Artificial Intelligence and Decision Support Systems

---

An Example of AI incorporated in a DSS

The following was taken from an article by Dolk & Kridel (1991):

---

Many decision support systems are used for forecasting economic variables. For example: the Treasury Short-Term Forecasting Model is used to forecast such variables as the level of unemployment, the growth of national income, the rate of inflation and balance of payments in the UK. Firms may use economic models to forecast future sales of their product or to predict market share one year ahead.

Regression Analysis estimates the relationship between several variables. This relationship may be used to predict the value of one of the variables in the future, given values for the other variables.

Thus, the government may assume a certain public expenditure figure and certain tax revenue figure for the year ahead and use the Treasury model to forecast the effect this will have on the unemployment level. Whether their forecast is accurate will depend partly on the strength of the statistical relationship between the variables. This may be assessed by such measures as the correlation coefficient and the coefficient of determination. Econometrics is the discipline concerned with building economic models based on sound statistical and econometric principles.

Most users of Econometric Models are not experts in the field. If they were to use an econometric model which did not give them satisfactory results it is unlikely that they would know what to do about it. An econometrician however would be able to suggest ways of solving the problem, perhaps by transformation of the variables or by altering the model’s specification.

---

The purpose of the Dolk & Kridel (1991) article is to outline a DSS which uses econometric models but which has built into it an AI component. The purpose of the AI component is to provide an “Expert Econometrician” who can advise the user on what to do next, once the expert has ascertained what the user is trying to achieve. It is rather as if you were using an econometrics package and a lecturer were standing behind you, monitoring your progress and giving advice, once he had established what you were doing. It also outlines a possible programming language for implementing such a system. However, in what follows we are only concerned in giving a bare outline of how such a system may work.

It is called an “Active” DSS, for rather than waiting for the user to invoke the expert part of the system, the expert part automatically intervenes to give the user assistance once it has established the user’s objectives.

Figure 42 gives the Basic Architecture of the System:

There are two process managers in the system:

• User-Directed Process Manager (UDPM)
• Computer-Directed Process Manager (CPDM)

The Basic Architecture of the System:

• The UDPM carries out the instructions inputted by the user. For example: the user may wish a simple regression analysis to be carried out on two sets of data. The output of this data will be the estimated regression equation, a value for the coefficient of deterioration and a test of significance on the regression coefficients.
• The History Recorder will monitor all the user commands and associated output and record them in the History Record.
• The History Inference Processor scans the History Record in order to ascertain the user’s problem-working process. It does this by trying to match the history record against a library of predefined models of problem-solving that reside in the Schema Library. If the inference processor finds such a model then the CPDM takes over and caries out the processes associated with the library model. The computer-directed output from the library model is sent to the user along with the user-directed output. If the History Inference Processor cannot identify an appropriate model in the Schema Library then control is returned to the user along with the user-directed output.
• The most difficult part of designing such a system is constructing the Schema Library. The model templates that reside in the library must accurately reflect real human problem-solving behaviour.

Dolk & Kridel (1991) suggest three approaches:

1. Protocol Analysis - this tries to establish human problem-solving processes empirically
2. Pattern recognition or script recognition, which carries out dynamic analysis of the History Record and perhaps tests a hypothesis of the user behaviour by simulation modelling
3. Predefined schemes based on conventional wisdom or heuristics

Although the application of AI to DSS creates many problems, it is a promising area of research to make DSS’s more user-friendly and therefore attractive to a wider class of users.

---

Reference(s)
	Journal		Dolk, D. R. & Kridel, D. J. (1991) An active modeling system for econometric analysis. Decision Support Systems, Volume: 7, Issue: 4, Page(s): 315-328. [doi: 10.1016/0167-9236(91)90061-F]. [Available on: ScienceDirect: http://www.sciencedirect.com/science/article/pii/016792369190061F].
	Book		Pfaffenberger, B. (2002) Computers in Your Future 2003. 5th Edition. Prentice Hall: United States of America (USA), New Jersey (NJ), Bergen, Upper Saddle River. [ISBN: 9780139227820]. [Available on: Amazon: https://amzn.to/3gv8n7D].

---

Reference (or cite) Article
	Kahlon, R. S. (2013) Artificial Intelligence & Decision Support Systems [Online]. dkode: United Kingdom, England, London. [Published on: 2013-02-09]. [Article ID: RSK666-0000103]. [Available on: dkode | Ravi - https://ravi.dkode.co/2013/02/artificial-intelligence-decision.html].

---