Back

Automated decision making

Introduction
Programers can design and write programs based on business rules that trigger events. They can do this because the decisions that are made are based on data held in a digital format. Where the system is repetitive, well understood, the rules are clear and the data can be relied on, automated decision making systems can remove the need for humans to act, improving response speeds and removing any subjective (and potentially flawed) decision making. 

Knowledge-based systems (Expert systems)
A 'knowledge-based system' is a synonym of an 'expert system'. It is "an application of artificial intelligence to a particular area of activity where traditional human expert knowledge and experience are made available through a computer package" (British Computer Society, 'A Glossary of Computing Terms'). In other words, it is a piece of software that has a go at replacing experts' knowledge and experience. In this section, we will describe the components of the software that make up an expert system. We will then look at some typical applications.

The components of knowledge-based software
Knowledge-based software systems have four identifiable parts to them. These are:

    1. The knowledge base.
    2. The rule base.
    3. The inference engine.
    4. The user interface.

1) The knowledge base
The knowledge base is the name given to the part of the software that holds the facts. It contains the knowledge for a particular area of expertise, such as a medical diagnostic expert system or an engine fault diagnostic system.

2) The rule base
This is an identifiable part of the knowledge base. It holds the rules that are applied to the facts. For example, one rule might be:

    • IF the toe is swollen
    • AND the nail is black
    • THEN it could be an in-growing toenail

3) The inference engine
This is a piece of software that takes requests for information from the user interface and then searches through the knowledge base by applying the rules in the rule base. The inference engine retrieves appropriate knowledge from the knowledge base and passes it to the user interface software. It separates the user interface from the 'clever' part of the software. That means that different applications can be written by different people, each one having their own user interface - and they will all be able to use the same expert system.

4) The user interface
The user enters requests for information. They may do this by entering in answers to closed questions (questions which have only a few possible answers). The answers will result in the knowledge base being reduced further and further until only limited facts are left. These can then be returned to the user along with a probability factor. These four components can be represented with the following diagram: 

Components

Talking about an expert system
We know that if we want to describe what an expert system is, we need to mention the knowledge base, the rule base, the inference engine and the user interface. We could also highlight the characteristics of an expert system in the following way:

    • Expert systems are quite specialised. Each expert system usually attempts to be an expert in a very focused area of expertise.
    • When answers are returned, they are often given probabilities. For example, in a medical diagnostic expert system, a high temperature, headaches and a cough might return
        • Common cold - 61.78% chance.
        • Influenza - 38.21% chance.
        • Bubonic plague - 0.01% chance.
    • A user is often asked a series of questions and must select an answer from perhaps two or three choices for each question. When the answers have been selected via the user interface, they are passed to the inference engine. This then triggers a search of the knowledge base using the rules in rule base. The resulting facts with a probability score are then passed back to the inference engine and on to the user interface.
    • Sometimes an explanation or advice may be part of the answers returned.

Types of knowledge-based systems
There are three types of knowledge-based systems that you should know about. These are:

    1. Diagnostic systems.
    2. Advice-giving systems.
    3. Decision making systems.

For each type, you should be able to describe a classic example.

1. Diagnostic systems
These are systems that ask a question, the answer given by the user resulting in the knowledge base being reduced in size. The classic example is a medical diagnostic system.

    • The first question might be 'Do you have constant headaches'? If the answer is 'Yes', then one portion of the knowledge base can be explored further. If the answer is 'No', then a different portion can be explored. Whatever the answer, the number of facts that will be looked at has been reduced.
    • Suppose the answer to the above question was 'Yes'. The next question might be 'Do you have a sore throat'? Again, the answer can be either 'Yes' or 'No'. If it is 'Yes' then that reduces the part of the knowledge base for people with headaches even further to a part for people with headaches and sore throats. If it is 'No', then that will reduce the part of the knowledge base for people with headaches but without sore throats.
    • Suppose the answer was 'No'. The next question might be 'Do you have a high temperature'? This process will continue until the number of facts left is very small. These can then be reported to the user, perhaps with a probability factor and some additional comments.
    • We can represent the process with the following diagram. Notice how the knowledge base gets smaller and smaller with each question answered. 
Diagnostic

Another example of this kind of system is a car engine diagnostic system. When a car has a problem with its engine, the problem can be diagnosed with the help of an expert system. For example, the first question might be 'Is there an oil leak'? If there is, that would reduce the knowledge base to an area that deals with problems involving oil leaks. If there isn't, that would reduce the knowledge base to an area that deals with problems not involving oil leaks. Another question could be asked that further reduces the knowledge base and so on, until the number of possible causes of the problem are small enough to report back to the user.

