人工智能与专家系统外文文献译文和原文

人工智能与专家系统外文文献译文和原文 ARTIFICIAL INTELLIGENCE AND EXPERT SYSTEM 1. History of AI The seed of AI were sown only two years after General Electric installed the first computer for business use. The year was 1956, and the term artificial intelligence (AI) was coined by john McCarthy as the theme of a conference held at Dartmouth College . That same year, the first AI computer program, called Logic Theorist was announced. Logic Theorist’s limited ability to the reason (proving calculus theorems) encourage researchers to develop another program called the General Problem Solver (GPS), which was intended to solve problems of all kinds. The task turned out to be more then the early pioneers could handle. AI research continued, but it took backseat to the less ambitious computer applications such as MIS and DSS. Over time, however, persistent research continued to push back the frontiers of using the computer for tasks that normally require human intelligence. 2. Areas of AI AI is currently being applied in business in the form of knowledge systems, which use human knowledge to solve problems. The most popular type of knowledge-based system is the expert system. An expert system is a computer program that attempts to represent the knowledge of human expert in the form of heuristics is derived from the same Greek root as the word eureka, which means “to discover”. A heuristic is, therefore, a rule of good guessing. Heuristics do not guarantee results as absolutely as do conventional algorithms that are incorporated into DSSs, but they offer results that are specific enough most of the time to be useful. The heuristics allow the expert system to function in a manner consistent with a human expert, advising the user on how to solve a problem. Since the expert system functions as a consultant, the act of using it is called a consultation--the user consults the expert system for advice. In addition to expert system, AI includes work in the following areas: neural networks, perceptive systems, learning, robotics, AI hardware, and natural language processing. These areas are illustrated the way that one area can benefit the others. 3. The Appeal of Expert System The concept of expert system is based on the assumption that an expert’s knowledge can be captured in computer storage and then applied by others when the need arises. An expert system offers unique capabilities as a decisions support system. First, an expert system offer the opportunity to make decisions that exceed the manager’s capabilities . For example, a new investment officer for a bank can use an expert system designed by a leading financial expert and, in doing so, incorporate the expert’s knowledge into his or reaching a particular solution. Very often, the explanation of how a solution was reached is more valuable than the solution itself. 4. An Expert System Model The model of an expert system consists of four main parts. The knowledge base houses the accumulated knowledge of the particular problem to be solved. The inference engine provides the reasoning ability that interprets the contents of the knowledge base. The expert and the knowledge engineer use the development engine to create the expert system. 1. The User interface The user interface enables the manager to enter instructions and information into the expert system and to receive information from it. The instructions specify the parameters that guide the expert system through its reasoning processing. The information is in the form of values assigned to certain variables. (1) Expert System Inputs The most popular interface format today is the graphical user interface, which features a Windows look. Some systems employ a custom interface tailored to the problem being solved. For example, the screen might display a drawing of a mechical assembly. (2) Expert System outputs Expert system are designed to recommend solutions. These solutions are supplemented by explanations. There are two types of explanation: Expert system are designed to recommend solutions. These solutions are supplemented by explanations while the expert system performs its reasoning. Perhaps the expert system will prompt the manager to enter some information. The manager asks why the information is needed. The expert system provides an explanation. Explanation of the problem solution. After the expert system provides a problem solution, the manager can ask for an explanation of how it was reached. The expert system will display each of the reasoning steps leading to the solution. Although the inner working of the expert system can be complex , the user interface is user-friendly. A manager accustomed to interacting with a computer should have no difficulty in using an expert system. 2. The Knowledge base The knowledge base contains both facts that describe the problem area and knowledge representation techniques that describe how the facts fit together in a logical manager. The term problem domain is used to describe the problem area. （1）Rules A popular knowledge representation technique is the use of rules specifies what to do in a given situation technique is the use of rules. A rule specifies what to do in a given situation and consists of two parts: a condition that may or may not be true and an action to be taken when the condition is true. An example of a rule is: IF ECONOMIC.INDEX>1.20 AND SEASONAL.INDEX>1.30 THEN SALES.OUTLOOK=”EXCELLENT” All of the rules contained in an expert system are called the rule set. The rule set can vary from a dozen of rules .A dozen of rules for a simple expert system,and 500, 1,000, or 10,000 rules for a complex one. (2) Network of Rules The rules of a role set are not