11-2   Expert systems and robotics

One practical application of AI has been in the area of expert systems. An expert system is a computer program that solves specialized problems at the level of a human expert. For example, Digital Equipment Corporation (DEC) uses an expert system called XCON to help its salespeople determine computer system configurations for their customers (Configurations include the components of a computer system and their interconnections). The system contains expert knowledge about configuring systems in the form of logical relationships or rules about how components must be connected together. For example, if a computer's configuration calls for communication, then the communication interfaces and cables must be included.

Prior to using XCON, tedious manual searches through parts catalogs and consultation with human experts were required to check the accuracy of each orders. With XCON, people can use the computer program that contains the knowledge of the configuration experts to simply and speed up filling complicated order, and DEC saves significant amounts of money in the process.

The preceding example points out some of the requirements for building an expert system. First, creating a knowledge base that is required for the application develops an expert system. A knowledge base contains facts and data that are specific to a particular problem area and rules that tell how to manipulate the facts or data stored. Unlike a conventional database, a knowledge base may include facts, assumptions, beliefs, expertise, and heuristic methods, which are exploratory methods for solving problems where an evaluation is made of the progress toward a goal using a series of approximate results. [1]

Creating a knowledge base requires that the system designer consult with a human expert. A person who creates an expert system is called a knowledge engineer—a computer scientist who designs and builds expert systems. The field of knowledge engineering grew out of the realization that building the knowledge base is one of the most important parts of an expert system. For example, obtaining knowledge from an expert is as important as actually developing the system itself, so a knowledge engineer must be trained in interviewing experts and in finding and developing reasoning procedures that match the expert's thought pattern. [2]

After the preliminary requirements and specifications of the expert system, a knowledge engineer has to develop two things: (1) the knowledge representation and (2) the reasoning strategies. The knowledge representation is the formal method for representing facts and rules about the area under consideration. A rule is a statement about the relationship of various facts or data.

For example, the first expert system, DENDRAL, helps chemists to identify the molecular structure of chemical compounds after they have obtained laboratory data using a tool called a mass spectrograph. Researchers know that the structure of a compound depends on a set of rules for chemical bonding. The knowledge engineer's job is to translate those rules into a set of rules that are used in the knowledge base.

The reasoning strategies are sometimes called an inference engine. An expert system reasons by processing symbols that represent objects. For example, when chemists make or discover a new compound, they can analyze it with a mass spectrograph, but they still need to figure out the specific shape that the compounds molecules will take. Even though the rules for chemical bonding are finite, there are millions of possible combinations based on a simple set of rules. The knowledge engineers involved in creating DENDRAL interviewed many chemists to find out their reasoning strategies for determining molecular structure from mass spectral data.

In traditional data processing, symbols generally represent numbers, letters, and mathematical operations. In an expert system, symbols can represent any type of object, and the simplest objects are called atoms, which are simply strings of characters. The processing of atoms consists of simple relational operations, such as matching character strings, joining character strings, and substituting one string for another.

By combining a knowledge base that consists of a number of facts about chemistry with a reasoning capability similar to that of a chemist, DENDRAL is able to outperform a human chemist in the specific task of narrowing down the choices of possible structures of a compound.

The word robot was chosen by Czech playwright Karel Capek in his play R.U.R. (Rossum's Universal Robots), in the early 1920s, because at that time the world was used to mean salve labor. Actually, the vision that many people have of robots is far from the reality of the working robots in factories. A robot is a programmable, general-purpose manipulator. Robots in the factory are replacing hard automation, that is, dedicated equipment that can perform only a single operation under a carefully controlled set of circumstances. Hard automation equipment is very specialized and, therefore, not adaptable to a wide variety of tasks. The design challenge in robotics is now to make general-purpose robots that can easily be programmed to perform a wide variety of tasks at costs that are less than that of a human worker.

Most people think of robots in terms of machines performing the work that people formerly did. Although this is certainly true, when robots are linked with information systems, factory automation becomes a much broader issue. Computer-integrated manufacturing (CIM)— computer-based systems in different company departments that are linked together—promises to transform the factory in much the way that office automation is transforming the office. [3] AI and robots are playing a major role in creating the factory of the future.

According to SRI International, a leading AI research organization, the goal of AI research in the factory is to produce programmable industrial automation, which is characterized by three features: (1) flexibility, or the capability to perform different tasks, (2) ease of training, or the capability of people to program new tasks, and (3) artificial intelligence, or the ability to perceive conditions that may be unpredictable and plan appropriate actions.

Now people are beginning to view robots not only as machines that replace people but as extensions of a company's information system, in which all aspects of manufacturing, materials handling, design, administration, production, assembly, quality control, packaging, and shipping are integrated and coordinated under one information system.

WORDS AND PHRASES

chemical bonding          化学结

Computer-integrated manufacturing (CIM) 计算机集成制造

hard automation           硬自动化

heuristic methods          启发式方法

inference engine           推理机

knowledge engineering     知识工程

knowledge representation   知识表示

mass spectrograph         质量光谱仪

outperform                超越，胜过

programmable industrial automation 可编程工业自动化

reasoning strategies       推理策略

robotics                 机器人技术

NOTES

[1] Unlike a conventional database, a knowledge base may include facts, assumptions, beliefs, expertise, and heuristic methods, which are exploratory methods for solving problems where an evaluation is made of the progress toward a goal using a series of approximate results.

与一般数据库不同，知识库应包含事实、假设、想法、专门知识和启发的方法，启发方法是一种解决问题的探寻法，它是用一系列近似结果构成向目标逼近的进程来求解问题。

[2] For example, obtaining knowledge from an expert is as important as actually developing the system itself, so a knowledge engineer must be trained in interviewing experts and in finding and developing reasoning procedures that match the expert's thought pattern.

例如，从专家那里获取知识与开发系统本身同样重要，因此知识工程师必须在与专家沟通和确立并开发具有专家思维方式的推理程序方面训练有素。

[3] Computer-integrated manufacturing (CIM)— computer-based systems in different company departments that are linked together—promises to transform the factory in much the way that office automation is transforming the office.

计算机集成制造（CIM）——分布于公司各部门、但相互连通的基于计算机的系统——使工厂就像办公室办公自动化一样改造一新。