THEORECTICAL INVESTIGATION OF THE LIMITS OF AI AND KNOWLEDGE REPRESENTATION

人工智能和知识表示的局限性的理论研究

• 批准号: 5203634
• 负责人: W WILBUR
• 金额: --
• 依托单位: NATIONAL LIBRARY OF MEDICINE
• 依托单位国家: 美国
• 资助国家: 美国
• 起止时间: 至
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/5203634
• 关键词: abstracting; artificial intelligence; information retrieval; statistics /biometry

Knowledge representation is a fundamental concern of AI research and has given rise to the subproblem of how to efficiently represent knowledge in a computer. We consider various aspects of the problem and have identified key concerns which any practical system applicable to large textual databases must deal with 1) No practical knowledge representation system has been developed at this time which allows the conversion of large amounts of text to machine readable form. This would require a solution to the language understanding problem. This is clearly not solved at this time. 2) There are theoretical limits to knowledge representation that must exist based on known limitations of human performance in interpreting natural language. These limits make it unreasonable to expect a machine solution unless we can achieve a performance with a machine superior to that of a human. The next step in investigating the limits of access to textual data in a large database is a study we have proposed to examine the role of subject knowledge in human performance at rating document pairs for relevance. We have already accumulated multiple human judgments made on two different tests sets of document pairs. The judges were all people knowledgeable in the field at some level. The proposal now is to have a number of judges who are literate in English, but who's area of expertise is not related to molecular biology, repeat the judgment task. The results will make a useful comparison to help delineate the contribution of subject knowledge to judgments. The goal is to better understand how humans use knowledge in selecting documents to meet a defined need.

知识表示是人工智能研究的一个基本问题， 引发了如何有效地表示知识的子问题 在电脑中。 我们考虑问题的各个方面并有 确定了任何适用于大型应用的实际系统的关键问题 文本数据库必须处理 1）尚未开发出实用的知识表示系统 这次允许将大量文本转换为机器 可读形式。 这需要语言的解决方案 理解问题。 这显然目前还没有解决。 2）知识表示存在理论上的限制，必须 存在基于人类口译表现的已知局限性 自然语言。 这些限制使得期望机器变得不合理 解决方案除非我们能够用一台机器实现优于 人类的。 下一步调查文本数据的访问限制 大型数据库是我们提出的一项研究，旨在检验 评级文档对时人类表现的学科知识 关联。 我们已经积累了多种人类判断 两个不同的文档对测试集。 评委都是人 在该领域具有一定程度的知识。 现在的建议是 许多法官都懂英语，但他们的领域 专业知识与分子生物学无关，重复判断任务。 结果将进行有用的比较，以帮助描述 学科知识对判断的贡献。 目标是为了更好 了解人类如何使用知识来选择文档来满足 明确的需求。