RI: Medium: Computation, Language, and Games

虽然博弈论已经能够在许多有趣的情况下做出有用的预测，但众所周知，在许多有趣的游戏中，博弈论的预测是完全不正确的。这个项目关注的是一个潜在的解释：玩家的计算能力限制。博弈论假定所有参与者都是理性的。在文献中，“理性”意味着玩家可以计算出自己的信念，并对他们认为其他玩家正在做的事情做出最佳反应。但是这样的计算可能不那么容易执行。该项目仍将假设玩家是理性的，但将尝试明确地建模这样一个事实，即人们是受计算限制的，并受到游戏框架/语言描述方式的影响。然后，它将考虑假设人们理性行事的后果，即尽其所能做到最好，受制于他们的计算限制和他们如何描述他们的世界。更具体地说，该项目将侧重于三个相互关联的主题。1)将计算考虑到人类行为：该项目将研究PI早期关于计算有限玩家的游戏工作中引入的正式模型如何用于解释人类行为，例如理性疏忽，人们似乎忽略了似乎相关的变量，以及在现实世界中将博弈论应用于野生动物和渔业安全，森林保护和毒品拦截等应用中的人类行为异常。2)动态游戏中的计算和语言问题：从直觉上看，当agent进行计算时，他对游戏的理解会提高，他对不确定性的判断会变得更加敏锐，他可能会意识到更多的选择，他的效用函数可能会发生变化。该项目将研究对这一过程建模的方法，包括拖延效应等问题（“我宁愿明天开始节食，今天吃巧克力蛋糕，也不愿今天开始节食”；如果每天都应用这种推理，智能体永远不会开始节食）。3)密码学和博弈论：现代密码学可以被看作是资源有限的代理之间的博弈。该项目将研究这些连接，希望能够设计出更好的加密协议。这项工作的一个具体成果将是实现预期成果的更好机制。例如，了解计算限制对人们行为的影响，可能有助于建立更好的机制来防止偷猎。一个特别关注的领域将是比特币，它可以被视为资源有限的矿工之间的游戏。

While game theory has been able to make useful predictions in many situations of interest, it is well known that, in many games of interest, the predictions of game theory are simply incorrect. This project focuses on one potential explanation: players' computational limitations. Game theory has assumed that all players are rational. In the literature, "rational" is taken to mean, among other things, that players can compute their beliefs and a best response to what they believe other players are doing. But such computations may not be so easy to perform. The project will still assume that players are rational, but will try to model explicitly the fact that people are computationally bounded and are affected by how games are framed/described in terms of language. It will then consider the consequences of assuming that people are playing rationally in the sense of doing the best that they can, subject to their computational limitations and how they describe their world. More specifically, the project will focus on three inter-related topics. 1) Taking computation into account in human behavior: The project will investigate how the formal models introduced in the PI's earlier work on games with computationally bounded players can be used to explain human behavior such as rational inattention, where people seem to ignore variables that seem relevant, and anomalies in human behavior in real-world deployments of game theory to applications such as security of wildlife and fisheries, forest protection, and drug interdiction. 2) Computational and language issues in dynamic games: Intuitively, as an agent does computation, his understanding of the game improves, his judgments of uncertainty becomes sharper, he may become aware of more options, and his utility function may change. The project will investigate approaches to modeling this process, including issues such as procrastination effects ("I'd rather start my diet tomorrow and eat the chocolate cake today than start my diet today"; if this reasoning is applied every day, the agent will never start his diet). 3) Cryptography and game theory: Modern cryptography can usefully be viewed as a game between resource-bounded agents. The project will investigate these connections, with the hope of being able to design better cryptographic protocols. One concrete outcome of the work will be better mechanisms for achieving desired outcomes. For example, understanding the effect of computational limitations on people's behavior may enable better mechanisms for preventing poaching. A particular focus area will be Bitcoin, which can be viewed as a game between resource-bounded miners.