion of performance (such as profit) but rather on the subjective evaluation of higher-ranking agents. If, depending on their traits, different agents interpret performance in different ways then performance is itself endogenous as the traits of higher-level agents are the product of past selection. This research embodies this feature of the selection process to provide a richer description of dynamics within a hierarchical society.