Introduction
This section introduces some of the software systems used in knowledge engineer- ing and shows how they work compared to human processing.
Objectives
By the end of the section you will be able to:
r explain the difference between knowledge engineering and artificial intelligence r define knowledge-based systems (KBSs)
r explain what a KBS can do
r explain the differences between human and computer processing
r provide a brief definition of expert systems, neural networks, case-based reason- ing, genetic algorithms, intelligent agents, data mining and intelligent tutoring systems
What Is the Difference Between Knowledge Engineering and Artificial Intelligence?
To try and provide a simple answer to this question, consider each of the following life forms:
r Plants r Fish
r Chimpanzees r Humans
Now, do the plant and animals in the example above exhibit evidence of intelli- gence?
Activity 9
We have noted that a knowledge engineer must be able to capture the be- havioural skills or knowledge of experts and code these into some KBS. If you were a knowledge engineer, what particular behavioural skills or knowledge, in generic terms, would you expect to find in the objects listed below?
If you picture the four objects listed below, this will help you see the different skills that are displayed by them. Think specifically of the movement (or lack of) for each object, as well as the communication skills that could be expected.
rA plant rA fish rA chimpanzee rA human.
Feedback 9
The skills may include:
A plant
rAdapt in time and evolve—an individual plant has no skills but as a species they do.
A fish
rNavigation rVisual recognition rAvoid danger.
A chimpanzee
rLanguage/communication about concrete concepts rUse of basic tools
rSimple problem solving rMimic humans
rBuild mental models.
A human
rLanguage/communication about complex concepts rLearn from being told
rLearn from the past experience rIdentify cause and effect relationships rTeach
rSolve complex problems rDesign, plan and schedule rCreate complex abstract models rShow initiative.
While we may not consider fish to be intelligent, they do exhibit some complex characteristics that can be considered aspects of intelligence. They navigate around the world, and visually recognise other animals. They can also plan to avoid danger.
All these are all aspects of intelligence, and when applied to computer systems could not be implemented by traditional computing techniques.
Chimpanzees are clearly more intelligent than fish. They have the ability to use language. They use basic tools, sticks and stones. They can solve simple problems, mimic humans and have been shown to build mental models of their environment.
Finally, humans are clearly more intelligent again. They can:
r learn by being told
r learn from past examples and from experience r teach
r solve complex problems, design, plan and schedule r create complex, abstract models of the universe.
Further, one common feature that fish, chimpanzees and humans share is that we are all unique individuals. Within the scope of our mental capacity we have individual choice and make our own decisions. This again is evidence of intelligence.
The application of artificial intelligence has tried to emulate all of these charac- teristics within computer systems. Knowledge engineers have the difficult job of attempting to build these characteristics into a computer program.
By using a range of techniques, including expert systems, neural networks, case- based reasoning, genetic algorithms, intelligent agents and data mining, we can get computer systems to emulate some aspects of intelligent behaviour such as:
r making decisions, diagnosing, scheduling and planning using expert systems or neural networks
r evolving solutions to very complex problems using genetic algorithms
r learning from a single previous example, where this is particularly relevant and using it to solve a current problem using case-based reasoning
r recognising hand writing or understanding sensory data—simulated by artificial neural networks
r identifying cause and effect relationships using data mining
r free will, i.e., the ability to take independent actions—simulated by intelligent agents.
For example, legal systems can suggest suitable fines based on past examples using case-based reasoning—a type of KBS you will encounter later in more detail.
Programs can also process human language including grammar checking, sum- marisation and translation—all of which use natural language processing tech- niques (not covered in this book).
Artificial intelligence aims to endow computers with human abilities. Often this involves research into new and novel technologies that might not be immediately usable.
Knowledge engineering, on the other hand, is the practical application of those as- pects of artificial intelligence that are well understood to real commercial business problems such as recognising signatures to detect potential fraud.
What Are KBSs?
Knowledge-based systems are computer programs that are designed to emulate the work of experts in specific areas of knowledge.
It is these systems that provide the main focus of this book.