physically linked, but their logical relationships can be illustrated with a hierarchical diagram. The rules at the bottom of the hierarchy provide evidence for the rules on the upper levels. The evidence enables the rules on the upper levels to produce conclusions. The top level might consist of a single conclusion, indicating that the problem has only a single solution. The term goal variable is used to describe the solution, which can be a computed value, an action to be taken, or some other recommendation. For example, if an expert system is to advise top-level management on whether to enter a new market area, a value of Yes or Not would be assigned to the single-goal variable MARKET DECISION. It is also possible for the top level of the hierarchy to include multiple conclusions, indicting the possibility of more than one solution. An example is an expert system that makes recommendations concerning the best strategy to follow in reacting to increased competitive activity. The system might select from among possible strategies of improving the quality of the firm’s products, investing more in advertising, or lowering prices. 3. The Inference Engine The inference engine is the portion of the expert system that performs reasoning by using the contents of the knowledge base in a particular sequence. During the consultation, the inference engine examines the rules of the knowledge base one at a time, and when a rule’s condition is true, the specified action is taken. In expert systems terminology, the rule is “fired” when the action is taken. Two main methods have been devised for the inference engine to use in examining the rules: forward reasoning and reverse reasoning. (1) Forward reasoning In forward reasoning, also called forward chaining, the rules are examined one after another in a certain order. The order might be the sequence in which the rules were entered in to the rule set, or it might be some other sequence specified by the user. As each rule is examined, the expert system attempts to evaluate whether the conditions true or false. RULE EVALUSTION. When the condition is true, the rule is fired and the next rule is examined. When the condition is false, the rule is not fired the next rule is examined. It is possible that a rule cannot be evaluated as true or false. Perhaps the condition includes one or more variables with unknown values. In that case, the rule condition is unknown. When a role condition is unknown, the rule is not fired and the next rule is examined. THE ITERAIIVE REASONING PROCESS. The process of examining one rule after the other continues until a complete pass has been made through the entire rule set. More than one pass usually is necessary to assign a value to the goal variable. Perhaps the information needed to evaluate one rule is produced by another rule that is examined subsequently. For example, after the eleventh rule is fired, the fifth rule can be evaluated on the next pass.The passes continue as long as it is possible to fire rules. When no more rules can be fired, the reasoning process ceases. (2) Reverse Reasoning In reverse reasoning, also called backward chaining, the inference engine selects a rule and regards it as a problem to be solved. Using the rule set as shown in figure 20-1. Rule 12 is the problem, since it assigns a value to the goal variable P. The inference engine attempts to evaluate Rule 12 but recognizes that Rule 10 or Rule 11 must be evaluated first. Rule 10 and 11 become sub problems of Rule 12. The inference engine then selects one of the subproblems to evaluate, and the selected subproblem becomes the new problem. Figure20-1 Rules set THE FIRST LOGCAL PATH IS PURSUED. We will assume that Rule 10 becomes the problem. The inference engine then determines that Rule 7 and 8 must be evaluated before Rule 10 can be evaluated. Rules 7and 8 become the subproblems in this manner, searching for a rule that can be evaluated. THE NEXT LOGICAL PATH IS PURSUED. When the expert system attempts to evaluate Rule 11, Rule 9 becomes the problem; it can be evaluated using the outcomes of Rules 4 and 5. Because both Rules 4 and 5 are true, Rule 9 can be evaluated as true without the need to examined Rule 6. Once Rule 9 is fired, Rule11 can be fired as well. This makes it possible to assign a value to goal variable P, since Rule 12 is fired if either Rule 10 or 11 is true. (3) Comparing Forward and Reverse Reasoning Reverse reasoning proceeds faster than forward reasoning, because it does not have to consider all of the rules and does not make multiple passes through the rule set. Reverse reasoning is especially appropriate when: l There are multiple goal variables. l There are many rules. l All or most all of the rule do not have to be examined in the process of reaching a solution. Some inference engines are designed to perform both forward and reverse reasoning.The user can specify which one to use. 4. The Development Engine The forth major component of the expert system is the development engine, which is used to create the expert system. When the inference engine consists of rules, this process involves building the rule set. There are two basic approaches: programming languages and expert system shells. (1) Programming Language You can create an expert system using any programming language; however, two are especially well suited to the symbolic representation of the knowledge base: Lisp and Prolog . Lisp was developed in 1959 by john McCarthy ( one of the members of that first AI meeting ) , and Prolog was begun by Alain Colmerauer at the University of Marseilles in 1972. (2) Expert System Shells One of the first expert systems was Mycin, developed