2. Advice-giving systems
You have already seen a classic example of this type of knowledge-based system, when you looked at stock-control systems. In this kind of system, a process is monitored by the software. When the software detects particular situations, it gives advice! In a stock control system in a shop, the stock is constantly monitored: the amount of a product in stock is decremented each time a customer buys a product and has it scanned and it is increased each time a delivery from the main warehouse is received. The stock control system 'knows' that when a product falls below a certain number, it should advise that a re-order take place. Stock control systems can also be called 'decision-making' knowledge-based systems if they are able to re-order automatically, without any human intervention.

Another example is the safety systems fitted on planes that help them avoid collisions. As the plane flies, it scans ahead, checking that the path is clear of other planes. If it does detect a plane coming towards it, it should advise the pilot to change course. Like stock control systems, these systems can also be called 'decision-making knowledge-based systems if they are able to automatically change course to a safe route, without human intervention.

3. Decision-making systems
In the section above, we saw that advice-giving systems can also be called 'decision-making systems' if they have enough information and the hardware and software to allow them to actually take decisions themselves, without human intervention. Many process systems are both, because they enable a user to select whether the system will be fully automated, or will simply report and advise. An automated share-selling expert system can be both. Instructions can be given to the software that tell it to automatically sell a share when it reaches a certain price or it can be told to give a message to the stockbroker, advising that the shares be sold. The stockbroker can then review the situation, discuss it with the owner of the shares and then take action. If you ever buy insurance for a car or a holiday, or need to apply for a loan, for example, you will be asked a series of questions and then given a quote based on your answers. This process can be done online automatically by software and usually requires no human intervention. If you phone up to speak to someone, they will be running through a list of tightly scripted questions and typing in your responses to a program and then giving you the program's answer. Very little if any flexibility is allowed and the job of an insurance consultant is reduced to a lower skilled one (and therefore a lower salary one). If you ever play a game like chess against a computer, you will be playing against a decision-making program. You can normally adjust how 'good' the computer should be and this increases or decreases the number of different permutations that are investigated. 

A stock control system -  an example of a decision-making expert system
Many shops now use stock control systems. The term 'stock control system' can be used to include various aspects of controlling the amount of stock on the shelves and in the stockroom and how reordering happens but they typically make use of automated decision making techniques. Features of any stock control system usually include:

    • Ensuring that products are on the shelf in shops in just the right quantity by automatically monitoring levels of stock.
    • Recognising when a customer has bought a product.
    • Automatically signalling when more products need to be put on the shelf from the stockroom.
    • Automatically reordering stock at the appropriate time from the main warehouse.
    • Automatically producing management information reports that could be used both by local managers and at Head office. These might detail what has sold, how quickly andat what price, for example. Reports could be used to predict when to stock up on extra products, for example, at Christmas or to make decisions about special offers, discontinuing products and so on.
    • Sending reordering information not only to the warehouse but also directly to the factory producing the products to enable them to optimise production.

Stock control systems ensure that just the right amount of stock are on the shelves. If there is too much stock, it ties up a company's money, money that might be better spent on reducing the overdraft, on advertising the business or on paying for better facilities for customers, for example. Too much stock means that some perishable products might not sell and would have to be thrown away and this would reduce a company's profit. If there were not enough products on the shelf, they might run out. If this happens, they would lose business and again, profits would not be as good as they ought to be. Stock control systems save a lot of staff time. Savings may be possible by reducing the number of staff needed in the business thereby improving profits. A stock control system will not remove the necessity for checking what is on the shelves regularly - things get stolen and these won't be recorded. Stock control systems also mean that a business may have to close down while the system is changed from a manual one. They also involve a considerable investment in equipment and support. Stock control systems require training and some staff may find them difficult to use. They can also break down so a procedure needs to be in place so the business can continue to trade. This may involve further costs as well, perhaps in the purchase of backup equipment or in the purchase of a support agreement. Usually, the benefits of a stock control system outweigh disadvantages.

Some potential disadvantages of automated decision-making systems
Software can never completely replace a human. For one thing, a human has hunches and intuition. Some might argue that removing these elements is a good thing! Humans can, however, evaluate completely new situations whereas an expert system is bound by rigid rules and their current knowledge base. Indeed, an expert system's knowledge base and rule base must be kept up-to-date. This may seem straightforward but if the expert system is in a third world country, for example, it means that some training will be needed so that somebody locally can do this. To have an expert system means that the hardware and software must be bought. Not only may this be a problem in some situations, but careful thought needs to be given to maintaining the hardware and software - you don't want the system to stop being used the moment there is a problem, simply because nobody has been trained to maintain the system. Expert systems may well be a wonderful substitute for situations where it is impossible to recruit highly trained people. They should also be much cheaper than the equivalent human. It may simply not be possible to pay for enough doctors to work in third world countries, or it may not be possible to recruit enough stockbrokers to work in a firm. There are some dangers, though. It may be that people come to rely too much on the power of the expert system! These systems are never perfect. They may contain software bugs. They may cause situations that have not previously been considered. Expert systems that automatically bought and sold shares around the world nearly caused a stock market meltdown in the 1990s because they fed off each other and caused each system to sell sell sell! 

Back