PART I: INVENTORY
2.4 Intelligent planning support
As mentioned earlier, in the 1980s the emphasis in planning shifted from information to knowledge as the required keystone for decision-making. The resultant need for access to sets of analytic tools within user-friendly environments started a line of research that focussed on the development of Decision Support Systems (DSS), and more specifically Spatial DSS (SDSS). From the background of Management Information Systems, this line of research has primarily been drawn on data processing and/or operations research techniques to support users in their decision-making concerning unstructured or semi- structured problems by means of utilizing decision rules and models (‘model base’) coupled with a comprehensive database (e.g., Turban, 1988; Darlington, 2000). The overall objective of DSS is to improve the quality of decision-making by providing users with sets of previously separate tools (data and models) to form a unified whole that is more valuable than the sum of the parts (Malczewski, 1999). Generally, DSS require extensive user involvement in the sense that it is the user who needs to select appropriate models from the model base, indicate the data to use from the database, and interpret the system outputs as a basis for making a decision. Hence, the system is essentially a repository for models and data (‘toolbox’) and an environment to run models. The system outputs – e.g., optimal values or locations – will not provide users with solutions but quantitatively assist them in the search for solutions.
Although this can already be of great value to planners, it should be considered the main drawback of (S)DSS that intelligence is lacking in terms of reasoning capabilities that would enable solving problems on behalf of the user. The significance of this drawback becomes evident when realizing that decision-making in planning requires
capabilities to process both formal (e.g., legal requirements, planning procedures, and problem situations) and informal information (e.g., obtained through personal judgements, hunches, intuition, hearsay, and personal experiences), demanding support in both quantitative and qualitative terms (Han and Kim, 1990). Furthermore, one of the many complexities that planners experience is the fact that the information to be used is often incomplete, imprecise and dynamic, thus requiring human knowledge and expertise to manipulate and digest the information effectively and efficiently (Leung, 1997). According to Klosterman (2001), intelligent support implies computer systems that are able to deal with novel situations and new problems, to apply knowledge acquired from experience, and to use the power of reasoning effectively as a guide to behaviour. Because such functionality is not included in the concept of (S)DSS, the typical complexities of planning remain poorly supported.
The issue of providing users with intelligent computer systems has been the topic of a parallel line of research that set off decades ago from the background of Artificial Intelligence and materialized in the form of Expert Systems (ES), and more specifically Spatial ES (SES). The basic intention of ES is to make (tentative) decisions by means of a reasoning mechanism for propagating inferences over a knowledge base that contains an expert’s knowledge for a particular problem domain (Turban, 1988). Human expertise is mimicked by capturing knowledge such as human experience, valuation, intuition and judgement (e.g., Leung, 1997; Darlington, 2000), which brings about the capability to solve (spatial) problems as well as or better than human experts, use expert knowledge in the form of rules or frames, and interact with users by transferring expertise and rendering advice or recommendations (Malczewski, 1999). However, whereas the problem area of DSS is usually broad and complex, ES are typically confined to narrow and well-defined problem domains and operate in highly isolated modes. The consequence is that they are most suitable for providing advice on repetitive problem areas such as diagnosis and malfunctions (e.g., Turban, 1988). Practically, the introversion of ES has been the main hindering factor for finding application in managerial decision support that requires much more negotiation and interaction with exogenous bodies of knowledge (Edwards, 1992). Theoretically, (S)ES have clear potentials to help out planners with various problems (Batty and Yeh, 1991). To date, however, only applications exist that address subfields – like problem formulation (George, 1995) – instead of providing solutions for the entire problem domain. Among a variety of plausible reasons (e.g., Han and Kim, 1990), this may be largely due to the fact that the domain of problems in planning is relatively large
and difficult to confine to boundaries, implying encoding more knowledge – perhaps even more than can be encoded – along various dimensions and disciplines, in order to enable decision-making (Arentze, 1999).