by Edward Shotlffle and Stanley Cohen of Stanford University, with the help of Stanton Axline, a physician. Mycin was created to diagnose certain infectious diseases. When the success of Mycin had been established, the developers looked for other ways tailored to apply their accomplishments. They discovered that the Mycin inference engine could be tailored to another type of problem by replacing the Mycin knowledge base with one reflecting the other problem domain. This finding signaled the start of a new approach to building expert system: the expert system sell. An expert system sell is a ready-made processor that can be tailored to a specific problem domain through the addition of the appropriate knowledge base. Today, most of the interest in applying expert system to business problems involves the use of sells. An example of a problem domain that lends itself to an expert system shell is help desk support. A help desk is a unit with-in the organization that provides technical help to users as well as to their own information specialists. In its most basic form, the help desk consists of one or more technical experts who receive users’ telephone calls for help. The user explains the problem and the technical expert suggests ways to solve it, perhaps referring to product manuals or other written sources. The help desk problem is so pervasive that a Helpdesk Institute was formed to facilitate dialogue among firms and industries with help desk expert system shells. When a firm uses one of the shells, it must populate the knowledge base with data concerning its own hardware and applications software. A software vendor can populate its knowledge base with data describing its software products, and so on. When a help desk expert system is used, either the user or the help desk staff member communicates directly with the system, and the system attempts to resolve the problem. One test of the degree of sophistication of artificial intelligence is whether the user cannot determine if he or she is interfacing with a human or a computer. This test has been called the Turing Test, in honor of the great pioneers in computer science, Alan Turing. The help desk expert systems use a variety of knowledge representation techniques. A popular approach is called case-based reasoning (CBR), which uses historical data as the basis for identifying problems and recommending solutions. Some systems employ knowledge expressed in the form of a decision tree, a network-like structure that enables the user to progress from the root through the network of branches by answering questions relating to the problem. The path leads the user to a solution at the end of branch. Expert system shells have brought artificial intelligence within the reach of firms that do not have the resources necessary to develop their own systems using programming language. In the business area, expert system shells are the most popular way for firms to implement knowledge-base system. 5. Advantages and Disadvantages of Expert Systems As with all computer applications, expert systems offer some real advantages; but there are also disadvantages. The advantages can accrue to both managers and the firm. 1. The Advantages of Expert Systems to Managers l Managers use expert systems with the intention of improving their decision-making. The improvement comes from being able to: l Consider More Alternative. An expert system can enable a manager to consider more alternatives in the process of solving a problem. For example, a financial manager who has been able to track the performance of only thirty stocks because of the volume of data that must be considered can, with the help of an expert system, track 300. By being able to consider a greater number of possible investment opportunities, the likelihood of selecting the best ones is increased. l Apply a Higher Level of Logic. A manager using an expert system can apply the same logic as that of a leading expert in field. l Devote More Time to Evaluating Decision Results. The manager can obtain advice from the expert system quickly, leaving more time to weigh the possible results before action has to be taken. l Make More Consistent Decisions. The computer does not have good days and bad days as the human manager does, Once the reasoning is programmed into the computer, the manager knows that the same solution process will be followed for each problem. 2. The Advantages of Expert Systems to the Firm l A firm that implements an expert system can expert: l Better Performance for the Firm. As the firm’s managers extend their problem solving abilities through the use of expert system, the form’s control mechanism is improved. The firm’s better able to meet its objectives. l To maintain Control over the Firm’s Knowledge. Expert systems afford the opportunity to make the experienced employees’ knowledge more available to newer, less experienced employees and to keep that knowledge in the firm longer—even after the employees have left. 3. The Disadvantages of Expert systems Two characteristics of expert systems limit their potential as a business problem-solving tool. First, they cannot handle inconsistent knowledge. This is a real disadvantage because, in business, few things hold true all the time because of the variability in human performance. Second, expert systems cannot apply the judgment and intuition that are important ingredients when solving semistructured or unstructured problems. 人工智能与专家系统 1．AI（人工智能）发展史 仅仅在通用电器公司开始将电脑应用于商业领域之后两年，即1956年，就出现了人工智能。人工智能这一术语是由John McCarthy在Ddartmouth大学的学术论坛上提出的。同年，第一个人工智能计算程序——Logic Theorist诞生了。Logic Theorist在推理方面的局限促使了研究人员开发另一个程序，那就是GPS(通用问题求解程序)。其目的是为了解决各种各样的问题，其解决问题的能力比前几代更强。

