基于互联网智商评测算法的搜索引擎智商测试研究

发布时间：2018-04-17 02:38

本文选题：互联网智商 + 人工智能　；参考：《北京交通大学》2016年博士论文

【摘要】：人工智能未来发展对于人类可能产生的威胁,目前越来越多的科学家和企业家表达了担心和忧虑,由此产生的人工智能威胁论引发了社会巨大争议。这争议背后本质上是人工智能系统能否定量评测的课题,本文以此问题作为研究的起点和基础,分析了评测人工智能智力发展水平面临的困难挑战,认为当前智能概念定义不清：人类智商测试与人工智能系统评测方法不统一；人工智能系统自身技术限制等诸多因素是造成上述问题的关键。虽然前人科学家已经为人工智能的评价体系做出了大量有意义的工作,但定量分析人工智能系统的智力水平问题一直存在瓶颈没有解决。在互联网快速发展的背景下,本文针对人工智能系统定量评测课题开展了以下工作并产生的创新点有：(1)在人工智能与互联网大数据深度结合,“互联网虚拟大脑”架构等研究基础上,提出了互联网及其应用也存在智力水平不断发展的问题。由此产生并定义了互联网智商和互联网应用的智商。(2)为解决统一测试人工智能系统和人类智力水平,建立“标准智能系统模型”,对人工智能系统和人类等生命系统进行了统一描述,本文同时建立了“标准智能系统知识交互模型”、“标准智能系统的评测模型”、“标准智能机”和“扩展冯诺依曼模型”。(3)在“标准智能系统模型”基础上,制定了互联网智商量表和评测方法,开发通用智力评测系统,使之能够同时对搜索引擎等人工智能系统,人类测试者进行测试,自动生成题库,自动计算其互联网智商。(4)利用开发的智力评测系统对世界50个搜索引擎和3个不同年龄段人类测试者进行测试,形成互联网绝对智商和离差智商排名。应用K-means算法进行聚类分析；应用支持向量机等算法验证聚类方法的有效性；应用分层抽样方法验证互联网智商测试题库的稳定性。根据测试结果对搜索引擎等系统未来发展提出建议,提出重点发展“知识的创新能力”可以帮助搜索引擎大幅度提高其智能水平。
[Abstract]:At present, more and more scientists and entrepreneurs have expressed their worries and worries about the threats that artificial intelligence may pose to human beings in the future, and the artificial intelligence threat theory has aroused great controversy in society.This controversy is essentially a question of whether artificial intelligence systems can be quantitatively evaluated. This paper takes this issue as the starting point and basis of the research, and analyzes the difficult challenges faced in evaluating the intelligence development level of artificial intelligence.It is considered that the concept of intelligence is not clearly defined, that the testing of human intelligence quotient is not consistent with the evaluation method of artificial intelligence system, and that many factors, such as the technical limitation of artificial intelligence system itself, are the key to the above problems.Although previous scientists have done a lot of meaningful work for the evaluation system of artificial intelligence, the problem of quantitative analysis of the intelligence level of artificial intelligence system has not been solved.In the context of the rapid development of the Internet, the following work has been carried out in this paper for the quantitative evaluation of artificial intelligence systems and the resulting innovations are: (1) the deep integration of artificial intelligence and Internet big data.On the basis of the research on the architecture of Internet virtual brain, it is pointed out that the intelligence level of the Internet and its applications is developing continuously.Thus, the Internet IQ and the intelligence quotient of Internet applications are defined. In order to solve the problem of unified testing of artificial intelligence system and human intelligence level, a "standard intelligent system model" is established.This paper gives a unified description of artificial intelligence system and human life system. At the same time, this paper establishes "Standard Intelligent system knowledge interaction Model", "Standard Intelligent system Evaluation Model","Standard Smart Machine" and "extended von Neumann Model". On the basis of "Standard Intelligent system Model", we developed the Internet IQ scale and evaluation method, and developed a universal intelligence evaluation system.To make it possible to test artificial intelligence systems such as search engines, human testers, and automatically generate question banks,Automatic calculation of its Internet IQ. 4) using the developed intelligence evaluation system to test 50 search engines around the world and 3 human testers of different ages to form the Internet absolute IQ and deviated IQ rankings.K-means algorithm is used to cluster analysis, support vector machine and other algorithms are used to verify the validity of the clustering method, and stratified sampling method is used to verify the stability of the Internet IQ test question bank.According to the test results, this paper puts forward some suggestions for the future development of search engine and other systems, and points out that the emphasis on developing "knowledge innovation ability" can help search engine to improve its intelligence level greatly.
【学位授予单位】：北京交通大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP391.3;TP18

【相似文献】