The Role of Ai in Decision Support Systems

Introduction

Since the 1960's scientists have promised that by the 21st century, robots and artificial intelligence would be available in every household, taking care of all the mundane tasks such as cleaning, cooking, and even gardening (May, 2008).

Today we live in the 21st century without robots that do all the cooking and cleaning. Artificial intelligence, however, is not a dead science either. AI is widely used in computer games, medical applications and prosthetics, space exploration, artistic development, and also in business.

This document will look into how the field of artificial intelligence (AI) is still very much alive today. The specific focus will be on how AI is used by businesses around the world to support the decision role and assist managers and executives to make better use of resources.

An Introduction to Artificial Intelligence

Although the theory of intelligent machines have been around for decades, it wasn't until 1956 when a mathematics professor named John McCarthy invited scientist to a conference in Dartmouth, that the ideas behind artificial intelligence finally took the first steps to becoming reality (Buchanan, 2005).

The conference addressed some of the basic issues related to problem solving namely, teaching a computer to use a language, teaching a computer how to abstract information from a problem and then calculate the best action to take to solve such a problem, and finally, teaching a computer to do this without user interaction (Buchanan, 2005).

In order to address these problems computer scientists have developed several different techniques, each of which attempts to learn from its environment in a unique and different way. Many of these techniques have been adopted by the business industry to support managers and professionals in making quick and correct decisions.

Decision Trees

Perhaps one of the most basic forms of artificial intelligence, a decision tree simply plots all possibilities that can arise from a particular action and weighs them and their consequences to determine if the action should be taken. Decision trees are often represented graphically to a user in order for them to quickly see all possible effects and outcomes that the particular action can have on the business (Matthews, 2003).

Businesses commonly make use of decision trees when deciding on investing finances in developing a new product. Using a pharmaceutical company as an example, a possible decision tree is illustrated in Figure 1.

Figure 1 - Decision tree for the development of a new drug (Matthews, 2003)

In the example illustrated above, a pharmaceutical company expects a profit of $200million if the drug is sold on the market. However, the company expects that there is only a 25% chance that the drug will be successful at curing the disease, thus leaving a 75% probability that the company will lose $12million on research and development costs. If the drug is successful, it still has to undergo FDA approval at which point there is a 60% probability of being approved and a 40% chance of the drug being rejected, which would result in an additional $20million loss for the company.

A decision tree can then be used to calculate the value of choosing to invest in the research and development of the drug by using the following risk calculations:

Value of FDA Approved Drug=($200mil=0.6)+ (-$20mil×0.4)

= $120mil - $8mil

= $112mil

The probable financial outcome of sending the drug for FDA approval is therefore a favourable $112million for the company. Therefore the company can now calculate the value of the node that considers whether or not the drug works by using the $112million calculated above.

New Drug Works=($112mil=0.25)+ (-$12mil×0.75)

= $28mil - $9mil

= $19mil

This calculation shows a still favourable $19million outcome to research and develop the new drug. Comparing this possible gain against the gain of not developing the drug ($0), it is clear that the company can gain more from investing in the development of the new drug, and thus the decision is made to do so. However, if the probability of the drug working was only 9%, the decision tree would change to that represented in Figure 2 below.

Figure 2 - Working Drug Probability only 9% (Matthews, 2003)

Now it is clear that it is financially more advantageous for the company not to invest in the development of the new drug and that the focus should be on other projects.

This demonstrates how decision trees can be used effectively by managers and executives to quickly calculate all possible outcomes from a single choice and select the best course of action to take.

Genetic Algorithms

Genetic algorithms is a form of artificial intelligence which mimics evolution over many generations. In a genetic algorithm, a problem is presented to the system which then attempts to solve the problem with a variety of different solutions. Each solution is then awarded a success factor which in turn is used to determine which solutions best solve the problem. The solutions with the most favourable success factors are then used to create a new series of solutions where the evaluation process is then again applied. At random intervals, a "mutation" is introduced in to ensure that the system considers several different possibilities. This mutation usually takes the form of and altered solution. This cycle continues until such a time that the best possible solution to the problem is derived (Skinner, n.d.).

One of the simplest and most common applications of a genetic algorithm is in the form of the so-called "Travelling Salesman Problem". In this scenario, the algorithm is presented with a series of locations of which each must be visited only once. The system is given the position of each location as well as the distances between them (although this can be calculated if the positions or literal). The algorithm then tests different paths that can be taken in order to identify the shortest possible path required to visit each position. Figure 3 shows a visual representation