AI研究仍在继续，但与MIS和DDS等计算机应用相比，研究热情的减弱使人工智能的研究相对落后。然而，在研究方面的不断努力一定会推动计算机向人工智能化方向发展。

2．AI领域 AI现在已经以知识系统的形式应用于商业领域，既利用人类知识来解决问题。专家系统是最流行的基于知识的系统，他是应用计算机程序以启发方式替代专家知识。Heuristic术语来自希腊eureka，意思是“探索”。因此，启发方式是一种良好猜想的规则。

启发式方法并不能保证其结果如同DSS系统中传统的算法那样绝对化。但是启发式方法提供的结果非常具体 ，以至于能适应于大部分情况启发式方法允许专家系统能像专家那样工作，建议用户如何解决问题。因为专家系统被当作顾问，所以，应用专家系统就可以被称为咨询。

除了专家系统外，AI还包括以下领域：神经网络系统、感知系统、学习系统、机器人、AI硬件、自然语言处理。注意这些领域有交叉，交叉部分也就意味着这个领域可以从另一个领域中收益。

3．专家系统的吸引力 专家系统的概念是建立在专家知识能够存储在计算机中并能被其他人应用这一假设的基础上的。

专家系统作为一种决策支持系统提供了独无二的能力。首先，专家系统为管理者提供了超出其能力的决策机会。比如，一家新的银行投资公司可以应用先进的专家系统帮助他们进行选择、决策。其次，专家系统在得到一个解决方案的同时给出一步步的推理。在很多情况下，推理本身比决策的结果重要的多。

4．专家系统模型 专家系统模型主要由4个部分组成：用户界面使得用户能与专家系统对话；
知识库收藏了要特殊解决的问题；
推理引擎提供了解释知识库的能力；
专家和工程师利用开发引擎建立专家系统。

1．用户界面 用户界面能够方便管理者向专家系统中输入命令、信息，并接受专家系统的输出。命令中有具体化的参数设置，引导专家系统的推理过程。信息以参数形式赋予某些变量。

（1）专家系统输入 现在流行的界面格式是图形化用户界面格式，这种界面与Windows有些相同的特征。有些系统采用了与所要解决问题相称的个性化界面例如，屏幕可能会显示机械装配图。

（2）专家系统输出 专家系统一般是提供解决方案的。这些解决方案都是以如下两种方始输出的：
①解决方案解释。在专家系统提供了问题解决方案后，管理者可能还想知道是如何得到这种方案的。专家系统就会显示一步步到达结果的推理过程。

②问题解释。管理者可能希望得到专家系统对问题的推理过程。专家系统可能还需要管理者输入一些信息。管理者问为什么需要信息，然后专家系统就会提供解释。

虽然专家系统的内部工作很复杂，但是用户界面相当友好，方便使用。一个会用计算机的管理者，使用专家系统对他来说也肯定没有问题。

2．知识库 知识库即包括描述问题域，也包括以一定的逻辑描述事实的表示技术。术语“问题域”描述了所解决问题的业务领域。

（1）规则 规则是比较常用的表示技术。规则具体规定了在一种特定的情况下做什么。他有两部分组成：一是条件，有真和假；
二是方法，是指在条件为真的条件下采取的行动。以下是规则的一个例子：
IF ECONOMIC.INDEX>1.20ANDSEASONAL.INDEX>1.30 THEN SALES.OUTLOOK=”EXCELLENT” 包含在专家系统里的所有规则叫做规则集每个专家系统；
每个专家系统里的规则集数量是不一样的。一个简单的专家系统有几十条规则，复杂的专家系统有500或1 000甚至10 000条规则。