There are seven main types of KBS:
1. Expert systems. Expert systems model the higher order cognitive functions of the brain. They can be used to mimic the decision-making process of human experts. Typical example applications include planning, scheduling and diag- nostics systems.
Expert systems are normally used to model the human decision-making process.
Although expert systems contain algorithms, many of those algorithms tend to be static, that is they do not change over time. While this provides some certainty in how the system will operate, it does mean that the expert system is not designed to learn from experience.
It is worth mentioning that expert systems are very often spoken of as synony- mous with KBSs. However, expert systems are simply a category of KBS.
2. Neural networks. Neural networks, on the other hand, model the brain at a biological level. Just as the brain is adept at pattern recognition tasks, such as vision and speech recognition, so are neural network systems. They can learn to read, can recognise patterns from experience and can be used to predict future trends, e.g. in the demand for electricity.
3. Case-based reasoning. Case-based reasoning systems model the human ability to reason via analogy. Typical applications include legal cases, where the knowl- edge of the law is not just contained in written documents, but in a knowledge base of how this has been applied by the courts in actual situations.
4. Genetic algorithms. A genetic algorithm is a method of evolving solutions to complex problems. For example, such a method could be used to find one of many goodsolutions to the problem of scheduling examinations (rooms,
students, invigilators and possibly even equipment) from the millions ofpossible solutions.
The term ‘genetic’ refers to the behaviour of algorithms. In this situation, the behaviour is very similar to biological processes involved in evolution.
5. Intelligent agents. An intelligent agent is, normally, a software program where its goal or overall task is specified but where the software can make some decisionson its own
Most agents work in the background (that is they are not seen by the user) and only appear to report their findings. They may work over the Internet looking for important information where the user simply does not have time to sift through all the reports presented to him or her.
Agents often have the ability to learn and make increasingly complex decisions on behalf of their users. The simplest agents simply retrieve information while the most complex learn and use deductive reasoning to make decisions.
6. Data mining. Data mining is a term used to describe knowledge discovery by identifying previously unknown relationships in data. Alternative terms for mining include knowledge extraction, data archaeology, data dredging and data harvesting.
The technique relates to the idea that large databases contain a lot of data, with many links within that data not necessarily becoming evident until the database is analysed thoroughly. One of the classic examples of data mining concerns the analysis of sales within supermarkets. Data mining techniques could potentially identify products often purchased at the same time such as nuts and crisps. By placing these items next to each other on the supermarket shelf the sales of both products can be increased as they can now be found easily.
Data mining is used in many different areas of business, including marketing, banking, retailing and manufacturing. The main aim of data mining in these situations is to uncover previously hidden relationships and then use this infor- mation to provide some competitive advantage for the organisation.
7. Intelligent tutoring systems. The interest in computer-based instructional envi- ronments increases with the demand for high-quality education at a low cost.
Meanwhile, computers become cheaper, more powerful and more user-friendly.
An environment that responds in a sophisticated fashion to adapt its teaching strategy to the specific learning style of a given student/user is highly attractive.
For a tutoring system to be intelligent, it must be able to react (teach) continu- ously according to a student’s learning. Most tutoring systems try to use a single teaching method but with various levels of explanations/examples/disclosure of domain materials to react to different student’s learning. However, a teacher in practice will use more than one teaching method in teaching a subject accord- ing to the type of domain knowledge. In order to be intelligent and effective in teaching, a tutoring system must be able to provide multiple teaching methods.
An example is available at: http://www.pitt.edu/∼vanlehn/andes.html.
With the exception of intelligent tutoring systems, the systems mentioned above are discussed in greater detail later in this book.
What Can KBSs Do?
A KBS can perform many of the tasks undertaken by humans. However, they do have some limitations, as the examples below explain.
When compared with human expertise—which is often not very accessible since only one or a few people can consult the expert at once—artificial expertise, once captured in some form of KBS, is permanent and open to inspection. Expert systems have been used to capture the knowledge of expert staff who are due to retire and cannot be replaced, for example.