The anticipated contributions of DSS and ES to the field of planning are clearly complementary (Table 2-1). In order to create more powerful and useful computer systems, the idea of integrating these technologies into systems that could yield synergy has developed into a new line of research since the late 1980s. Commonly, this integration is considered to be a matter of using ES techniques within a DSS framework for enhancing the modelling capabilities of the system or improving the intelligence of the system in various components including data management, model management and user interface (Arentze, 1999). Resulting applications have been given various names, such as intelligent DSS (IDSS), intelligent support systems, expert DSS (EDSS), expert support system (ESS), and knowledge-based DSS (KBDSS) (e.g., El-Najdawi and Stylianou 1993). Albeit less common due to the wide scope required for planning support, it is also possible to integrate DSS into conventional ES, which would mean recognizing the fact that experts often use quantitative models to support their experience, intuition, or rules-of-thumb (Han and Kim, 1990).
From the viewpoint of spatial decision-making, Malczewski (1999) described a generic framework for a Spatial Expert Support System (SESS) in which an ES component is linked (or integrated) to each of the three basic components of an SDSS, i.e., a geographical database management system (GDBMS), a model-based management system (MBMS), and a dialogue generation and management system (DGMS). As opposed to this system architecture that advocates enhancing each of the SDSS components with dedicated ES techniques, Leung (1997) suggested one central ES shell to operate as the core or brain of SDSS, providing linkages – interfaces – with the various SDSS components. In addition, the necessity of knowledge acquisition modules is
Table 2-1. Synergetic contributions of DSS and ES to integrated systems (source: Turban, 1988)
DSS contribution ES contribution
• Experience in data collection • Experience in implementation
• Personalized advice to users to match their decision styles
• Quantitative, mathematical, and computational reasoning
• Intelligent advice (faster and cheaper than human) to DSS or its user
• Explanatory capabilities
• Computerization of decision-making process • Qualitative analysis (e.g., analogical reasoning,
explicitly stressed within this system architecture. As such, the ES shell directs control and information flows, while providing facilities to represent and store domain specific knowledge acquired from human experts and learning examples, and to contain metaknowledge for inference control, systems and user interface, and external communication. Albeit in conceptual terms, El-Najdawi and Stylianou (1993) proposed a model for DSS-ES integration that basically stands midway between these two architectures, as the enhanced DSS components are linked through a blackboard, implying that communications between the components is arranged through a shared database. Several other expert components are suggested as part of this system architecture, among which a knowledge cache that, by means of a collection of intelligent agents, represents the knowledge of specialists in the organization. These agents are intended to be available to the various system components through the blackboard, and to help the user come to a better understanding of the structure and formulation of a problem at hand if necessary. Additionally, the agents are assumed to act as expert critics by analyzing the user’s decision-making process – including the choice of models and databases, interpretation of results and conclusions – and providing suggestions.
In general, consensus appears to exist over the fact that DSS-ES integrations require multiple experts – either in terms of components or interfaces – in order to have the resulting system operate intelligently. But there are more possible ways to adopt the notion of multi-experts than only for interconnecting DSS components. In this respect, the collection of intelligent agents as suggested by El-Najdawi and Stylianou (1993) is noteworthy, as it advocates the use of a multi-expert concept for the inclusion of a repository of multidisciplinary knowledge that can support both system components and users. Some prototype systems exist that embody this type of concept, like the Intelligent CAD System (ICADS) developed in the area of building design and physics (Pohl et al., 2000) that incorporates multiple experts in domains such as natural and artificial lightning, noise control, structural system selection, climatic determinants, and energy conservation. The system assists architects in the development, analysis, and evaluation of solutions during the early design process.
The repository concept could be expanded with an extra dimension, when adopting the way in which the notion of multi-experts is commonly used in the field of pattern recognition (e.g., Cordella et al., 1999): in order to improve the reliability of system output (decision), several compensatory experts are assigned to solve the same problem under the assumption that, by suitably combining the results of these experts according to
a certain rule (combining rule), the performance obtained can be better than that of a single expert. In general terms, this approach is resembled by some recent applications in the field of urban planning. In Ligtenberg et al, (2001) and Ligtenberg et al. (2004) simulation studies were described in which different stakeholders with various objectives are represented to perform the same task (e.g., determining the preferred location for new urbanization), after which the varying outcomes are subjected to a voting procedure in order to select the outcome for implementation.