（2）规则网络 规则集里的规则再物理上并没有联系。但是他在逻辑上的关系可用层次图表示最底层的规则为上一级提供了依据。这些依据有助于上层的规则得出结论。

最顶层的可能只包含一个结论，这说明只有一个解决方案。目标变量是用来描述解决方案的。他可以是一个计算值一个可识目标，一种措施，或者一些建议。例如，如果一个专家系统是用来给管理者在是否要进入一个新市场决策上提供建议的，那么，单目标变量MARKET.DECISION的值就是Yes或No。

当然，也有可能在最高层得到多个结论，也就意味着有多种解决方案。例如，在关于提高市场竞争力战略决策中，专家系统可能就会提供所有可能的方案，如提高公司产品质量、增加广告投入量或降低价格。

3．推理引擎 推理引擎是专家系统的一部分，他根据特定顺序在知识库内容的基础上进行推理。

在咨询阶段，推理引擎挨个检查知识库规则，当某条规则的条件为真时就采取规定的行动。在专家系统中，当采取行动时，就称规则被激活。

在检查规则中，一般采用以下两种方法：正向推理和反向推理。

（1） 正向推理 在正向推理（也称为正向连接）中，规则是按照一定顺序逐个检查的。这种顺序可能是输入到规则集中的顺序，也可能是由用户自己定义的顺序。当检查每个规则之后，专家系统开始求值，既为“真”还是为“假”。

规则求值。当条件为真时，规则就被激活，然后再检查下一个规则。当然还存在规则的值即非“真”又非“假”的情况。这种情况下，规则的条件是不知到的，这是，规则不被取消，继续检查下一条规则。

迭代推理过程。挨个检查规则集中的规则，直到规则集中所有的规则都检查完毕。有时为了设定一个目标变量值往往要通过好几轮测试。可能测试这个规则所需要的信息是来自另一个规则测试的结果。比如，在第11个规则被激活后，第5个规则才进行测试。只要有规则被激活了，测试就继续，直到规则没有激活推理过程才结束。

（2） 反向推理 在反向推理（也称为反向连接）中，推理引擎将规则视为一个待解决的问题。如图20-1所视的规则集中，规则12是一个问题，因为他分配了一个值给目标变量P 。推理引擎试图得出规则12的值，但是，有图中可知，我们必须先要知道规则10和11的结果。规则10和11是规则12的子问题。推理引擎先要对子问题进行求值。

图20-1 规则集 选择第一条逻辑路径。我们假设当前规则10是待解决的问题。推理引擎在解决问题前首先要确定规则7和8的值。现在规则7和8是子问题，同样要解决这个子问题，先要用之前讲过的方法细分问题域，直到能够求值。

选择下一条逻辑路径。当专家系统尝试对规则11求值时，规则9成为问题。利用规则4和5的结果来对其求值。因为规则4和5都为真，所以规则9的值也为真。没有必要对规则6进行求值了。

规则9被激活后。规则11也被激活了。因为只要规则10或规则11其中一个为真，就可以激活规则12了，目标变量P的值也就可以得知。

（3） 正向推理和反向推理的比较 反向推理比正向推理要快。因为反向推理不必考虑所有的规则，也不用一轮一轮在规则中求值。反向推理尤其适用于以下几种情况：
①多个目标变量；

②有很多的规则；

③在求的问题结的过程中无须将所有的或几乎所有的规则都检查一便。

有些推理引擎即适合正向推理也适合反向推理，视具体情况而定。

4．开发引擎 专家系统的第4个重要组件就是开发引擎。他用来建造专家系统。当推理引擎包含许多规则时，建造专家系统的过程就涉及到建立规则集。有两种基本方法：程序语言或专家系统外壳程序。

