A scalable model of believable performance agents: Using case-based planning in the control of information access and application.

这个项目使用可信的即兴性能代理领域来调查可伸缩性和健壮性的更大问题，这些问题长期以来一直存在于人工智能系统中。总体而言，人工智能系统要么容易出现脆性，导致其知识库边缘的能力急剧下降，要么在知识库庞大的情况下出现推理的超级爆炸。这些现象反映了一个系统复盖一个以上领域所需的知识广度与确保所产生的知识库能够有效和有效地使用所需的控制之间的紧张关系。在这个项目中，控制和广度问题是分离的，并由两种非常不同的系统处理。控制由基于案例(或基于实例)的计划的变体来处理，它将为信息获取、评估和应用提供任务级的指导。广度是通过使用信息检索社区的词汇技术来提供的，这些技术将被用于挖掘在线资源。基于案例的规划部分将用于创建查询和要求，然后使用这些查询和要求控制检索机制执行的信息访问。目标是构建一个健壮且可扩展的系统，该系统能够在不牺牲推理系统提供的控制能力的情况下提供广泛的覆盖范围。选择可信的性能代理作为应用领域提供了高度可见的约束和评估指标。一方面，代理人必须利用人工智能的控制，以便产生适当的体验，同时，另一方面，也必须在信息相关性、话题性和观点的问题上。这项工作在一定程度上是为了给机器在世界上扮演一个新的角色：机器被用来向世界展示通过在线系统和在线系统掌握的联系在一起的沟通和文化联系。这项工作的更广泛影响将从创建特定的性能代理到更强大和可扩展的人工智能愿景。该项目旨在吸引更多来自传播学和戏剧专业的学生(包括在工程学中代表性不足的女性)，她们希望在机器上工作，作为一种沟通工具，使沟通成为可能，并得到智能系统的支持。

This project uses the realm of believable improvisational performance agents to investigate the larger issues of scalability and robustness, which have long been problematic in AI systems. In general, AI systems have either been prone to a brittleness that results in a dramatic drop off in competence at the edges of their knowledge bases or a hyper explosion of inference in those cases where the knowledge bases are massive. These phenomena reflect a tension between the breadth of knowledge needed by a system to cover more than a single domain and the control needed to ensure that the resulting knowledge base can be used efficiently and effectively. In this project, issues of control and breadth are decoupled and handled by two very different sorts of systems. Control is handled by a variant of case-based (or example-based) planning, which will provide task-level guidance to information acquisition, evaluation, and application. Breadth is provided by the use of lexical technologies from the information retrieval community, which will be used to mine online resources. The case-based planning component will be used to create the queries and requirements that will then be used to control the information access performed by the retrieval mechanisms. The goal is to build a robust and scalable system that is able to provide breadth of coverage without sacrificing the power of control provided by the reasoning system. The choice of believable performance agents as an application domain provides highly visible constraints and evaluation metrics. On one hand, the agents must make use of the control of AI in order to produce the appropriate experience while, on the other, also be on point with regard to issues of information relevance, topicality, and point of view. This work is, in part, aimed at a new role for the machine in the world: a role in which the machine is used to expose the world of communication and cultural connections that are linked together by and within the grasp of online systems. The broader impacts of this work would range from creation of particular performance agents to more robust and scalable visions for AI. The project seeks to attract a wider set of students (including the underrepresented group of women in engineering) from communications studies and theater, who want to do work on the machine as a device for communication, communication made possible and supported by the mediation of intelligent systems.