Where human expertise is difficult to transfer between people, the knowledge within any KBS can be re-used and copied around the world. Where humans can be unpredictable, KBSs are consistent. Where human expertise can be expensive and take decades to develop, KBS can be relatively cheap.
On the other hand, humans are creative and adaptable, where KBSs are unin- spired and developed for fixed purpose. Humans have a broad focus and a wide understanding. Knowledge-based systems are focused on a particular problem and cannot be used to solve other problems.
Humans can fall back on common sense knowledge and are robust to error.
Knowledge-based systems are limited to the technical knowledge that has been built into them. Humans are also very good at processing sensory information.
While neural networks can also handle sensory data, expert systems are generally limited to symbolic information.
Summary
There are a variety of KBSs, each designed to attempt to emulate different aspects of human intelligence, knowledge and behavioural skills.
Self-Assessment Question
For each of the four entities listed below, identify the different behavioural skills or knowledge they display that contribute or provide evidence of their ‘intelligence’.
For one example of each skill, indicate what type of KBS is designed to emulate it.
A plant A dog A dolphin A human.
Answer to Self-Assessment Question
You should have been able to answer approximately as follows:
A plant
r Adapt in time and evolve—genetic algorithms.
A dog
r Navigation-expert systems
r Visual recognition—neural networks r Avoid danger—neural networks.
A dolphin
r Language—neural networks (used in speech recognition) and natural language processing (not covered in this book)
r Simple problem solving—expert systems
r Build mental models—case-based reasoning-expert systems.
A human
r Reason by analogy—case-based reasoning r Learn from being told—expert systems
r Learn from past experience—case-based reasoning and neural networks r Identify cause and effect relationships—data mining and neural networks r Teach—intelligent tutoring systems
r Solve complex problems—expert systems/genetic algorithms r Design, plan and schedule—expert systems and genetic algorithms r Create complex abstract models—expert systems
r Show initiative (or at least emulate individual choice and decision making)—
intelligent agents.
References
Davis, R. (1979). Interactive transfer of expertise: Acquisition of new inference rules.
Artificial intelligence, 12: 121–157.
Debenham, J. K. (1988).Knowledge Systems Design. Prentice-Hall: Englewood Cliffs, NJ.
Drucker, P. F. (1988). The coming of the new organisation.Harvard Business Review, 66(1):39–48.
Fensel, D. (1995).The Knowledge Acquisition and Representation Language KARL. Kluwer Academic Publishers: Amsterdam.
Harry, M. (1994).Information Systems in Business. Pitman Publishing: Boston, MA, p. 50.
Hayes, R. (1992). The measurement of information. In Vakkari, P. and Cronin, B. (editors), Conceptions of Library and Information Science. Taylor Graham: London, pp. 97–108.
Laudon, K. C. and Laudon, J. P. (1998).Management Information Systems: New Approaches to Organisation and Technology, 5th ed. Prentice-Hall: Englewood Cliffs, NJ, p. 8.
Long, L. and Long, N. (1998).Computers, 5th ed. Prentice-Hall: Englewood Cliffs, NJ, p. 5.
McNurlin, B. and Sprague, R. H., Jr. (1998).Information Systems Management in Practice, 4th ed. Prentice-Hall: Englewood Cliffs, NJ, p. 197.
Senn, J. A. (1990).Information Systems in Management. Wadsworth Publishing: Belmont, CA, p. 58.
Zachman, J. (1987). A framework for information systems architecture.IBM Systems Jour- nal, 26(3):276–292.
2
Types of Knowledge-Based Systems
Introduction
This chapter builds on the brief introduction to different types of knowledge-based systems from the first chapter and provides you with the opportunity to explore them in greater depth.
The chapter consists of six sections:
1. Expert systems 2. Neural networks (NNs) 3. Case-based reasoning (CBR) 4. Genetic algorithms
5. Intelligent agents 6. Data mining.
Objectives
By the end of the chapter you will be able to:
r describe the characteristics of a knowledge-based system
r explain the main elements of knowledge-based systems and how they work r evaluate the advantages and limitations of knowledge-based systems
r identify appropriate contexts for the use of particular types of knowledge-based systems.
26