（1） 程序语言 你可以应用任何语言创建专家系统，但最适合符号化表示知识库的两种语言是：Lisp和Polog。Lisp是在1959年由McCarthy（首届AI会议的成员之一）开发的。Prolog是在1972年由Alain Colmerauer在Marseilles大学开发的。

（2） 专家系统外壳程序 第一个专家系统是Mycin，是由Stanford大学的Edward Shortliffe和Stanley Cohen在物理学家Stanton Axline的帮助下开发的。Mycin是用来诊断某种传染病的。

当成功开发第一个专家系统Mycin后，开发者们试图在别的各个领域应用这个成果。他们发现如果将知识库更换成反映另一个问题的相关知识Mycin推理引擎能够适用于该类型的问题域。这种发现开创了建立专家系统的新方法：专家系统外壳程序。他是一段预先编写好的程序，只要增加相应的知识库就能够适用于一个具体的问题域。如今应用专家系统解决商业问题的焦点在于外壳程序的应用。

由问题域导出专家系统外壳程序，其中的一个例子就是桌面帮助支持。桌面帮助支持就是系统的一个单元，为用户提供技术帮助。信息服务单元典型的给用户和信息专家提供桌面帮助。桌面帮助最基本的形式就是一两个专家给用户进行电话答疑。用户提出问题，专家予以解答。

桌面帮助问题是如此的普遍，以致于再公司成立了桌面帮助部门以方便对话。在年会上，最重要的一项活动就是演示专家系统的外壳程序的桌面帮助。当一个公司应用其中一个外壳程序时，他必须扩充相关生产线的知识库。比如，信息服务单元应该扩充硬件和应用软件的相关数据，在软件的帮助库中扩充软件描述等。

当桌面帮助专家系统得以应用，用户以及桌面帮助员工就可以直接跟专家系统对话，系统就可以解决问题。人工智能的智能化程度的一个测试就是用户是否不能判别出 他是在跟机器还是在跟人对话，这种测试称为Turing测试。Alan Turing是计算机学伟大的先驱之一。

桌面帮助专家系统利用不同的信息表示技术。比较流行的方法是CBR(case-based reasoning，基于事实的推理)。他是根据历史数据作为识别问题的基础，然后提出解决方案。有些系统是以决策树的形式来表示的。他是一个网状结构，使用户能够回答与解决相关的问题。

专家系统外壳程序引入了人工智能，使公司没有必要开发他们自己的系统。在商业领域，公司经常使用专家系统外壳程序来实施基于知识的系统。

5．专家系统的优缺点 跟其他计算机应用一样，专家系统提供了一些实际利益，但也有一些不足之处。管理者和公司都可以从专家系统中收益。

1.家系统为管理者带来得好处 管理者应用专家系统改进决策。这些改进表现如下：
（1）提供更多的选择。在解决问题过程中专家系统能促使管理者考虑到更多的选择。比如，没有专家系统，由于考虑范围有限，财务经理只能跟踪30种股票的表现。但是有了专家系统，就可以跟踪300种股票。考虑的投资范围的扩大，也就增加了选择最佳方案的可能性。

（2）应用更高的逻辑层。管理者借助于专家系统，能够达到最先进的专家逻辑水平。

（3）倾注更多的时间于评估方案之上。管理者能够快速的从专家系统中得到建议，给管理者在行动之前留下更多选择和权衡的时间。

（4）决策更加一致。与管理者相比，计算机不会有搀杂个人情感的波动因素，一旦将推理输入到计算机，管理者就会得到确定的方案。

2.为公司带来得好处 专家系统为公司带来如下好处：
（1）公司有更好的业绩。因为管理者是借助于专家系统解决问题的，所以公司的管理机制得到大大的改善公司能够更好的接近目标。

（2）保持对公司知识的控制。专家系统为老员工传授丰富的经验给新员工创造了机会。即使员工离开后，也能够使知识自成一体。

6.专家系统的缺点 专家系统的两个特征限制了将其作为商务问题解决工具的潜能。第一，他们不能处理一致性知识的问题。这是一个实实在在的不足之处，因为在商业中，由于人为因素的可变性，没有事情时时正确。第二，专家系统不能应用判断和指导，而在解决结构化问题时他们是很重要的因素。