**Abstract**

Uses for DSS range from the simple yes/no to building hypothetical worlds where consequences of decisions are played out. Active Structure, with the ability for structure to build structure within the automated decision support process, can be used to simulate the effects of decisions in these hypothetical worlds.

Decision Support Systems (DSS) are used in a wide range of applications and over a wide range of complexity. This paper is aimed at systems supporting decisions involving a high level of complexity and where many low level decisions need to be made along the way to support the high level decision. Decision support becomes a means of finding a path to the decision, as well as support of the decisionmaking.

**Classification**

Different forms of DSS can be classified using three metrics:

Static or dynamic structure Directed or undirected structure Propagation of simple or complex elements

These metrics allow a classification, both of the DSS and the problems it is designed to handle. A problem which is directly computable from a set of inputs requires relatively limited computing activity so efficiency is of little concern. A problem which metaphorically requires a foundation to be laid and scaffolding to be erected before the structure that will support the decision begins to emerge can be extremely wasteful of computing resources unless the structure itself controls those resources.

**Static Directed Structure – Simple Elements**

The structure may be able to handle complex calculations or combine many influences, but its structure is completely known beforehand and may be programmed as an algorithm. The structure is directed, so that all inputs and outputs are known. Only simple elements – numbers or truth values - are propagated. This is the realm of decision trees and neural networks.

**Static Undirected Structure – Simple Elements**

This is the realm of simple Constraint Reasoning. A static structure is created having sets of numbers at the nodes of the structure. These sets are applied to the links in the structure, which represent relations among the nodes. A reduction in any set causes calculation across other links, so the structure is undirected and self-phasing. The structure is completely known beforehand but is not initially directed. The output may be a value at a node, or a pattern of values. The method uses hypothesising, but only on a structure which remains constant throughout. No propagation of values takes place.

**Static Directed Structure – Complex Elements**

The structure may be able to handle complex calculations or combine many influences, but its structure is completely known beforehand. The structure is directed, so that all inputs and outputs are known. Complex elements – probability spectra, objects – can be propagated, as well as Bayesian values for logic and existence. This is a typical algorithmic approach to a complex problem – the complexity of the algorithm may mirror the complexity of the problem when the algorithm is devised, but the problem will change over time, unlike the algorithm.

**Static Undirected Structure – Complex Elements**

Simple Constraint Reasoning can handle this case if the complex objects have stereotypical attributes – investment analysis for example. If the complex objects differ in structure – choosing between building a hotel or a golf course, say, where the common attributes are limited and the consequences of a partial decision are different, simple Constraint Reasoning does not work, as it cannot handle complex objects at the nodes.

**Dynamic Directed Structure**

A structure which changes itself cannot be directed, otherwise it knows a priori how it should modify itself for the particular problem, and is therefore static.

**Dynamic Undirected Structure – Complex Elements**

Most conceptually hard decisions fit under this classification. The general outline of the problem is known, but many of the details only become known as decisions are made within the problem space – the shape of the problem changes as it is worked on, as do what is seen as the inputs and outputs. There is another metric that becomes available when the structure is dynamic and undirected – there can be essentially one structure, which changes itself but remains one structure, or there can be many structures relatively independent of each other and able to operate on or change each other. The many structures approach (two structures are enough to give this flexibility), where one structure moves across another, modifying as it goes, allows a higher level of complexity to be reached. Each modification or addition allows new sites for further modification.

Active Structure meets the criteria for a dynamic undirected structure propagating complex objects. These properties allow it to operate as a dynamic state machine, responding to changes as they occur. It is capable of carrying out simple or complex cognitive tasks, so it can support the most difficult decisions.