CAREER: Evaluation as the Basis for AI Agent Design

许多软件开发人员将评估视为一种负担，因为它目前需要付出巨大的努力，而且大多数程序员没有接受过这方面的培训。 在 PI 的专业领域（人工智能中的规划和代理设计）的背景下，该项目的目的是解决当前方法论和教学法中的两个差距：人工智能软件代理评估方法的缺乏以及本科生和研究生计算机科学课程中缺乏实证和评估方法。 为了实现这些目标，研究部分创建了新的评估方法，支持在两个代理项目的背景下设计和开发人工智能代理：模拟机器人和万维网信息收集代理。 教学部分将评估方法整合到本科生和研究生计算机科学课程中，尽可能将模块纳入现有课程，并在必要时设计新课程。 每个组成部分都利用现有和新的研究项目来探索激励设计的评估问题。预期的结果不仅是新的软件评估方法、其使用的记录示例和两个经过全面评估的软件代理，而且还包括经验丰富的新从业者。

Many software developers view evaluation as a burden both because it currently takes significant effort and because most programmers are not trained in it. In the context of the PI's area of expertise (planning and agent design in Artificial Intelligence), the purpose of this project is to address two gaps in current methodology and pedagogy: the dearth of evaluation methods for AI software agents and the absence of empirical and evaluation methods in the undergraduate and graduate computer science curricula. Towards these ends, the research component creates new methods for evaluation that support AI agent design and development in the context of two agent projects: a simulated robot and an information gathering agent for the WWW. The teaching component integrates evaluation methods into the undergraduate and graduate CS curriculum by incorporating modules into existing courses where possible and designing new courses where necessary. Each component exploits existing and new research projects for exploring issues of evaluation in motivating design. The expected results are not only new software evaluation methods, documented examples of their use and two thoroughly evaluated software agents, but also experienced, new practitioners.