I was among those who proposed problem solving methods (PSMs) in the late 1970s and early 1980s as a knowledge-level description of strategies useful in building knowledge-based systems. This paper summarizes the evolution of my ideas in the last two decades. I start with a review of the original ideas. From an artificial intelligence (AI) point of view, it is not PSMs as such, which are essentially high-level design strategies for computation, that are interesting, but PSMs associated with tasks that have a relation to AI and cognition. They are also interesting with respect to cognitive architecture proposals such as Soar and ACT-R: PSMs are observed regularities in the use of knowledge that an exclusive focus on the architecture level might miss, the latter providing no vocabulary to talk about these regularities. PSMs in the original conception are closely connected to a specific view of knowledge: symbolic expressions represented in a repository and retrieved as needed. I join critics of this view, and maintain with them that most often knowledge is not retrieved from a base as much as constructed as needed. This criticism, however, raises the question of what is in memory that is not knowledge as traditionally conceived in AI, but can support the construction of knowledge in predicate–symbolic form. My recent proposal about cognition and multimodality offers a possible answer. In this view, much of memory consists of perceptual and kinesthetic images, which can be recalled during deliberation and from which internal perception can generate linguistic–symbolic knowledge. For example, from a mental image of a configuration of objects, numerous sentences can be constructed describing spatial relations between the objects. My work on diagrammatic reasoning is an implemented example of how this might work. These internal perceptions on imagistic representations are a new kind of PSM.

Problem solving methods (PSMs) are software components that represent and encode reusable algorithms. They can be combined with representations of domain knowledge to produce intelligent application systems. A goal of research on PSMs is to provide principled methods and tools for composing and reusing algorithms in knowledge-based systems. The ultimate objective is to produce libraries of methods that can be easily adapted for use in these systems. Despite the intuitive appeal of PSMs as conceptual building blocks, in practice, these goals are largely unmet. There are no widely available tools for building applications using PSMs and no public libraries of PSMs available for reuse. This paper analyzes some of the reasons for the lack of widespread adoptions of PSM techniques and illustrate our analysis by describing our experiences developing a complex, high-throughput software system based on PSM principles. We conclude that many fundamental principles in PSM research are useful for building knowledge-based systems. In particular, the task–method decomposition process, which provides a means for structuring knowledge-based tasks, is a powerful abstraction for building systems of analytic methods. However, despite the power of PSMs in the conceptual modeling of knowledge-based systems, software engineering challenges have been seriously underestimated. The complexity of integrating control knowledge modeled by developers using PSMs with the domain knowledge that they model using ontologies creates a barrier to widespread use of PSM-based systems. Nevertheless, the surge of recent interest in ontologies has led to the production of comprehensive domain ontologies and of robust ontology-authoring tools. These developments present new opportunities to leverage the PSM approach.

During the 1980s, a community of artificial intelligence researchers became interested in formalizing problem solving methods (PSMs) as part of an effort called “second-generation expert systems.” We provide an example of how we are applying second-generation expert systems concepts in an agent-based system for space flight operations, the orbital communications adapter mirroring system (OCAMS), which was developed in the Brahms multiagent framework. Brahms modeling language provides an ontology for simulating work practices, including groups, agents, activities, communications, movements, and geographic areas. Activities are a behavioral unit of analysis to be contrasted with tasks, a functional unit of analysis. Problem solving occurs in the context of activities in the service of tasks; appropriate PSMs depend on the context: which people/roles are participating, what tools are available, how the results will be evaluated, and so forth. A work practice simulation facilitates designing workflow tools that appropriately interact with the physical and organizational context in which work occurs. OCAMS was developed using a simulation-to-implementation methodology, in which a prototype workflow tool was embedded in a Brahms simulation of how people would use the tool. The reusable components in a workflow system like OCAMS include entire “problem solvers” (e.g., a planning subsystem), interoperability frameworks, and agents that inspect and change the world. Thus, a tool kit for building workflow tools requires more than a library of PSMs, which play a relatively small role in the overall multiagent, systems-integration architecture. Our research concern has shifted to situations that may arise that are outside the OCAMS' capability. In practical decision making, people must reflect on the validity of their models. As programs becoming actors in the workplace, we need to develop systems that help people to understand the limitations of the models that drive the automated operations, which means in part detecting when the formalizations in the system are inadequate.

We believe that the future for problem solving method (PSM) derived work is very promising. In short, PSMs provide a solid foundation for creating a semantic layer supporting planetary-scale networks. Moreover, within a world-scale network where billions services are used and created by billions of parties in ad hoc dynamic fashion we believe that PSM-based mechanisms provide the only viable approach to dealing the sheer scale systematically. Our current experiments in this area are based upon a generic ontology for describing Web services derived from earlier work on PSMs. We outline how platforms based on our ontology can support large-scale networked interactivity in three main areas. Within a large European project we are able to map business level process descriptions to semantic Web service descriptions, to enable business experts to manage and use enterprise processes running in corporate information technology systems. Although highly successful, Web service-based applications predominately run behind corporate firewalls and are far less pervasive on the general Web. Within a second large European project we are extending our semantic service work using the principles underlying the Web and Web 2.0 to transform the Web from a Web of data to one where services are managed and used at large scale. Significant initiatives are now underway in North America, Asia, and Europe to design a new Internet using a “clean-slate” approach to fulfill the demands created by new modes of use and the additional 3 billion users linked to mobile phones. Our investigations within the European-based Future Internet program indicate that a significant opportunity exists for our PSM-derived work to address the key challenges currently identified: scalability, trust, interoperability, pervasive usability, and mobility. We outline one PSM-derived approach as an exemplar.

This paper presents a comparison of two previous and separate efforts to develop an ontology in the engineering design domain, together with an ontology proposal from which ontologies for a specific application may be derived. The research contrasts an empirical, user-centered approach to developing the ontology engineering design integrated taxonomies (EDIT) with a theoretical approach in which concepts and relations are elicited from engineering design theories design ontology (DO). The limitations and advantages of each approach are discussed. The research methodology adopted is to map the ontology through examining each of the concepts and relations contained within each of the ontologies DO and EDIT with respect to the other. The comparison process results in an examination of both ontologies, with a few changes resulting from this. The importance of the two different approaches, one that is theoretically sound and another that is applicable, is recognized and argued. Finally, the merged ontology for engineering design is proposed as a template ontology that can be tailored by researchers and practitioners for a specific context.

When designing a product, the earlier the potential risks can be identified, the more costs can be saved, as it is easier to modify a design in its early stages. Several methods exist to analyze the risk in a system, but all require a mature design. However, by applying the concept of “common interfaces” to a functional model and utilizing a historical knowledge base, it is possible to analyze chains of failures during the conceptual phase of product design. This paper presents a method based on these common interfaces to be used in conjunction with other methods such as risk in early design to allow a more complete risk analysis during the conceptual design phase. Finally, application of this method is demonstrated in a design setting by applying it to a thermal control subsystem.

Most past approaches to data mining have been based on association rules. However, the simple application of association rules usually only changes the user's problem from dealing with millions of data points to dealing with thousands of rules. Although this may somewhat reduce the scale of the problem, it is not a completely satisfactory solution. This paper presents a new data mining technique, called knowledge cohesion (KC), which takes into account a domain ontology and the user's interest in exploring certain data sets to extract knowledge, in the form of semantic nets, from large data sets. The KC method has been successfully applied to mine causal relations from oil platform accident reports. In a comparison with association rule techniques for the same domain, KC has shown a significant improvement in the extraction of relevant knowledge, using processing complexity and knowledge manageability as the evaluation criteria.